SXM (socket)

SXM (Server PCI Express Module)[1] is a high bandwidth socket solution for connecting Nvidia Compute Accelerators to a system. Each generation of Nvidia Tesla since P100 models, the DGX computer series and the HGX boards come with an SXM socket type that realizes high bandwidth, power delivery and more for the matching GPU daughter cards.[2] Nvidia offers these combinations as an end-user product e.g. in their models of the DGX system series. Current socket generations are SXM for Pascal based GPUs, SXM2 and SXM3 for Volta based GPUs, SXM4 for Ampere based GPUs, and SXM5 for Hopper based GPUs. These sockets are used for specific models of these accelerators, and offer higher performance per card than PCIe equivalents.[2] The DGX-1 system was the first to be equipped with SXM-2 sockets and thus was the first to carry the form factor compatible SXM modules with P100 GPUs and later was unveiled to be capable of allowing upgrading to (or being pre-equipped with) SXM2 modules with V100 GPUs.[3][4]

Computing node of TSUBAME 3.0 supercomputer showing four NVIDIA Tesla P100 SXM modules
Bare SXM sockets next to sockets with GPUs installed

SXM boards are typically built with four or eight GPU slots, although some solutions such as the Nvidia DGX-2 connect multiple boards to deliver high performance. While third party solutions for SXM boards exist, most System Integrators such as Supermicro use prebuilt Nvidia HGX boards, which come in four or eight socket configurations.[5] This solution greatly lowers the cost and difficulty of SXM based GPU servers, and enables compatibility and reliability across all boards of the same generation.

SXM modules on e.g. HGX boards, particularly recent generations, may have NVLink switches to allow faster GPU-to-GPU communication. This as well reduces bottlenecks which would normally be located within CPU and PCIe.[2][6] The GPUs on the daughter cards are just using NVLink as their main communication protocol. For example a Hopper-based H100 SXM5 based GPU can use up to 900 GB/s of bandwidth across 18 NVLink 4 channels, with each contributing a 50 GB/s of bandwidth;[7] This compared to PCIe 5.0, which can handle up to 64 GB/s of bandwidth within a x16 slot.[8] This high bandwidth also means that GPUs can share memory over the NVLink bus, allowing an entire HGX board to present to the host system as a single, massive GPU.[9]

Power delivery is also handled by the SXM socket, negating the need for external power cables such as those needed in PCIe equivalent cards. This, combined with the horizontal mounting allows cooling options of higher efficiency which in turn allows the SXM based GPUs to operate at a much higher TDP. The Hopper-based H100, for example, can draw up to 700W solely from the SXM socket.[10] The lack of cabling also makes assembling and repairing of large systems much easier, and also reduces the possible points of failure.[2]

The early Nvidia Tegra automotive targeted evaluation board, 'Drive PX2', had two MXM (Mobile PCI Express Module) sockets on both sides of the card, this dual MXM design can be considered a predecessor to the Nvidia Tesla implementation of the SXM socket.

Comparison of accelerators used in DGX:[11][12][13]

ModelArchitectureSocketFP32
CUDA
cores
FP64 cores
(excl. tensor)
Mixed
INT32/FP32
cores
INT32
cores
Boost
clock
Memory
clock
Memory
bus width
Memory
bandwidth
VRAMSingle
precision
(FP32)
Double
precision
(FP64)
INT8
(non-tensor)
INT8
dense tensor
INT32FP4
dense tensor
FP16FP16
dense tensor
bfloat16
dense tensor
TensorFloat-32
(TF32)
dense tensor
FP64
dense tensor
Interconnect
(NVLink)
GPUL1 CacheL2 CacheTDPDie sizeTransistor
count
Process
B200 BlackwellN/AN/AN/AN/AN/AN/A8 Gbit/s HBM3e8192-bit8 TB/sec192 GB HBM3eN/AN/AN/A4.5 POPSN/A9 PFLOPSN/A2.25 PFLOPS2.25 PFLOPS1.2 PFLOPS40 TFLOPS1.8 TB/secGB100N/AN/A1000 WN/A208 BTSMC 4NP
B100 BlackwellN/AN/AN/AN/AN/AN/A8 Gbit/s HBM3e8192-bit8 TB/sec192 GB HBM3eN/AN/AN/A3.5 POPSN/A7 PFLOPSN/A1.98 PFLOPS1.98 PFLOPS989 TFLOPS30 TFLOPS1.8 TB/secGB100N/AN/A700 WN/A208 BTSMC 4NP
H200 HopperSXM516896460816896N/A1980 MHz6.3 Gbit/s HBM3e6144-bit4.8 TB/sec141 GB HBM3e67 TFLOPS34 TFLOPSN/A1.98 POPSN/AN/AN/A990 TFLOPS990 TFLOPS495 TFLOPS67 TFLOPS900 GB/secGH10025344 KB (192 KB × 132)51200 KB1000 W814 mm280 BTSMC 4N
H100 HopperSXM516896460816896N/A1980 MHz5.2 Gbit/s HBM35120-bit3.35 TB/sec80 GB HBM367 TFLOPS34 TFLOPSN/A1.98 POPSN/AN/AN/A990 TFLOPS990 TFLOPS495 TFLOPS67 TFLOPS900 GB/secGH10025344 KB (192 KB × 132)51200 KB700 W814 mm280 BTSMC 4N
A100 80GB AmpereSXM4691234566912N/A1410 MHz3.2 Gbit/s HBM2e5120-bit1.52 TB/sec80 GB HBM2e19.5 TFLOPS9.7 TFLOPSN/A624 TOPS19.5 TOPSN/A78 TFLOPS312 TFLOPS312 TFLOPS156 TFLOPS19.5 TFLOPS600 GB/secGA10020736 KB (192 KB × 108)40960 KB400 W826 mm254.2 BTSMC N7
A100 40GB AmpereSXM4691234566912N/A1410 MHz2.4 Gbit/s HBM25120-bit1.52 TB/sec40 GB HBM219.5 TFLOPS9.7 TFLOPSN/A624 TOPS19.5 TOPSN/A78 TFLOPS312 TFLOPS312 TFLOPS156 TFLOPS19.5 TFLOPS600 GB/secGA10020736 KB (192 KB × 108)40960 KB400 W826 mm254.2 BTSMC N7
V100 32GB VoltaSXM351202560N/A51201530 MHz1.75 Gbit/s HBM24096-bit900 GB/sec32 GB HBM215.7 TFLOPS7.8 TFLOPS62 TOPSN/A15.7 TOPSN/A31.4 TFLOPS125 TFLOPSN/AN/AN/A300 GB/secGV10010240 KB (128 KB × 80)6144 KB350 W815 mm221.1 BTSMC 12FFN
V100 16GB VoltaSXM251202560N/A51201530 MHz1.75 Gbit/s HBM24096-bit900 GB/sec16 GB HBM215.7 TFLOPS7.8 TFLOPS62 TOPSN/A15.7 TOPSN/A31.4 TFLOPS125 TFLOPSN/AN/AN/A300 GB/secGV10010240 KB (128 KB × 80)6144 KB300 W815 mm221.1 BTSMC 12FFN
P100 PascalSXM/SXM2N/A17923584N/A1480 MHz1.4 Gbit/s HBM24096-bit720 GB/sec16 GB HBM210.6 TFLOPS5.3 TFLOPSN/AN/AN/AN/A21.2 TFLOPSN/AN/AN/AN/A160 GB/secGP1001344 KB (24 KB × 56)4096 KB300 W610 mm215.3 BTSMC 16FF+

References

  1. Michael Brown, W.; et al. (2012). "An Evaluation of Molecular Dynamics Performance on the Hybrid Cray XK6 Supercomputer". Procedia Computer Science. 9: 186–195. doi:10.1016/j.procs.2012.04.020.
  2. Proud, Matt. "Achieving Maximum Compute Throughput: PCIe vs. SXM2". The Next Platform. Retrieved 2022-03-31.
  3. Volta architecture whitepaper nvidia.com
  4. DGX 1 User Guide nvidia.com
  5. servethehome (2020-05-14). "NVIDIA A100 4x GPU HGX Redstone Platform". ServeTheHome. Retrieved 2022-03-31.
  6. "NVLink & NVSwitch for Advanced Multi-GPU Communication". NVIDIA.
  7. "Nvidia's H100 – What It Is, What It Does, and Why It Matters". Data Center Knowledge | News and analysis for the data center industry. 2022-03-23. Retrieved 2022-03-31.
  8. "Is PCIe 5.0 Worth It? The Benefits of PCIe 5.0 (2022)". www.techreviewer.com. Retrieved 2022-03-31.
  9. "NVIDIA HGX A100: Powered by A100 GPUs and NVSwitch". NVIDIA. Retrieved 2022-03-31.
  10. "NVIDIA H100 GPU full details: TSMC N4, HBM3, PCIe 5.0, 700W TDP, more". TweakTown. 2022-03-23. Retrieved 2022-03-31.
  11. Smith, Ryan (March 22, 2022). "NVIDIA Hopper GPU Architecture and H100 Accelerator Announced: Working Smarter and Harder". AnandTech.
  12. Smith, Ryan (May 14, 2020). "NVIDIA Ampere Unleashed: NVIDIA Announces New GPU Architecture, A100 GPU, and Accelerator". AnandTech.
  13. "NVIDIA Tesla V100 tested: near unbelievable GPU power". TweakTown. September 17, 2017.
This article is issued from Wikipedia. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.