|
Zestawienie procesorów NVIDIA Compute (SXM) |
|
|
Chip |
GPU /
nazwa |
Rdzenie |
Wydajność |
Pamięć |
Uwagi |
|
TFLOPS |
|
Tensor |
CUDA |
FP64 |
FP32 |
FP16 |
GB |
GB/s |
Typ |
Szyna |
|
|
Rubin Ultra
(X) |
|
|
|
|
|
|
|
HBM4 12hi |
|
|
|
X200
X100 |
GX200 |
|
|
|
|
|
|
|
HBM4 |
|
NVLink 6, 3.6TB/s,
Grace ARM
(GX200 NVL) |
|
X100 |
|
|
|
|
|
|
|
HBM4 8hi |
|
NVLink 6, 3.6TB/s |
|
B300 |
Blackwell
Ultra |
|
|
|
|
|
288
(8x36) |
|
HBM3e 12hi |
|
NVLink 5, 1.8TB/s, 1400W
(DGX B300) |
|
NVIDIA
GB300 |
|
|
|
|
|
|
|
NVLink 5,
1.8TB/s, 1400W, Grace ARM
(GB300 NVL72) |
|
B200 |
NVIDIA B200 |
|
|
40 |
80 |
|
180
(8x24) |
8000 |
HBM3e 8hi |
4096 |
NVLink 5, 1.8TB/s, 1000W
(DGX B200) |
|
NVIDIA GB200 |
|
|
40 |
80 |
|
192
(8x24) |
8000 |
HBM3e 8hi |
|
NVLink-C2C, 1.8TB/s, 1000W, Grace ARM
(GB200 NVL72) |
|
B102 |
NVIDIA
B200A |
|
|
|
|
|
144
(4x36) |
|
HBM3e 12hi |
|
|
|
B100 |
NVIDIA B100 |
|
|
30 |
60 |
|
180
(8x24) |
8000 |
HBM3e 8hi |
|
NVLink 5, 1.8TB/s
700W |
|
B10 |
NVIDIA GB10 |
|
|
|
|
|
|
|
|
|
NVlink-C2C, Grace ARM
(DGX
SPARK) |
|
H200 H100 |
NVIDIA
H200 |
|
|
34 |
67 |
|
141
(6x24) |
4800 |
HBM3e 8hi |
|
NVLink 4,
900GB/s, 700W
(DGX H200) |
|
NVIDIA
GH200 |
|
16896 |
34 |
67 |
|
141
(6x24) |
4700 |
HBM3e 8hi |
|
NVLink 4,
900GB/s, 1000W, Grace ARM
(GH200
96GB) |
|
96
(6x16) |
|
HBM3 8hi |
|
|
NVIDIA H20 |
|
|
|
44 |
|
96
(6x16) |
4000 |
HBM3 8hi |
|
400W, HGX,
Chiny |
|
NVIDIA
H100 SXM5 |
528 |
16896 |
34 |
67 |
120 |
80
(5x16) |
3000 |
HBM3 8hi |
10512 |
NVLink 4, 900GB/s, 700W
(DGX
H100) |
|
GA100 |
NVIDIA A100
SXM4 |
432 |
6912 |
9.7 |
19.5 |
|
80 |
2039 |
HBM2 |
5120 |
NVLink 3, 600GB/s
(DGX
A100) |
|
40 |
1555 |
|
NVIDIA A800
SXM4 |
432 |
6912 |
9.7 |
19.5 |
|
80 |
2039 |
400GB/s, Chiny |
|
GV100 |
Tesla V100
SXM3 |
640 |
5120 |
7.8 |
15.7 |
|
32 |
900 |
HBM2 |
4096 |
NVLink 2,
300GB/s
(DGX-2) |
|
Tesla V100
SXM2 |
NVLink 2,
50GB/s
(DGX-1) |
|
32 |
|
16 |
|
GP100 |
Tesla P100
SXM1 |
- |
3584 |
5.3 |
10.6 |
21.2 |
16 |
732 |
HBM2 |
4096 |
NVLink 1,
40 GB/s
(DGX-1) |
|
12 |
549 |
|
|
Zestawienie procesorów NVIDIA Compute (PCIe) |
|
|
Chip |
GPU |
Rdzenie |
Wydajność |
Pamięć |
Zasilanie |
Uwagi |
|
TFLOPS |
|
SM |
Tensor |
CUDA |
FP64 |
FP32 |
FP16 |
INT8 |
INT4 |
GB |
GB/s |
Typ |
Szyna |
W |
Złącze |
Adapter |
|
X10x |
GX200 NVL |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
X40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
B20x |
B40 |
|
|
|
|
|
|
|
|
96 |
|
|
|
|
|
|
|
|
B202 |
RTX PRO 6000
Blackwell Server Edition |
188 |
752 |
24064 |
1.97 |
117 |
|
|
|
96 |
1600 |
GDDR7 |
512 |
400-600 |
PCIe
CEM5 16-pin
(12V-2x6) |
od
producenta serwera |
4x DP 2.1b |
|
H200 H100 |
H200 NVL |
132 |
528 |
16896 |
34 |
67 |
|
|
|
141 |
4800 |
HBM3e |
|
600 |
PCIe
CEM5 16-pin |
od
producenta serwera |
AI |
|
H100 NVL |
114 |
456 |
14592 |
30 |
60 |
|
|
|
94 |
3360 |
HBM3 |
6016 |
400 |
1xCPU
8-pin -> 1xCEM5 16-pin (930-00030-1546-000) -
maks. 310W |
|
H100 |
24 |
48 |
96 |
|
|
80 |
2040 |
HBM2e |
5120 |
350 |
|
AD104 |
L4 |
60 |
240 |
7680 |
0.49 |
30.3 |
|
|
|
24 |
300 |
GDDR6 |
192 |
72 |
- |
- |
AI / Video |
|
L2 |
|
|
|
- |
24,1 |
|
|
|
300 |
|
|
|
Chiny |
|
AD102 |
L40S |
142 |
568 |
18176 |
1.4 |
91.6 |
|
|
|
48 |
864 |
GDDR6 |
384 |
350 |
PCIe
CEM5 16-pin |
1xCPU 8-pin
-> 1xCEM5 16-pin
(930-00030-1546-000) |
Omniverse / vWS
4x DP |
|
L40 |
1.4 |
90.5 |
|
|
|
300 |
|
L20 |
|
|
|
- |
59,8 |
|
|
|
864 |
275 |
|
|
Chiny |
|
GA107 |
A16 |
4x 20 |
4x 80 |
4x 2560 |
4x 0.1 |
4x 8.7 |
4x 8.7 |
|
|
4x 16 |
4x 232 |
GDDR6 |
4x128 |
250 |
CPU 8-pin |
|
4x GPU, vPC |
|
A2 |
10 |
40 |
1280 |
0.07 |
4.5 |
4.5 |
|
|
16 |
200 |
GDDR6 |
128 |
60 |
- |
- |
vPC |
|
GA102 |
A10 |
72 |
288 |
9216 |
0.98 |
31.2 |
31.2 |
|
|
24 |
600 |
GDDR6 |
384 |
150 |
PCIe 8-pin |
2xPCIe
6-pin -> 1xPCIe 8-pin |
Omniverse / vWS |
|
A40 |
84 |
336 |
10752 |
0.59 |
37.7 |
37.7 |
|
|
48 |
696 |
300 |
CPU 8-pin |
|
Omnivers /
vWS
3x DP |
|
GA100 |
A30 |
- |
224 |
3804 |
5.2 |
10.3 |
10.3 |
|
|
24 |
933 |
HBM2 |
3072 |
165 |
CPU 8-pin |
|
AI / vCS
NVLINK 600GB/s |
|
A100 |
- |
432 |
6912 |
9.75 |
19.5 |
78 |
624 |
1248 |
80 |
1935 |
5120 |
300 |
CPU 8-pin |
2xPCIe
8-pin -> 1xCPU 8-pin |
|
40 |
1555 |
250 |
|
A800 |
80 |
1935 |
300 |
NVLINK 400GB/s |
|
40 |
1555 |
250 |
|
TU104 |
Tesla T4 |
40 |
320 |
2560 |
0,25 |
8.1 |
65.1 |
130 |
260 |
16 |
320 |
GDDR6 |
256 |
70 |
- |
- |
|
|
TU102 |
RTX 8000 |
72 |
576 |
4608 |
- |
14.9 |
29.9 |
238.9 |
|
48 |
624 |
GDDR6 |
384 |
260 |
|
|
NVIDIA RTX 8000 Passive |
|
RTX
6000 |
- |
24 |
NVIDIA RTX 6000 Passive |
|
GV100 |
Tesla V100s |
- |
640 |
5120 |
8.4 |
16.4 |
32.7 |
- |
- |
32 |
1134 |
HBM2 |
4096 |
250 |
|
|
|
|
Tesla V100 |
7 |
14 |
28.2 |
- |
- |
32 / 16 |
900 |
|
|
GP104 |
Tesla P6 |
- |
- |
2048 |
|
6,2 |
|
|
|
16 |
192 |
GDDR5 |
256 |
|
|
|
Moduł MXM |
|
Tesla P4 |
- |
- |
2560 |
|
5.5 |
|
22 |
- |
8 |
|
|
|
|
|
GP102 |
Tesla P40 |
- |
- |
3840 |
|
12 |
|
47 |
- |
24 |
346 |
GDDR5 |
384 |
|
|
|
|
|
GP100 |
Tesla P100 |
- |
- |
3584 |
4.7 |
9.3 |
18.7 |
22 |
- |
16 |
732 |
HBM2 |
4096 |
|
|
|
|
|
12 |
549 |
|
|
|
|
|
GM206 |
Tesla M4 |
- |
- |
1024 |
|
2.2 |
|
|
|
4 |
88 |
GDDR5 |
128 |
|
|
|
|
|
GM204 |
Tesla M60 |
- |
- |
4096 |
0.3 |
8 |
|
|
|
16 |
160 |
GDDR5 |
256 |
|
|
|
2x GPU |
|
Tesla M6 |
- |
- |
1536 |
|
3,6 |
|
|
|
8 |
|
|
|
Moduł MXM |
|
GM200 |
Tesla M40 |
- |
- |
3072 |
|
7 |
|
|
|
24 / 12 |
288 |
GDDR5 |
384 |
|
|
|
|
|
GM107 |
Tesla M10 |
- |
- |
2560 |
|
6.7 |
|
|
|
32 |
83 |
GDDR5 |
256 |
|
|
|
4x GPU |
|
GK210 |
Tesla K80m |
- |
- |
4992 |
2.91 |
8.74 |
|
|
|
24 |
240 |
GDDR5 |
384 |
|
|
|
2x GPU |
|
GK110B |
Tesla K40m |
- |
- |
2880 |
1.66 |
5 |
|
|
|
12 |
288 |
GDDR5 |
384 |
|
|
|
|
|
Tesla K40C |
|
|
|
Karta aktywna |
|
|