모종의 음모/ai 프로그램

LLM / 1080 ti 모드 별 벤치마크

구차니 2026. 9. 17. 10:36

궁금한 것 을 해소하기 위해 대충 5시간에 걸쳐 GPU 갈굼

  • MTP 가 정말 속도에 영향을 주냐
  • GPU 2개인데 pcie3.0 x8과 x4에 영향을 받는가
  • 컨텍스트 최대 길이가 성능에 영향을 미치는가
  • 입력 토큰길이에 따라 성능 저하는 어느정도 발생하는가
  • 단일과 -sm layer로 나누었을때 성능은 어떠한가

 

결과만 요약하면

  • GPU 0 / 1이 우선권 문제인지 lane 폭 문제인진 모르겠으나 GPU0이 3~10% 정도 이상 성능이 좋게 나옴
  • MTP는 켤수 있으면 켜라. 조건에 따라 다르지만 MTP 1이 1.3~1.5배 향상됨
  • 컨텍스트 최대 길이는 의외로 생성 성능에 영향을 안준다.
  • 한쪽에 가능하지만, 억지로 레이어 쪼개서 양쪽으로 올리면 1.1~1.4배 성능 차이가 난다. 되도록이면 1대로 돌리자

 

아래는 테스트용 스크립트들. 추후 쓰려면 바이너리 경로등을 손봐야 함

bench.zip
0.01MB

 

 

 

아래는 원본 내용

report.md
0.08MB

 

 

llama.cpp Benchmark Report

Short conversation (3-turn, context accumulates)

Questions: see questions.txt. Columns show prompt_t/s / generation_t/s per turn.

gemma4-12b

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

gemma-4-12b-it-Q4_0 none 0 131072 off 139.2/34.8 143.4/34.6 270.3/34.4 38.4 OK
gemma-4-12b-it-Q4_0 none 0 131072 on(n=1) 123.9/57.5 141.8/56.1 246.3/51.2 21.5 OK
gemma-4-12b-it-Q4_0 none 0 131072 on(n=2) 123.3/61.5 187.3/61.7 238.3/59.1 18.6 OK
gemma-4-12b-it-Q4_0 none 0 131072 on(n=3) 118.4/65.5 227.8/59.7 231.2/51.6 18.9 OK
gemma-4-12b-it-Q4_0 none 0 131072 on(n=4) 123.2/65.5 181.6/57.1 240.6/45.5 20.0 OK
gemma-4-12b-it-Q4_0 none 0 65536 off 140.0/33.5 146.5/33.2 266.5/32.8 27.0 OK
gemma-4-12b-it-Q4_0 none 0 65536 on(n=1) 118.5/54.5 138.2/55.1 245.5/51.2 20.0 OK
gemma-4-12b-it-Q4_0 none 0 65536 on(n=2) 121.3/59.5 226.9/62.4 229.2/51.9 19.0 OK
gemma-4-12b-it-Q4_0 none 0 65536 on(n=3) 122.9/66.3 243.9/52.7 222.2/52.1 19.0 OK
gemma-4-12b-it-Q4_0 none 0 65536 on(n=4) 118.3/65.2 139.6/60.8 245.8/48.4 19.1 OK
gemma-4-12b-it-Q4_0 none 0 32768 off 136.6/33.6 146.1/33.2 263.9/32.7 27.0 OK
gemma-4-12b-it-Q4_0 none 0 32768 on(n=1) 125.1/57.3 140.9/55.5 242.6/51.8 19.7 OK
gemma-4-12b-it-Q4_0 none 0 32768 on(n=2) 120.7/65.7 139.8/58.8 243.7/57.5 18.3 OK
gemma-4-12b-it-Q4_0 none 0 32768 on(n=3) 125.6/63.6 139.2/60.2 244.5/53.6 18.7 OK
gemma-4-12b-it-Q4_0 none 0 32768 on(n=4) 113.1/60.4 182.1/63.9 242.7/43.8 19.8 OK
gemma-4-12b-it-Q4_0 none 0 16384 off 139.6/33.5 146.0/32.9 267.0/32.7 27.1 OK
gemma-4-12b-it-Q4_0 none 0 16384 on(n=1) 122.8/55.8 213.0/56.2 230.9/53.9 19.6 OK
gemma-4-12b-it-Q4_0 none 0 16384 on(n=2) 124.2/62.4 139.8/60.7 243.0/54.1 18.7 OK
gemma-4-12b-it-Q4_0 none 0 16384 on(n=3) 122.5/60.7 140.1/63.2 244.8/53.5 18.7 OK
gemma-4-12b-it-Q4_0 none 0 16384 on(n=4) 120.8/60.6 141.5/58.4 245.8/50.0 19.4 OK
gemma-4-12b-it-Q4_0 none 1 131072 off 135.6/32.8 249.3/32.7 225.6/32.5 27.0 OK
gemma-4-12b-it-Q4_0 none 1 131072 on(n=1) 106.7/52.1 130.0/51.5 223.8/49.9 22.8 OK
gemma-4-12b-it-Q4_0 none 1 131072 on(n=2) 117.4/59.5 129.0/57.8 224.2/53.2 21.3 OK
gemma-4-12b-it-Q4_0 none 1 131072 on(n=3) 117.7/61.5 155.4/54.7 218.6/52.5 21.5 OK
gemma-4-12b-it-Q4_0 none 1 131072 on(n=4) 112.3/62.0 128.8/56.6 223.6/45.7 22.1 OK
gemma-4-12b-it-Q4_0 none 1 65536 off 130.6/31.5 243.5/30.9 223.0/30.6 29.4 OK
gemma-4-12b-it-Q4_0 none 1 65536 on(n=1) 117.7/50.5 130.1/50.2 225.2/48.2 23.2 OK
gemma-4-12b-it-Q4_0 none 1 65536 on(n=2) 117.1/57.9 128.0/56.5 222.8/49.6 21.9 OK
gemma-4-12b-it-Q4_0 none 1 65536 on(n=3) 117.0/51.0 129.5/54.2 222.5/51.0 22.5 OK
gemma-4-12b-it-Q4_0 none 1 65536 on(n=4) 117.0/61.7 225.9/59.6 200.0/45.0 21.5 OK
gemma-4-12b-it-Q4_0 none 1 32768 off 133.9/31.7 212.7/31.0 227.7/30.7 30.3 OK
gemma-4-12b-it-Q4_0 none 1 32768 on(n=1) 119.1/49.3 130.3/51.8 222.2/48.9 22.9 OK
gemma-4-12b-it-Q4_0 none 1 32768 on(n=2) 119.2/61.1 130.8/57.4 224.5/48.8 21.4 OK
gemma-4-12b-it-Q4_0 none 1 32768 on(n=3) 119.4/58.1 127.6/53.1 216.0/45.0 22.4 OK
gemma-4-12b-it-Q4_0 none 1 32768 on(n=4) 118.7/60.7 130.8/53.0 225.1/45.8 22.2 OK
gemma-4-12b-it-Q4_0 none 1 16384 off 132.8/31.8 137.5/31.0 244.6/30.7 30.1 OK
gemma-4-12b-it-Q4_0 none 1 16384 on(n=1) 107.6/52.2 227.2/51.0 200.4/48.9 21.2 OK
gemma-4-12b-it-Q4_0 none 1 16384 on(n=2) 110.5/57.1 209.0/56.6 204.2/51.1 21.1 OK
gemma-4-12b-it-Q4_0 none 1 16384 on(n=3) 117.0/59.2 129.6/57.1 208.3/46.9 21.8 OK
gemma-4-12b-it-Q4_0 none 1 16384 on(n=4) 117.1/63.0 193.5/54.3 208.2/43.3 22.3 OK
gemma-4-12b-it-Q4_0 layer both 131072 off 138.0/33.4 145.7/33.3 260.4/33.2 28.0 OK
gemma-4-12b-it-Q4_0 layer both 131072 on(n=1) 123.1/49.2 136.9/48.5 238.2/47.3 23.3 OK
gemma-4-12b-it-Q4_0 layer both 131072 on(n=2) 122.3/55.9 239.8/50.8 213.4/45.0 22.8 OK
gemma-4-12b-it-Q4_0 layer both 131072 on(n=3) 122.4/49.1 135.9/48.8 236.3/43.6 23.7 OK
gemma-4-12b-it-Q4_0 layer both 131072 on(n=4) 118.0/53.8 136.3/47.4 237.4/38.9 24.4 OK
gemma-4-12b-it-Q4_0 layer both 65536 off 136.4/33.1 143.0/33.0 255.2/33.0 28.1 OK
gemma-4-12b-it-Q4_0 layer both 65536 on(n=1) 122.1/48.4 225.8/49.4 218.4/47.4 22.5 OK
gemma-4-12b-it-Q4_0 layer both 65536 on(n=2) 122.3/54.4 136.8/51.9 237.7/44.2 22.5 OK
gemma-4-12b-it-Q4_0 layer both 65536 on(n=3) 122.4/45.0 136.5/47.1 237.8/44.8 24.0 OK
gemma-4-12b-it-Q4_0 layer both 65536 on(n=4) 122.4/53.7 136.4/47.1 238.4/42.6 23.3 OK
gemma-4-12b-it-Q4_0 layer both 32768 off 140.6/33.1 145.9/33.1 260.2/33.0 27.9 OK
gemma-4-12b-it-Q4_0 layer both 32768 on(n=1) 122.9/48.3 137.3/49.0 239.4/47.7 22.8 OK
gemma-4-12b-it-Q4_0 layer both 32768 on(n=2) 109.1/52.3 137.7/52.0 239.2/46.8 22.2 OK
gemma-4-12b-it-Q4_0 layer both 32768 on(n=3) 113.3/53.9 138.1/47.0 239.0/45.8 22.7 OK
gemma-4-12b-it-Q4_0 layer both 32768 on(n=4) 124.3/52.7 237.4/48.5 217.2/38.7 23.7 OK
gemma-4-12b-it-Q4_0 layer both 16384 off 135.4/32.8 143.2/32.8 257.2/32.8 28.1 OK
gemma-4-12b-it-Q4_0 layer both 16384 on(n=1) 118.5/50.7 135.7/49.0 237.0/47.7 22.5 OK
gemma-4-12b-it-Q4_0 layer both 16384 on(n=2) 120.0/53.6 216.3/52.3 219.0/46.5 21.9 OK
gemma-4-12b-it-Q4_0 layer both 16384 on(n=3) 119.0/53.8 235.9/49.9 214.2/46.3 21.2 OK
gemma-4-12b-it-Q4_0 layer both 16384 on(n=4) 98.1/51.2 236.1/46.7 214.0/40.8 23.8 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 131072 off 85.4/31.3 93.9/31.1 203.3/30.5 36.9 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=1) 78.3/42.7 166.5/42.6 181.6/40.2 24.0 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=2) 78.5/45.1 92.4/43.7 184.5/34.0 24.7 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=3) 78.6/40.4 91.7/41.9 189.3/31.6 26.2 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=4) 78.9/42.5 153.1/39.7 183.6/26.9 27.7 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 65536 off 85.2/30.8 92.6/29.7 203.0/29.5 29.2 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=1) 78.9/45.4 163.3/42.7 181.7/40.6 22.6 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=2) 78.0/43.3 91.8/42.6 190.4/35.2 24.6 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=3) 79.2/41.9 131.7/42.2 185.5/32.4 25.5 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=4) 78.1/42.2 92.5/41.4 191.5/35.2 25.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 32768 off 84.8/30.6 93.7/29.6 203.1/29.4 29.4 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=1) 73.8/44.8 157.0/42.1 182.4/39.7 23.7 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=2) 70.6/43.6 155.4/46.0 181.4/36.8 23.9 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=3) 77.4/40.6 93.5/43.4 189.3/29.6 26.3 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=4) 78.5/43.6 92.2/36.1 188.0/30.9 26.7 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 16384 off 83.1/30.8 96.2/29.6 198.6/29.4 29.3 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=1) 78.3/44.4 93.2/43.3 191.1/40.6 23.4 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=2) 76.5/45.0 92.9/44.7 187.7/36.5 23.9 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=3) 78.8/43.9 92.3/44.5 188.9/33.0 24.8 OK
gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=4) 68.9/45.3 173.1/37.3 181.7/31.7 25.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 131072 off 84.3/30.6 93.4/30.5 192.8/30.2 30.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=1) 76.0/43.4 178.7/40.8 164.0/37.0 26.3 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=2) 76.1/44.7 90.1/42.6 174.1/36.0 26.0 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=3) 71.8/38.8 177.5/38.2 163.8/37.9 27.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=4) 75.7/38.1 88.6/40.9 178.3/28.7 29.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 65536 off 81.6/28.6 87.9/28.2 188.6/28.0 32.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=1) 70.9/43.4 81.7/39.7 176.3/37.0 26.2 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=2) 75.9/42.5 178.2/38.9 164.5/37.8 25.7 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=3) 76.2/40.1 123.6/36.9 173.4/29.7 28.8 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=4) 73.4/43.0 87.9/36.1 170.6/27.2 29.4 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 32768 off 82.1/28.5 185.9/28.1 176.6/28.1 32.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=1) 75.6/42.5 87.4/42.0 175.6/36.3 26.1 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=2) 76.1/42.0 160.3/44.1 166.9/35.7 24.6 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=3) 66.4/36.6 88.3/35.5 171.7/30.3 29.6 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=4) 76.0/46.1 148.8/46.1 162.2/26.7 27.3 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 16384 off 81.2/28.1 91.4/28.0 187.5/28.0 32.3 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=1) 76.3/42.1 88.9/42.3 174.9/35.7 26.2 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=2) 76.3/39.3 174.5/41.0 164.8/35.1 26.9 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=3) 74.7/40.4 123.5/38.3 170.6/31.1 28.2 OK
gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=4) 76.3/40.4 89.3/40.0 178.1/30.9 27.9 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 131072 off 81.7/31.8 92.9/31.7 196.6/31.6 28.8 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=1) 75.9/42.6 159.0/41.0 174.5/40.2 25.7 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=2) 73.9/44.2 90.3/38.6 184.9/35.4 26.7 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=3) 73.9/39.7 90.3/39.6 185.0/29.4 28.6 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=4) 71.8/35.7 87.9/35.7 182.8/26.0 31.2 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 65536 off 81.6/31.5 90.1/31.2 196.0/31.3 28.8 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=1) 72.9/43.2 90.6/41.7 184.6/37.1 25.7 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=2) 75.7/41.7 90.3/41.2 184.0/36.1 26.2 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=3) 75.7/38.8 90.0/37.7 184.7/30.1 28.6 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=4) 64.3/37.0 88.7/40.3 184.7/28.0 29.2 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 32768 off 81.7/31.5 129.4/31.3 186.1/30.8 28.9 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=1) 73.2/42.8 138.4/41.9 175.3/39.9 25.1 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=2) 75.6/42.0 90.6/38.4 185.3/37.5 26.1 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=3) 71.1/37.6 167.8/38.1 173.2/27.0 29.6 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=4) 76.0/34.6 178.5/30.4 174.5/27.6 31.6 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 16384 off 80.9/31.6 91.1/31.3 190.5/31.0 28.8 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=1) 75.9/41.5 90.1/43.4 184.0/37.2 25.2 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=2) 73.9/41.2 102.5/40.8 183.7/36.6 25.8 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=3) 73.8/38.1 90.5/40.6 185.0/31.5 27.4 OK
gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=4) 72.1/34.2 90.2/36.8 184.3/30.5 29.3 OK

gemma4-26b

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

gemma-4-26B-A4B-it-UD-IQ2_M none 0 131072 off - - - 21.8 OOM
gemma-4-26B-A4B-it-UD-IQ2_M none 0 65536 off - - - 3.6 OOM
gemma-4-26B-A4B-it-UD-IQ2_M none 0 32768 off 92.5/65.4 175.2/65.2 341.6/64.8 16.1 OK
gemma-4-26B-A4B-it-UD-IQ2_M none 0 16384 off 135.9/64.9 196.1/64.4 338.4/63.2 16.2 OK
gemma-4-26B-A4B-it-UD-IQ2_M none 1 131072 off - - - 6.1 OOM
gemma-4-26B-A4B-it-UD-IQ2_M none 1 65536 off - - - 6.1 OOM
gemma-4-26B-A4B-it-UD-IQ2_M none 1 32768 off 147.6/60.0 177.3/59.9 317.9/59.4 19.5 OK
gemma-4-26B-A4B-it-UD-IQ2_M none 1 16384 off 149.8/59.8 176.1/59.6 314.3/58.2 19.7 OK
gemma-4-26B-A4B-it-UD-IQ2_M layer both 131072 off 148.0/61.8 196.3/61.8 333.4/61.7 18.1 OK
gemma-4-26B-A4B-it-UD-IQ2_M layer both 65536 off 147.7/61.7 196.1/61.6 332.7/61.5 18.1 OK
gemma-4-26B-A4B-it-UD-IQ2_M layer both 32768 off 149.5/61.5 194.9/61.5 330.8/61.4 17.9 OK
gemma-4-26B-A4B-it-UD-IQ2_M layer both 16384 off 149.1/61.5 195.4/61.5 331.4/61.3 17.8 OK

gemma4-31b

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

gemma-4-31B-it-UD-IQ2_XXS none 0 131072 off - - - 19.0 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=1) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=2) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=3) - - - 4.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=4) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 65536 off - - - 3.4 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=1) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=2) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=3) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=4) - - - 4.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 32768 off - - - 3.4 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=1) - - - 4.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=2) - - - 4.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=3) - - - 4.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=4) - - - 4.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 16384 off 43.5/15.0 48.2/14.9 96.1/14.2 57.6 OK
gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=1) - - - 6.0 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=2) - - - 5.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=3) - - - 5.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=4) - - - 5.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 131072 off - - - 5.5 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=1) - - - 6.3 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=2) - - - 6.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=3) - - - 6.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=4) - - - 6.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 65536 off - - - 5.5 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=1) - - - 6.3 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=2) - - - 6.4 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=3) - - - 6.3 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=4) - - - 6.3 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 32768 off - - - 5.5 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=1) - - - 6.3 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=2) - - - 6.4 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=3) - - - 6.3 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=4) - - - 6.4 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 16384 off 43.2/14.8 46.4/14.0 89.6/13.3 62.3 OK
gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=1) - - - 7.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=2) - - - 7.2 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=3) - - - 7.1 OOM
gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=4) - - - 7.1 OOM
gemma-4-31B-it-UD-IQ2_XXS layer both 131072 off 41.9/15.3 60.1/15.2 94.0/15.1 57.2 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=1) - - - 6.6 OOM
gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=2) - - - 6.7 OOM
gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=3) - - - 6.6 OOM
gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=4) - - - 6.6 OOM
gemma-4-31B-it-UD-IQ2_XXS layer both 65536 off 42.1/15.2 79.4/15.1 92.2/15.0 57.5 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=1) 39.3/22.3 67.4/21.9 85.3/21.0 44.1 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=2) 39.2/23.4 91.9/20.7 81.8/21.2 44.0 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=3) 37.0/23.9 138.9/16.4 49.6/15.9 52.9 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=4) 39.2/20.4 45.8/19.0 88.5/18.2 48.7 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 32768 off 41.3/15.1 85.4/15.0 133.1/14.8 57.8 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=1) 38.7/22.9 92.4/20.9 80.9/20.6 42.9 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=2) 37.8/22.7 67.6/21.7 85.1/19.9 42.0 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=3) 39.2/20.7 45.9/19.4 88.9/17.5 48.6 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=4) 39.2/20.8 45.7/20.3 87.5/17.2 48.2 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 16384 off 41.6/15.1 47.9/14.8 95.1/14.6 58.1 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=1) 39.3/22.6 66.8/21.9 86.5/20.7 43.8 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=2) 39.3/24.3 90.5/20.5 81.3/17.9 46.0 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=3) 39.4/20.3 52.6/19.9 88.4/17.8 48.3 OK
gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=4) 39.3/19.7 145.8/17.8 53.0/15.5 54.2 OK
gemma-4-31B-it-UD-Q2_K_XL none 0 131072 off - - - 24.0 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 65536 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=2) - - - 3.7 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 32768 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 16384 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 131072 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 65536 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 32768 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 16384 off - - - 2.9 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=1) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=2) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=3) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=4) - - - 3.5 OOM
gemma-4-31B-it-UD-Q2_K_XL layer both 131072 off - - - 5.5 OOM
gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=1) - - - 6.2 OOM
gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=2) - - - 6.3 OOM
gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=3) - - - 6.3 OOM
gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=4) - - - 6.2 OOM
gemma-4-31B-it-UD-Q2_K_XL layer both 65536 off 35.3/13.8 40.7/13.7 86.1/13.6 63.6 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=1) 32.7/20.5 40.3/20.7 79.4/20.2 47.5 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=2) 32.7/22.4 64.2/21.0 76.9/18.7 47.3 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=3) 32.9/21.9 64.0/20.4 76.9/18.2 46.8 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=4) 31.0/21.0 39.6/20.1 80.4/17.7 49.4 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 32768 off 34.1/13.6 68.3/13.4 82.1/13.3 60.9 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=1) 30.9/20.1 48.1/20.7 78.7/19.4 48.0 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=2) 30.9/20.9 40.1/21.4 79.6/18.7 47.6 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=3) 31.2/20.9 40.0/20.6 80.0/17.5 49.0 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=4) 32.6/20.2 64.3/22.5 75.7/18.2 46.0 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 16384 off 34.3/13.6 67.6/13.4 81.0/13.4 64.6 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=1) 30.8/20.3 63.9/20.1 76.7/19.3 48.4 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=2) 32.6/20.8 40.0/21.9 79.9/18.8 47.3 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=3) 32.8/19.9 39.9/17.8 79.8/17.7 51.3 OK
gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=4) 32.8/22.5 40.2/21.1 79.5/16.9 48.3 OK

gemma4-e2b

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

gemma-4-E2B-it-Q4_K_M none 0 131072 off 251.7/92.4 268.8/92.4 827.5/92.3 16.6 OK
gemma-4-E2B-it-Q4_K_M none 0 65536 off 177.7/92.2 380.4/92.2 847.8/92.1 11.0 OK
gemma-4-E2B-it-Q4_K_M none 0 32768 off 178.4/91.9 892.3/92.1 837.1/92.0 8.7 OK
gemma-4-E2B-it-Q4_K_M none 0 16384 off 262.1/88.7 430.0/92.2 846.8/92.1 8.3 OK
gemma-4-E2B-it-Q4_K_M none 1 131072 off 243.8/86.7 358.6/86.7 817.0/86.6 12.0 OK
gemma-4-E2B-it-Q4_K_M none 1 65536 off 258.9/85.4 352.0/85.3 803.6/85.4 12.1 OK
gemma-4-E2B-it-Q4_K_M none 1 32768 off 262.6/81.3 408.6/86.0 804.0/85.8 8.9 OK
gemma-4-E2B-it-Q4_K_M none 1 16384 off 263.2/82.2 406.5/85.9 799.4/85.0 8.9 OK
gemma-4-E2B-it-Q4_K_M layer both 131072 off 205.5/65.2 290.7/64.6 660.5/62.9 15.2 OK
gemma-4-E2B-it-Q4_K_M layer both 65536 off 215.5/63.5 340.7/65.2 672.0/63.5 11.0 OK
gemma-4-E2B-it-Q4_K_M layer both 32768 off 163.0/66.0 355.8/67.8 707.0/65.9 10.7 OK
gemma-4-E2B-it-Q4_K_M layer both 16384 off 167.1/62.6 337.5/64.7 664.8/63.0 11.0 OK

gemma4-e4b

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

gemma-4-E4B-it-Q4_K_M none 0 131072 off 137.3/61.6 439.9/61.5 510.4/61.2 24.2 OK
gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=1) 162.3/81.3 557.5/87.7 486.8/88.6 11.5 OK
gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=2) 154.5/91.4 550.3/92.6 488.2/88.9 10.9 OK
gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=3) 163.7/89.2 338.7/95.8 496.6/91.6 10.5 OK
gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=4) 167.1/69.7 559.5/89.2 475.0/84.6 11.3 OK
gemma-4-E4B-it-Q4_K_M none 0 65536 off 180.4/60.7 517.8/60.9 497.2/60.0 12.0 OK
gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=1) 163.0/88.7 232.9/92.9 490.9/85.8 13.4 OK
gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=2) 163.9/87.5 548.8/91.6 490.6/87.5 10.9 OK
gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=3) 123.7/66.3 544.5/101.7 484.8/76.2 11.3 OK
gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=4) 161.4/63.4 547.0/86.8 486.2/76.4 11.7 OK
gemma-4-E4B-it-Q4_K_M none 0 32768 off 137.9/60.7 527.3/60.9 503.6/59.6 12.1 OK
gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=1) 163.6/78.8 600.7/93.1 475.7/87.3 11.1 OK
gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=2) 165.2/74.1 399.6/99.9 495.0/90.8 10.4 OK
gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=3) 164.3/74.0 589.4/93.0 469.5/88.0 11.1 OK
gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=4) 125.3/59.3 477.0/85.9 489.8/72.6 11.8 OK
gemma-4-E4B-it-Q4_K_M none 0 16384 off 158.4/60.9 576.5/60.8 495.3/59.6 12.3 OK
gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=1) 155.8/79.5 510.5/86.7 484.5/87.8 11.0 OK
gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=2) 160.8/73.3 565.4/97.5 484.7/86.0 10.9 OK
gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=3) 165.7/86.4 232.9/105.1 492.4/94.8 12.8 OK
gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=4) 166.7/65.1 567.5/91.5 483.8/83.0 11.4 OK
gemma-4-E4B-it-Q4_K_M none 1 131072 off 163.1/57.4 225.8/57.4 492.2/57.1 17.2 OK
gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=1) 155.3/73.1 368.1/87.5 467.6/80.4 11.9 OK
gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=2) 136.0/75.6 476.1/91.2 457.9/83.2 11.9 OK
gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=3) 139.9/76.5 570.5/90.8 441.5/83.2 12.2 OK
gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=4) 143.7/67.1 533.1/78.7 447.0/77.4 12.8 OK
gemma-4-E4B-it-Q4_K_M none 1 65536 off 134.1/56.9 581.6/55.2 444.0/54.5 14.1 OK
gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=1) 133.0/83.1 575.9/81.1 438.6/83.2 12.6 OK
gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=2) 120.5/80.7 530.4/88.3 442.6/81.0 12.2 OK
gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=3) 159.8/63.8 429.5/92.1 445.3/87.8 11.7 OK
gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=4) 157.2/76.2 504.1/77.3 448.0/67.6 13.2 OK
gemma-4-E4B-it-Q4_K_M none 1 32768 off 177.3/57.0 630.6/55.3 454.5/54.9 14.1 OK
gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=1) 151.4/84.0 517.6/82.6 437.5/79.5 12.4 OK
gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=2) 162.7/66.7 562.5/86.3 432.2/75.0 12.5 OK
gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=3) 153.3/66.4 533.8/92.9 442.7/77.1 12.1 OK
gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=4) 158.0/58.9 512.7/93.8 437.6/68.2 12.8 OK
gemma-4-E4B-it-Q4_K_M none 1 16384 off 169.7/55.6 316.9/54.5 443.6/54.5 13.4 OK
gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=1) 160.8/77.1 323.0/83.8 457.0/81.1 11.7 OK
gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=2) 159.0/75.8 554.9/87.0 438.5/80.5 12.3 OK
gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=3) 121.6/87.3 224.7/88.1 475.7/76.6 14.8 OK
gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=4) 159.6/88.0 225.4/91.6 476.4/76.9 14.6 OK
gemma-4-E4B-it-Q4_K_M layer both 131072 off 171.6/57.4 559.6/57.7 494.4/57.6 13.3 OK
gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=1) 150.0/76.6 455.2/80.8 478.5/76.7 12.9 OK
gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=2) 156.5/64.9 533.9/82.7 472.6/81.3 12.9 OK
gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=3) 157.8/78.4 559.7/92.1 473.4/77.8 12.6 OK
gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=4) 152.5/62.7 628.6/87.3 463.4/61.2 14.0 OK
gemma-4-E4B-it-Q4_K_M layer both 65536 off 174.8/57.5 548.8/57.6 492.1/57.5 13.3 OK
gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=1) 154.3/76.4 227.1/83.8 487.5/79.0 15.3 OK
gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=2) 158.9/65.6 572.5/85.0 465.8/80.8 12.6 OK
gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=3) 152.1/61.2 540.7/86.7 474.2/78.9 12.5 OK
gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=4) 162.1/66.4 470.2/75.9 478.2/68.5 13.0 OK
gemma-4-E4B-it-Q4_K_M layer both 32768 off 160.8/57.3 439.7/57.5 494.9/57.4 12.8 OK
gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=1) 154.6/77.8 627.5/80.9 464.0/80.2 12.5 OK
gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=2) 159.5/74.9 567.8/84.8 468.9/77.7 12.6 OK
gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=3) 153.4/84.3 432.2/79.2 477.6/73.6 12.4 OK
gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=4) 159.2/54.2 489.7/70.3 474.7/69.8 13.4 OK
gemma-4-E4B-it-Q4_K_M layer both 16384 off 172.1/57.1 476.7/57.5 495.0/57.4 12.9 OK
gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=1) 161.2/82.0 226.4/85.5 489.0/79.1 14.5 OK
gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=2) 159.3/82.0 226.6/89.0 486.9/80.5 14.3 OK
gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=3) 128.7/73.9 531.0/83.8 474.7/70.0 12.5 OK
gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=4) 127.7/69.2 597.7/89.2 464.9/67.4 12.7 OK

gemma4-e4b-qat

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 131072 off 225.8/43.5 338.4/43.3 547.1/43.0 26.3 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 65536 off 254.4/42.6 205.3/43.4 536.7/42.7 20.6 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 32768 off 234.9/41.0 290.8/43.4 532.0/43.2 14.8 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 16384 off 226.2/43.0 361.1/42.8 537.0/42.9 18.1 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 131072 off 243.3/42.9 244.4/41.9 931.4/41.9 16.0 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 65536 off 217.4/41.7 211.8/42.1 520.5/42.1 21.6 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 32768 off 222.5/42.1 339.9/41.7 512.0/41.2 16.1 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 16384 off 219.6/42.5 339.9/42.1 499.4/42.0 16.0 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 131072 off 204.2/39.2 319.1/39.3 480.3/39.5 22.9 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 65536 off 218.9/42.2 286.2/40.6 488.3/38.5 16.4 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 32768 off 176.8/34.6 284.8/42.1 528.9/42.0 15.6 OK
gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 16384 off 212.6/41.5 389.0/42.0 522.5/41.6 16.0 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 131072 off 263.9/69.5 420.4/70.1 641.8/70.0 17.8 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 65536 off 265.9/69.4 686.1/69.3 632.7/69.2 10.9 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 32768 off 305.0/69.5 510.7/69.7 636.5/69.3 10.4 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 16384 off 302.9/69.5 314.8/69.2 632.7/68.1 13.8 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 131072 off 280.7/65.8 298.1/65.7 602.2/65.4 15.2 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 65536 off 284.6/65.5 642.7/65.5 576.5/65.4 12.0 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 32768 off 280.9/65.3 652.1/64.6 569.5/63.6 12.2 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 16384 off 288.0/65.3 690.6/63.7 557.8/63.3 12.4 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 131072 off 262.2/63.4 658.0/65.6 608.9/65.5 11.9 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 65536 off 289.4/65.6 305.1/65.6 870.7/65.4 15.1 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 32768 off 289.7/65.2 619.3/65.2 606.4/65.2 11.5 OK
gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 16384 off 289.4/64.9 430.1/65.4 612.2/65.3 11.1 OK

qwen3.5

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

Qwen3.5-0.8B-UD-IQ2_M none 0 131072 off 140.5/175.7 346.4/175.9 690.4/174.6 7.1 OK
Qwen3.5-0.8B-UD-IQ2_M none 0 65536 off 349.3/175.1 1051.4/175.9 1812.4/175.8 3.8 OK
Qwen3.5-0.8B-UD-IQ2_M none 0 32768 off 353.0/175.5 1439.3/175.9 2042.0/175.7 4.1 OK
Qwen3.5-0.8B-UD-IQ2_M none 0 16384 off 343.2/173.7 570.3/175.7 3438.5/175.4 4.9 OK
Qwen3.5-0.8B-UD-IQ2_M none 1 131072 off 293.2/160.5 1865.5/161.2 685.3/161.3 6.3 OK
Qwen3.5-0.8B-UD-IQ2_M none 1 65536 off 296.6/160.8 2372.6/0.0 681.2/160.9 6.6 OK
Qwen3.5-0.8B-UD-IQ2_M none 1 32768 off 267.1/161.8 256.9/162.1 678.0/162.0 6.6 OK
Qwen3.5-0.8B-UD-IQ2_M none 1 16384 off 233.3/159.8 819.6/160.8 684.7/160.8 5.4 OK
Qwen3.5-0.8B-UD-IQ2_M layer both 131072 off 238.6/152.9 660.8/160.3 792.1/163.3 5.6 OK
Qwen3.5-0.8B-UD-IQ2_M layer both 65536 off 276.1/163.9 298.6/163.9 802.6/164.1 6.8 OK
Qwen3.5-0.8B-UD-IQ2_M layer both 32768 off 287.7/162.3 298.6/164.4 3081.2/164.1 6.6 OK
Qwen3.5-0.8B-UD-IQ2_M layer both 16384 off 213.0/163.5 1300.6/164.4 804.1/164.0 5.5 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 0 131072 off 238.6/161.0 359.9/163.9 965.5/163.6 7.6 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 0 65536 off 251.8/162.7 536.5/162.8 951.2/162.7 6.6 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 0 32768 off 142.0/162.8 357.6/162.8 990.8/162.7 6.6 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 0 16384 off 152.1/162.7 358.2/162.8 990.2/162.7 6.6 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 1 131072 off 205.4/150.2 260.6/151.3 678.7/151.1 7.2 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 1 65536 off 208.9/149.9 262.3/149.9 681.2/149.8 7.2 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 1 32768 off 150.8/150.1 513.2/150.1 678.1/150.1 7.2 OK
Qwen3.5-0.8B-UD-Q4_K_XL none 1 16384 off 183.8/149.8 261.0/149.9 675.6/149.9 7.2 OK
Qwen3.5-0.8B-UD-Q4_K_XL layer both 131072 off 166.1/143.6 298.3/146.0 778.9/145.7 7.7 OK
Qwen3.5-0.8B-UD-Q4_K_XL layer both 65536 off 169.0/152.9 301.5/153.5 802.8/153.1 7.2 OK
Qwen3.5-0.8B-UD-Q4_K_XL layer both 32768 off 171.6/152.9 304.3/153.7 805.5/153.2 7.1 OK
Qwen3.5-0.8B-UD-Q4_K_XL layer both 16384 off 164.4/152.0 299.7/152.5 2021.4/152.0 7.0 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 0 131072 off 266.9/134.6 336.2/135.8 2332.9/135.7 9.8 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 0 65536 off 267.2/134.9 337.0/135.1 899.7/134.9 7.7 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 0 32768 off 153.0/135.3 337.3/135.2 897.0/135.1 7.8 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 0 16384 off 278.5/135.0 342.8/135.0 883.3/134.9 7.5 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 1 131072 off 180.9/127.2 248.8/126.9 629.6/126.9 8.4 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 1 65536 off 202.9/125.7 376.4/125.8 628.9/125.7 7.5 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 1 32768 off 222.9/125.2 713.0/125.3 643.7/125.2 7.6 OK
Qwen3.5-0.8B-UD-Q8_K_XL none 1 16384 off 235.2/125.6 249.6/125.6 641.1/125.5 8.4 OK
Qwen3.5-0.8B-UD-Q8_K_XL layer both 131072 off 177.1/120.8 542.3/121.8 726.5/121.8 8.9 OK
Qwen3.5-0.8B-UD-Q8_K_XL layer both 65536 off 163.5/125.7 282.3/124.7 733.4/124.5 8.5 OK
Qwen3.5-0.8B-UD-Q8_K_XL layer both 32768 off 163.1/128.6 287.0/128.7 750.1/128.7 8.3 OK
Qwen3.5-0.8B-UD-Q8_K_XL layer both 16384 off 176.6/128.1 578.8/128.2 752.3/128.0 8.3 OK

qwen3.6_27B

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

Qwen3.6-27B-Q4_K_S none 0 131072 off - - - 31.0 OOM
Qwen3.6-27B-Q4_K_S none 0 65536 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S none 0 32768 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S none 0 16384 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S none 1 131072 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S none 1 65536 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S none 1 32768 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S none 1 16384 off - - - 2.7 OOM
Qwen3.6-27B-Q4_K_S layer both 131072 off - - - 6.0 OOM
Qwen3.6-27B-Q4_K_S layer both 65536 off 40.7/15.8 38.0/15.8 82.1/15.7 56.3 OK
Qwen3.6-27B-Q4_K_S layer both 32768 off 43.6/15.8 41.4/15.7 81.7/15.6 56.6 OK
Qwen3.6-27B-Q4_K_S layer both 16384 off 41.8/15.7 37.8/15.7 82.6/15.6 56.6 OK

qwen3.6_35B

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 131072 off - - - 24.6 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 65536 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 32768 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 16384 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 131072 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 65536 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 32768 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 16384 off - - - 2.3 OOM
Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 131072 off 52.2/56.9 92.9/57.2 202.7/57.1 17.8 OK
Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 65536 off 69.6/56.7 108.3/56.7 207.9/56.5 19.3 OK
Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 32768 off 68.5/56.4 107.3/56.6 208.1/56.4 19.3 OK
Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 16384 off 75.2/56.6 108.1/56.6 208.5/56.4 19.2 OK

qwen3.8_27B

quantsmgpuctxmtpturn1 (p/g t/s)turn2 (p/g t/s)turn3 (p/g t/s)wall(s)status

Qwen3.8-27B-UD-IQ2_XXS none 0 131072 off - - - 19.4 OOM
Qwen3.8-27B-UD-IQ2_XXS none 0 65536 off - - - 3.0 OOM
Qwen3.8-27B-UD-IQ2_XXS none 0 32768 off 98.4/14.2 37.4/14.1 82.0/13.5 42.7 OK
Qwen3.8-27B-UD-IQ2_XXS none 0 16384 off 99.5/13.7 37.4/13.1 81.8/12.9 49.0 OK
Qwen3.8-27B-UD-IQ2_XXS none 1 131072 off - - - 5.0 OOM
Qwen3.8-27B-UD-IQ2_XXS none 1 65536 off - - - 5.0 OOM
Qwen3.8-27B-UD-IQ2_XXS none 1 32768 off 90.0/14.3 31.3/14.1 70.1/13.6 49.0 OK
Qwen3.8-27B-UD-IQ2_XXS none 1 16384 off 89.1/13.3 30.8/12.5 69.6/12.3 52.5 OK
Qwen3.8-27B-UD-IQ2_XXS layer both 131072 off 86.0/15.6 34.7/15.4 76.3/15.2 45.4 OK
Qwen3.8-27B-UD-IQ2_XXS layer both 65536 off 91.9/14.7 34.3/14.7 76.3/14.7 44.3 OK
Qwen3.8-27B-UD-IQ2_XXS layer both 32768 off 86.0/14.9 34.1/14.8 75.2/14.7 45.9 OK
Qwen3.8-27B-UD-IQ2_XXS layer both 16384 off 92.3/15.5 34.2/15.4 75.5/15.3 43.3 OK

Long context (single turn, >10k tokens)

Prompt: see long_prompt.txt (needle-in-haystack recall test).

familyquantsmgpuctxmtpprompt t/sgen t/swall(s)status

gemma4-12b gemma-4-12b-it-Q4_0 none 0 131072 off 415.9 30.7 37.6 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 131072 on(n=1) 405.2 48.8 38.7 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 131072 on(n=2) 403.5 52.4 38.6 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 131072 on(n=3) 402.4 53.1 38.7 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 131072 on(n=4) 401.9 43.0 39.3 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 65536 off 401.0 28.4 38.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 65536 on(n=1) 398.3 46.7 39.1 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 65536 on(n=2) 398.7 51.7 38.8 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 65536 on(n=3) 400.7 53.7 38.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 65536 on(n=4) 399.5 50.1 38.8 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 32768 off 397.4 29.7 38.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 32768 on(n=1) 395.0 47.8 39.3 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 32768 on(n=2) 397.6 50.6 38.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 32768 on(n=3) 397.0 48.8 39.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 32768 on(n=4) 397.5 43.2 39.3 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 16384 off 396.9 29.4 39.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 16384 on(n=1) 395.7 47.6 39.2 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 16384 on(n=2) 397.1 47.7 39.1 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 16384 on(n=3) 399.0 49.2 38.8 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 0 16384 on(n=4) 400.2 48.1 38.8 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 131072 off 412.5 28.7 39.8 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 131072 on(n=1) 393.7 46.3 41.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 131072 on(n=2) 386.5 41.0 42.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 131072 on(n=3) 387.7 45.0 42.1 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 131072 on(n=4) 389.2 43.3 42.1 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 65536 off 394.8 26.9 41.2 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 65536 on(n=1) 387.5 42.8 42.2 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 65536 on(n=2) 387.1 43.6 42.2 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 65536 on(n=3) 387.8 39.8 42.4 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 65536 on(n=4) 387.5 48.4 41.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 32768 off 391.2 27.1 41.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 32768 on(n=1) 392.1 42.3 41.7 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 32768 on(n=2) 387.5 48.8 41.6 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 32768 on(n=3) 387.5 47.0 41.7 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 32768 on(n=4) 386.7 45.3 41.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 16384 off 394.7 26.8 41.2 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 16384 on(n=1) 389.4 42.5 42.1 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 16384 on(n=2) 388.9 45.8 41.6 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 16384 on(n=3) 386.2 43.2 42.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 none 1 16384 on(n=4) 388.6 48.0 41.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 131072 off 726.7 29.9 25.8 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 131072 on(n=1) 605.9 35.3 31.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 131072 on(n=2) 603.5 29.9 31.7 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 131072 on(n=3) 599.3 31.4 31.6 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 131072 on(n=4) 602.7 21.1 33.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 65536 off 713.9 29.7 26.1 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 65536 on(n=1) 598.1 36.1 30.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 65536 on(n=2) 599.1 35.0 31.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 65536 on(n=3) 601.8 31.4 31.3 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 65536 on(n=4) 599.8 27.5 31.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 32768 off 706.8 29.6 26.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 32768 on(n=1) 598.9 35.8 30.6 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 32768 on(n=2) 597.8 31.0 31.2 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 32768 on(n=3) 595.9 29.8 31.5 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 32768 on(n=4) 596.5 31.0 31.3 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 16384 off 708.1 29.5 26.0 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 16384 on(n=1) 597.9 36.2 30.4 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 16384 on(n=2) 593.4 34.6 30.7 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 16384 on(n=3) 596.3 26.8 31.9 OK
gemma4-12b gemma-4-12b-it-Q4_0 layer both 16384 on(n=4) 596.9 25.6 32.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 131072 off 352.9 26.0 43.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=1) 353.3 34.6 43.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=2) 351.9 35.4 43.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=3) 352.7 31.9 44.2 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 131072 on(n=4) 352.7 34.5 43.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 65536 off 354.9 26.1 42.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=1) 351.5 37.7 43.4 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=2) 351.3 34.7 43.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=3) 351.3 35.3 43.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 65536 on(n=4) 353.2 26.9 44.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 32768 off 355.2 25.9 42.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=1) 352.1 40.1 43.2 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=2) 352.1 37.6 43.4 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=3) 350.8 30.1 44.4 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 32768 on(n=4) 351.1 26.7 44.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 16384 off 355.5 26.5 42.7 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=1) 350.9 36.3 43.6 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=2) 352.8 34.4 43.6 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=3) 352.0 37.2 43.5 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 0 16384 on(n=4) 351.5 27.8 44.7 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 131072 off 366.7 25.5 43.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=1) 355.2 33.6 45.2 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=2) 356.2 33.5 45.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=3) 351.6 29.5 46.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 131072 on(n=4) 353.7 26.5 46.3 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 65536 off 359.3 24.6 43.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=1) 355.1 31.8 45.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=2) 352.1 35.9 45.0 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=3) 352.6 32.0 45.3 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 65536 on(n=4) 354.5 31.7 45.2 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 32768 off 357.9 24.7 44.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=1) 353.7 33.6 45.0 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=2) 354.3 32.9 45.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=3) 349.9 31.1 45.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 32768 on(n=4) 353.1 28.7 45.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 16384 off 359.8 24.6 43.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=1) 354.0 36.5 44.7 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=2) 351.5 31.0 45.6 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=3) 351.9 30.2 45.6 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL none 1 16384 on(n=4) 352.0 34.7 45.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 131072 off 639.1 27.8 27.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=1) 531.0 29.5 34.0 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=2) 525.5 25.6 34.9 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=3) 525.3 23.5 35.3 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 131072 on(n=4) 522.1 19.6 36.5 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 65536 off 618.7 27.6 28.5 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=1) 528.3 31.8 33.5 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=2) 528.9 26.3 34.3 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=3) 525.8 22.5 35.3 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 65536 on(n=4) 526.0 20.8 35.7 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 32768 off 614.3 27.9 28.4 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=1) 525.6 31.0 33.5 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=2) 524.3 32.0 33.4 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=3) 524.7 23.9 34.8 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 32768 on(n=4) 525.1 22.1 35.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 16384 off 612.1 27.8 28.5 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=1) 524.9 31.9 33.2 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=2) 525.0 25.9 34.1 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=3) 528.6 22.2 34.7 OK
gemma4-12b gemma-4-12b-it-UD-Q2_K_XL layer both 16384 on(n=4) 523.7 27.2 34.1 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 0 131072 off - - 3.7 OOM
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 0 65536 off - - 3.6 OOM
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 0 32768 off 727.3 57.4 23.1 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 0 16384 off 709.7 51.5 23.8 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 1 131072 off - - 6.1 OOM
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 1 65536 off - - 6.1 OOM
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 1 32768 off 724.9 52.5 26.0 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M none 1 16384 off 708.0 48.5 26.4 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M layer both 131072 off 1293.5 54.7 17.2 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M layer both 65536 off 1293.4 54.6 17.1 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M layer both 32768 off 1282.3 54.8 17.0 OK
gemma4-26b gemma-4-26B-A4B-it-UD-IQ2_M layer both 16384 off 1282.9 54.8 16.9 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 131072 off - - 3.4 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=1) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=2) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=3) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 131072 on(n=4) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 65536 off - - 3.4 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=1) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=2) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=3) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 65536 on(n=4) - - 4.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 32768 off - - 3.4 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=1) - - 4.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=2) - - 4.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=3) - - 4.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 32768 on(n=4) - - 4.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 16384 off - - 4.6 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=1) - - 5.0 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=2) - - 5.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=3) - - 5.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 0 16384 on(n=4) - - 5.0 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 131072 off - - 5.5 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=1) - - 6.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=2) - - 6.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=3) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 131072 on(n=4) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 65536 off - - 5.5 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=1) - - 6.4 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=2) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=3) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 65536 on(n=4) - - 6.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 32768 off - - 5.5 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=1) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=2) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=3) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 32768 on(n=4) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 16384 off 149.8 12.5 99.1 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=1) - - 7.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=2) - - 7.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=3) - - 7.1 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS none 1 16384 on(n=4) - - 7.2 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 131072 off 262.7 13.3 62.0 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=1) - - 6.6 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=2) - - 6.6 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=3) - - 6.6 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 131072 on(n=4) - - 6.5 OOM
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 65536 off 258.9 13.2 62.8 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=1) 220.2 16.0 71.5 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=2) 220.0 15.3 71.9 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=3) 220.0 13.2 73.2 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 65536 on(n=4) 220.7 13.7 72.7 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 32768 off 256.0 13.3 63.0 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=1) 219.9 16.0 71.3 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=2) 219.9 13.8 72.6 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=3) 219.5 14.4 72.3 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 32768 on(n=4) 220.0 13.2 73.0 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 16384 off 254.6 13.4 63.2 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=1) 218.9 16.2 71.5 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=2) 219.6 15.9 71.4 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=3) 220.2 14.1 72.3 OK
gemma4-31b gemma-4-31B-it-UD-IQ2_XXS layer both 16384 on(n=4) 219.7 10.7 75.3 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 131072 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 131072 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 65536 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 65536 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 32768 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 32768 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 16384 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 0 16384 on(n=4) - - 3.7 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 131072 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 131072 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 65536 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 65536 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 32768 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 32768 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 16384 off - - 2.9 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=1) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=2) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=3) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL none 1 16384 on(n=4) - - 3.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 131072 off - - 5.5 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=1) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=2) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=3) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 131072 on(n=4) - - 6.3 OOM
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 65536 off 245.3 12.3 67.1 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=1) 207.3 15.1 76.6 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=2) 207.2 13.8 77.3 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=3) 207.3 14.5 76.8 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 65536 on(n=4) 207.0 14.0 77.2 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 32768 off 241.0 12.2 67.9 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=1) 206.6 14.9 76.5 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=2) 206.7 14.9 76.6 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=3) 207.0 14.3 76.8 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 32768 on(n=4) 207.1 11.2 79.3 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 16384 off 241.0 12.3 67.8 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=1) 207.0 15.6 76.1 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=2) 207.1 14.3 76.8 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=3) 207.7 14.4 76.5 OK
gemma4-31b gemma-4-31B-it-UD-Q2_K_XL layer both 16384 on(n=4) 207.5 10.4 80.0 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 0 131072 off 1424.6 85.0 12.8 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 0 65536 off 1419.1 85.0 12.8 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 0 32768 off 1415.2 82.6 12.9 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 0 16384 off 1409.1 84.8 12.8 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 1 131072 off 1420.8 79.4 13.3 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 1 65536 off 1395.5 78.1 13.5 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 1 32768 off 1393.3 74.5 13.5 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M none 1 16384 off 1334.5 76.5 13.9 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M layer both 131072 off 2339.7 18.3 15.3 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M layer both 65536 off 2333.8 18.5 15.0 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M layer both 32768 off 2351.6 18.3 15.0 OK
gemma4-e2b gemma-4-E2B-it-Q4_K_M layer both 16384 off 2348.7 18.5 14.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 131072 off 958.7 56.4 18.2 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=1) 948.6 77.3 17.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=2) 948.3 75.5 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=3) 948.5 86.9 19.1 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 131072 on(n=4) 947.0 82.5 19.2 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 65536 off 942.2 55.0 16.1 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=1) 942.1 72.0 19.4 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=2) 945.1 76.4 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=3) 946.9 80.1 19.2 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 65536 on(n=4) 945.3 80.6 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 32768 off 937.2 54.0 18.4 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=1) 943.9 76.4 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=2) 946.2 76.4 19.3 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=3) 946.8 97.9 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 32768 on(n=4) 945.9 80.9 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 16384 off 931.8 55.0 18.4 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=1) 943.3 75.0 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=2) 946.7 75.9 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=3) 944.4 95.5 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 0 16384 on(n=4) 947.5 85.0 17.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 131072 off 964.7 51.7 16.6 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=1) 945.2 72.0 18.6 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=2) 930.7 68.0 18.8 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=3) 924.8 87.6 18.8 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 131072 on(n=4) 916.5 77.7 19.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 65536 off 921.0 47.8 19.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=1) 923.9 67.8 20.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=2) 937.7 84.9 20.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=3) 918.3 67.3 20.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 65536 on(n=4) 911.5 65.6 20.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 32768 off 925.4 46.7 17.2 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=1) 929.2 67.6 18.8 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=2) 914.2 71.3 19.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=3) 931.0 92.9 18.5 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 32768 on(n=4) 921.1 76.4 20.4 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 16384 off 922.1 47.6 19.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=1) 918.5 68.7 20.4 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=2) 921.6 66.1 20.5 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=3) 924.5 85.4 18.6 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M none 1 16384 on(n=4) 901.6 64.5 21.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 131072 off 1697.8 53.3 13.2 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=1) 1432.4 66.3 14.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=2) 1429.3 65.2 16.3 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=3) 1430.5 54.5 16.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 131072 on(n=4) 1430.8 64.6 16.3 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 65536 off 1695.8 52.5 10.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=1) 1428.5 65.9 16.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=2) 1426.1 65.4 16.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=3) 1426.6 78.4 14.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 65536 on(n=4) 1428.6 65.6 14.2 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 32768 off 1692.2 53.2 12.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=1) 1424.0 67.9 14.0 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=2) 1423.1 64.9 15.8 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=3) 1425.7 60.2 15.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 32768 on(n=4) 1427.5 65.6 13.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 16384 off 1650.6 51.8 10.9 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=1) 1423.1 68.0 13.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=2) 1423.0 62.9 13.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=3) 1423.9 78.7 13.7 OK
gemma4-e4b gemma-4-E4B-it-Q4_K_M layer both 16384 on(n=4) 1420.6 62.1 15.7 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 131072 off 1024.0 40.8 17.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 65536 off 1022.1 40.7 17.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 32768 off 1022.5 40.7 17.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 0 16384 off 1019.5 40.8 17.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 131072 off 1024.6 39.9 18.5 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 65536 off 1018.0 38.6 18.7 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 32768 off 1014.1 37.7 18.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL none 1 16384 off 1003.5 37.7 18.9 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 131072 off 1037.1 39.9 18.2 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 65536 off 1033.6 39.6 18.2 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 32768 off 1032.2 38.6 18.3 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q2_K_XL layer both 16384 off 1031.6 39.5 18.0 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 131072 off 1051.2 63.5 16.7 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 65536 off 1041.5 60.0 14.6 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 32768 off 1025.4 58.2 16.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 0 16384 off 1034.2 58.7 16.7 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 131072 off 1046.0 59.7 17.4 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 65536 off 1026.5 57.1 17.7 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 32768 off 1022.3 55.6 17.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL none 1 16384 off 1017.2 58.1 15.7 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 131072 off 1850.1 59.4 10.1 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 65536 off 1855.4 59.9 12.0 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 32768 off 1848.5 59.3 9.8 OK
gemma4-e4b-qat gemma-4-E4B-it-qat-UD-Q4_K_XL layer both 16384 off 1849.1 59.9 11.8 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 0 131072 off 4451.0 161.5 5.0 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 0 65536 off 4453.3 155.3 4.3 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 0 32768 off 4448.8 161.6 5.0 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 0 16384 off 4449.4 161.7 5.0 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 1 131072 off 4420.6 148.2 5.3 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 1 65536 off 4407.3 149.0 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 1 32768 off 4393.5 149.1 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M none 1 16384 off 4388.0 147.9 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M layer both 131072 off 7569.6 151.9 4.3 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M layer both 65536 off 7531.3 150.5 4.3 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M layer both 32768 off 7533.3 151.2 4.0 OK
qwen3.5 Qwen3.5-0.8B-UD-IQ2_M layer both 16384 off 7511.7 150.3 4.0 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 0 131072 off 4460.4 151.0 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 0 65536 off 4436.3 150.7 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 0 32768 off 4422.0 150.1 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 0 16384 off 4415.1 150.3 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 1 131072 off 4387.4 139.5 5.4 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 1 65536 off 4375.3 139.4 5.4 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 1 32768 off 4353.9 139.5 5.4 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL none 1 16384 off 4349.8 139.3 5.4 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL layer both 131072 off 7328.9 142.1 4.4 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL layer both 65536 off 7368.6 142.4 4.4 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL layer both 32768 off 7343.5 141.4 4.3 OK
qwen3.5 Qwen3.5-0.8B-UD-Q4_K_XL layer both 16384 off 7363.9 140.2 4.3 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 0 131072 off 4128.0 126.5 5.7 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 0 65536 off 4109.7 126.4 5.7 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 0 32768 off 4084.1 126.3 5.5 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 0 16384 off 4090.5 125.8 5.7 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 1 131072 off 4092.0 118.4 5.9 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 1 65536 off 4062.2 118.1 6.0 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 1 32768 off 4054.0 117.9 6.0 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL none 1 16384 off 4026.7 113.0 6.0 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL layer both 131072 off 6803.8 120.2 5.1 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL layer both 65536 off 6906.1 120.4 4.8 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL layer both 32768 off 6870.5 120.5 4.8 OK
qwen3.5 Qwen3.5-0.8B-UD-Q8_K_XL layer both 16384 off 6862.7 119.3 4.8 OK
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 0 131072 off - - 2.7 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 0 65536 off - - 2.7 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 0 32768 off - - 2.7 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 0 16384 off - - 2.5 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 1 131072 off - - 2.5 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 1 65536 off - - 2.7 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 1 32768 off - - 2.5 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S none 1 16384 off - - 2.7 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S layer both 131072 off - - 5.9 OOM
qwen3.6_27B Qwen3.6-27B-Q4_K_S layer both 65536 off 397.0 14.7 44.1 OK
qwen3.6_27B Qwen3.6-27B-Q4_K_S layer both 32768 off 386.4 14.6 44.9 OK
qwen3.6_27B Qwen3.6-27B-Q4_K_S layer both 16384 off 386.7 14.7 44.8 OK
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 131072 off - - 2.4 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 65536 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 32768 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 0 16384 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 131072 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 65536 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 32768 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL none 1 16384 off - - 2.3 OOM
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 131072 off 1415.3 53.2 15.9 OK
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 65536 off 1406.3 53.1 15.7 OK
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 32768 off 1388.6 51.4 13.8 OK
qwen3.6_35B Qwen3.6-35B-A3B-UD-Q2_K_XL layer both 16384 off 1401.5 53.1 15.7 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 0 131072 off - - 3.0 OOM
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 0 65536 off - - 2.9 OOM
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 0 32768 off 207.2 11.9 64.0 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 0 16384 off 207.1 12.3 69.3 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 1 131072 off - - 5.0 OOM
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 1 65536 off - - 5.0 OOM
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 1 32768 off 206.2 11.6 68.0 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS none 1 16384 off 204.5 11.4 73.0 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS layer both 131072 off 379.1 14.1 44.1 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS layer both 65536 off 371.6 13.3 40.7 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS layer both 32768 off 367.4 13.7 45.0 OK
qwen3.8_27B Qwen3.8-27B-UD-IQ2_XXS layer both 16384 off 366.7 13.9 41.2 OK