Datasets:
witcheer
/
rtx-5090-benchmarks

Dataset card Data Studio Files
xet
 Community
1
model
stringclasses
25 values
architecture
stringclasses
5 values
params_b
float64
8.95
79.7
⌀
quant
stringclasses
22 values
size_gib
float64
0
58.3
engine
stringclasses
5 values
backend
stringclasses
1 value
gpu
stringclasses
1 value
vram_gb
int64
32
32
test
stringclasses
7 values
tokens_per_sec
float64
28.9
30.7k
stddev
float64
0.04
497
⌀
date
stringdate
2026-05-28 00:00:00
2026-08-23 00:00:00
Qwen3.6-35B-A3B
MoE (3B active)
34.66
Escha-W2 2-bit
12.3
escha-sglang
CUDA
RTX 5090
32
tg128
285.7
null
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
pp128
4,559.37
24.28
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
pp512
11,697.92
39.68
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
pp2048
11,692.84
69.29
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
pp4096
11,571.07
37.62
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
pp8192
11,440.21
13.35
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
pp16384
11,197.72
21.7
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
Q4_K_M
22.83
llama.cpp
CUDA
RTX 5090
32
tg128
377.39
1.7
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
31.58
NVFP4
21.6
vLLM 0.25.1
CUDA
RTX 5090
32
tg128
410.5
null
2026-08-11
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
pp128
4,862.76
90.41
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
pp512
12,223.14
27.52
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
pp2048
12,319.38
98.47
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
pp4096
12,217.11
25.17
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
pp8192
12,122.5
36.74
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
pp16384
11,917.42
17.82
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_M
17.56
llama.cpp
CUDA
RTX 5090
32
tg128
399.26
1.8
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
pp128
4,956.77
53.01
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
pp512
12,638.4
47.63
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
pp2048
12,450.45
103.15
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
pp4096
12,324.66
27.71
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
pp8192
12,200.37
47.88
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
pp16384
11,989.76
14.96
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ2_XXS
17.54
llama.cpp
CUDA
RTX 5090
32
tg128
402.86
2.19
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
pp128
4,711.7
37.7
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
pp512
11,828.18
79.39
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
pp2048
12,001.73
51.13
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
pp4096
11,946.19
47.24
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
pp8192
11,782.47
14.2
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
pp16384
11,573.66
25.92
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ3_XXS
18.44
llama.cpp
CUDA
RTX 5090
32
tg128
390.12
2.37
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
pp128
4,779.87
36.75
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
pp512
12,029.45
74.7
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
pp2048
12,171.49
31.77
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
pp4096
12,068.53
34.41
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
pp8192
11,918.81
36.27
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
pp16384
11,705.79
57.32
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
IQ4_XS
17.62
llama.cpp
CUDA
RTX 5090
32
tg128
394.51
1.7
2026-08-12
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
pp128
4,661.01
26.33
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
pp512
11,925.81
54.11
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
pp2048
11,890.47
35.47
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
pp4096
11,782.65
31.64
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
pp8192
11,653.06
47.18
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
pp16384
11,391.36
27.45
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q3_K_M
18.46
llama.cpp
CUDA
RTX 5090
32
tg128
390.76
2.83
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
pp128
4,317.5
39.57
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
pp512
11,067.4
52.14
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
pp2048
11,098.54
55.67
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
pp4096
10,997.06
51.87
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
pp8192
10,914.82
17.75
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
pp16384
10,664.91
33.25
2026-08-13
NVIDIA-Nemotron-3.5-Lightning-30B-A3B
MoE (3B active)
32.91
Q5_K_M
25.11
llama.cpp
CUDA
RTX 5090
32
tg128
348.71
1.58
2026-08-13
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
pp128
2,787.01
291.87
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
pp512
3,451.36
47.21
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
pp2048
3,420.26
5.35
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
pp4096
3,383.03
2.88
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
pp8192
3,324.24
3.18
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
pp16384
3,205.55
5.21
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_XXS
8.38
llama.cpp
CUDA
RTX 5090
32
tg128
114.61
0.18
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
pp128
2,994.41
333.9
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
pp512
3,790.07
88.02
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
pp2048
3,766.3
2.72
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
pp4096
3,724.73
3.01
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
pp8192
3,659.91
0.81
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
pp16384
3,537.69
0.36
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ2_M
9.6
llama.cpp
CUDA
RTX 5090
32
tg128
103.3
0.14
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
pp128
2,990.75
325.66
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
pp512
3,944.05
62.59
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
pp2048
3,910.74
3.84
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
pp4096
3,862.58
2.23
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
pp8192
3,792.96
2.29
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
pp16384
3,661.88
1.38
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-IQ3_XXS
11.09
llama.cpp
CUDA
RTX 5090
32
tg128
96.23
0.13
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
pp128
3,004.1
323.1
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
pp512
3,881.34
52.05
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
pp2048
3,838.37
14.45
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
pp4096
3,777.83
4.16
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
pp8192
3,705.33
4.04
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
pp16384
3,560.3
6.35
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q4_K_M
15.92
llama.cpp
CUDA
RTX 5090
32
tg128
78.94
0.07
2026-08-14
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
pp128
2,895.97
277.62
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
pp512
3,837.74
88.6
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
pp2048
3,813.05
12.26
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
pp4096
3,765.95
10.79
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
pp8192
3,693.59
2.67
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
pp16384
3,566.36
0.68
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
UD-Q4_K_XL
16.68
llama.cpp
CUDA
RTX 5090
32
tg128
76.22
0.13
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
pp128
2,578.57
217.51
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
pp512
3,223.28
38.41
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
pp2048
3,214.67
10.49
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
pp4096
3,171.59
2.73
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
pp8192
3,123
2.19
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
pp16384
3,034.29
0.44
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q6_K
21.3
llama.cpp
CUDA
RTX 5090
32
tg128
63.08
0.05
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
pp128
2,761.64
266
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
pp512
3,833.96
64.6
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
pp2048
3,800.41
7.61
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
pp4096
3,755.65
5.24
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
pp8192
3,680.71
5.43
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
pp16384
3,543.55
2.17
2026-08-15
Qwen3.8-27B
Dense (hybrid attn)
27.32
Q8_0
27.04
llama.cpp
CUDA
RTX 5090
32
tg128
53.07
0.06
2026-08-15