Jump to content

User:SHJX/sandbox

From Wikipedia, the free encyclopedia
SKU Launch GPU
Die
Transistors (billion)
Die
size
Core SM
count
Cache Fillrate[a][b] Memory Processing power (TFLOPS) Interface TDP
Config[c] Clock
(MHz)[d]
L1 L2 Pixel
(Gpx/s)
Texture
(Gtex/s)
Type Size Clock
(Gb/s)
Band-
width
(GB/s)
Bus
width
FP6/FP8
Tensor
INT8
Tensor
FP16/BF16
Tensor
TF32
Tensor
FP32 FP64 FP64
Tensor
Dense
[Sparse]
Dense
[Sparse]
Dense
[Sparse]
Dense
[Sparse]
Dense
[Sparse]
Dense
[Sparse]
Dense
[Sparse]
H100 CNX Mar 21, 2023 GH100 80 814 mm2 14,592
456:24:456
690
1845
114 28.5 MB 50 MB HBM2e 80 GB 3.2 2,039 5120-bit PCIe 5.0 x16 350 W
H100 PCIe 80GB Mar 21, 2023 14,592
456:24:456
1095
1755
114 28.5 MB 50 MB 3.2 2,039 5120-bit 350 W
H100 PCIe 96GB Mar 21, 2023 16,896
528:24:528
1665
1837
132 33 MB 50 MB HBM3 1,681 5120-bit 700 W
H100 SXM5 64GB Mar 21, 2023 16,896
528:24:528
1665
1920
132 33 MB 50 MB HBM3 64 GB 1,008 3072-bit SXM5 350 W
H100 SXM5 80GB Mar 21, 2023 16,896
528:24:528
1590
1980
132 33 MB 50 MB HBM3 80 GB 1,681 5120-bit 350 W
H100 SXM5 96GB Mar 21, 2023 16,896
528:24:528
1665
1837
132 33 MB 50 MB HBM3 96 GB 5.3 1,681 5120-bit 350 W
H200 SXM5 Nov 13, 2023 16,896
528:24:528
132 33 MB 50 MB HBM3e 141 GB 4,915.2 6144-bit ?
[3,958]
?
[3,958]
?
[1,979]
?
[989]
67
 ?
34
 ?
67
 ?
700 W
H800 PCIe 80GB Mar 21, 2023 16,896
528:24:528
1095
1755
132 33 MB 50 MB HBM2e 80 GB 5.3 2,039 5120-bit PCIe 5.0 x16 350 W
H800 SXM5 16,896
528:24:528
1095
1755
132 33 MB 50 MB HBM3 5.3 2,039 5120-bit SXM5 350 W
  1. ^ Pixel fillrate is calculated as the number of render output units (ROPs) multiplied by the base (or boost) core clock speed.
  2. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.
  3. ^ CUDA Cores
    Texture Mapping Units : Render Output Units : Tensor Cores
  4. ^ Core boost values (if available) are stated below the base value inside brackets.



SKU Launch GPU
Die
Transistors (billion)
Die
size
Core SM
count
Cache Fillrate[a][b] Memory Processing power (TFLOPS) Interface TDP
Config[c] Clock
(MHz)[d]
L1 L2 Pixel
(Gpx/s)
Texture
(Gtex/s)
Type Size Clock
(Gb/s)
Band-
width
(GB/s)
Bus
width
FP4
Tensor
FP6/FP8
Tensor
INT8
Tensor
FP16/BF16
Tensor
TF32
Tensor
FP64
Tensor
Dense [Sparse] Dense [Sparse] Dense [Sparse] Dense [Sparse] Dense [Sparse] Dense [Sparse]
B100 PCIe 192GB Mar 21, 2024 GB100 208  mm2 24,576
768:384:768
690
1845
 MB  MB HBM3e 192 GB 8,192 5120-bit 7,000 [14,000] 3,500 [7,000] 1,800 [3,500] 900 [1,800] 30 [-] PCIe 6.0 x16 700 W
B100 SMX5 192GB Mar 21, 2024 GB100 208  mm2 24,576
768:384:768
690
1845
 MB  MB HBM3e 192 GB 8,192 5120-bit SXM5 700 W
B200 PCIe 288GB Mar 21, 2024 GB100 208  mm2 24,576
768:384:768
690
1845
 MB  MB HBM3e 288 GB 8.0 8,192 5120-bit PCIe 6.0 x16 1,000 W
B200 SMX5 288GB Mar 21, 2024 GB100 208  mm2 24,576
768:384:768
690
1845
 MB  MB HBM3e 288 GB 8.0 8,192 5120-bit SXM5 1,000 W
  1. ^ Pixel fillrate is calculated as the number of render output units (ROPs) multiplied by the base (or boost) core clock speed.
  2. ^ Texture fillrate is calculated as the number of texture mapping units (TMUs) multiplied by the base (or boost) core clock speed.
  3. ^ CUDA Cores
    Texture Mapping Units : Render Output Units : Tensor Cores
  4. ^ Core boost values (if available) are stated below the base value inside brackets.