Nvidia Revs North American Channel To Promote Banned-In-China A800 GPU

November 03, 2023, 11:52 AM EDT

The AI chip big begins to reallocate the A800, initially designed for China, to North America and different areas after expanded U.S. export restrictions halted shipments of the GPU to the Asian nation final month.

Nvidia is getting companions in North America and different areas set as much as promote the A800 GPU, an AI chip initially designed to sidestep U.S. export restrictions towards China earlier than new guidelines halted gross sales to the nation final month.

Nvidia companions, together with U.S.-based electronics producer PNY Applied sciences and system integrator Colfax Worldwide, have begun selling this week the introduction of the Nvidia A800 40GB Energetic PCIe card, which the chip designer is looking the “final workstation growth platform for AI, knowledge science and high-performance computing” on its American web site.

[Related: Analysis: Nvidia, AMD Give Partners New AI Selling Points For GPUs In PCs]

PNY started promoting the Nvidia A800 40GB Energetic GPU on Monday via companions in North America, Latin America, Europe, the Center East, Africa and India, a spokesperson informed CRN. Excluded nations embrace China and dozens of others, equivalent to Russia, Cuba, Iran, Iraq and Vietnam (full listing beneath).

Different companions overtly selling the A800 40GB card embrace Japan-based ASK Corp. and Elsa.

A U.S. distribution govt informed CRN he expects the A800 to begin transport within the subsequent few weeks.

“I count on they’ll promote out pretty rapidly given the demand for Nvidia’s high-end AI GPUs,” mentioned Kent Tibbils, vice chairman of promoting at Fremont, Calif.-based ASI. “General, the AI/ML market remains to be driving server development throughout a number of markets, and we count on this to proceed via 2024.”

Nvidia didn’t reply to a request for remark.

The A800’s Historical past: Constructed For China, Then Banned A Yr Later

Nvidia initially designed the A800 to fulfill U.S. export restriction guidelines set towards China for AI chips in October 2022 after the American authorities ordered the corporate to cease promoting its strongest knowledge middle GPUs, the A100 and H100, to prospects within the Asian nation.

The primary objective of the export restrictions, set by the U.S. Division of Commerce, is to forestall China from getting access to state-of-the-art applied sciences to spice up its army.

Utilizing the identical Ampere structure that powers the A100, Nvidia launched the A800 to Chinese language prospects a 12 months in the past, sidestepping U.S. export restrictions on the time by designing the GPU to supply a chip-to-chip knowledge switch fee decrease than the brink for AI chips focused by the sanctions.

Whereas the A100 has a chip-to-chip bandwidth of 600 GB/s, which is the minimal threshold for chips banned by U.S. export restrictions towards China, the A800’s chip-to-chip bandwidth falls beneath that threshold with solely 400 GB/s.

Nvidia subsequently launched the H800 as a substitute for the corporate’s most up-to-date flagship knowledge middle GPU, H100, and just like the A800, it has a chip-to-chip bandwidth underneath 600 GB/s.

Rivals Intel and AMD reportedly pursued related methods after the U.S. export restrictions prevented the businesses from promoting highly effective AI chips into China.

Final month, the U.S. authorities successfully banned Nvidia from promoting the A800 and H800 into China by increasing export restrictions to incorporate AI chips that exceed a sure efficiency degree when a number of chips are related inside a system, which is essential to coaching more and more giant AI fashions.

The principles additionally impacted Nvidia’s lately launched L40S GPU in addition to chips from Intel and AMD.

The Commerce Division mentioned the brand new efficiency density parameter is supposed to forestall corporations from introducing workarounds that will enable Chinese language companies to buy a “bigger variety of smaller datacenter AI chips which, if mixed, can be equally as highly effective as restricted chips.”

The brand new U.S. guidelines enacted in October additionally restricted the export of superior chips to a broader set of nations, together with Iran and Russia, in line with Reuters.

The Wall Avenue Journal reported on Monday that the expanded export restrictions resulted in Nvidia cancelling greater than $5 billion in AI chip orders set for Chinese language prospects subsequent 12 months.

An Nvidia spokesperson informed the newspaper that the chip designer was within the technique of reallocating provide of impacted AI chips, such because the A800, to the U.S. and different areas.

Nvidia Pitches A800 40GB PCIe Card As ‘Final’ Workstation Platform

Nvidia’s A800 40GB Energetic product is a dual-slot PCIe card that comprises the A800 GPU. Along with the PCIe card, the A800 was additionally offered within the SXM type issue for servers in China.

Whereas the PCIe and SXM variations of the A800 have been primarily designed to energy servers in China, Nvidia is positioning the A800 40GB Energetic PCIe card for highly effective desktop PCs referred to as workstations.

Known as the “final workstation growth platform for AI, knowledge science and high-performance computing,” the A800 40GB Energetic GPU is designed to “carry the ability of a supercomputer to your workstation and speed up end-to-end knowledge science workflows,” in line with Nvidia’s web site.

Nvidia’s A800 40GB Energetic PCIe card shares a number of of the identical specs as Nvidia’s A100 40GB PCIe card, equivalent to 6,912 CUDA cores, 432 Tensor cores, 40GB of high-bandwidth HBM2 reminiscence and a 240-watt most energy consumption.

The A800 can be able to attaining the identical 9.7 teraflops in double-precision efficiency and 19.5 teraflops in single-precision efficiency because the A100’s PCIe and SXM type components.

The primary distinction is the speed at which the A800 can talk with different A800s, with the GPU that includes an NVLink chip-to-chip bandwidth of 400 GB/s versus the A100’s 600 GB/s.

In advertising and marketing supplies, Nvidia in contrast the A800 40GB Energetic GPU to the corporate’s Quadro GV100 PCIe card, which launched in 2018, and mentioned the previous is 4.2 occasions sooner for AI inference with the BERT Massive mannequin, 90 % sooner for AI coaching with the BERT Massive mannequin, 90 % sooner for the GTC benchmark and 70 % sooner for the LAMMPS benchmark.

Nvidia mentioned the A800 40GB Energetic GPU, like different AI chips in its portfolio, comes with a three-year subscription to Nvidia AI Enterprise, the corporate’s software program suite that features AI frameworks, libraries, pre-trained fashions and instruments for growing and operating AI functions.

PNY’s Full Checklist Of Excluded Nations For A800 Energetic GPU

PNY’s full listing of excluded nations for the Nvidia A800 Energetic GPU are as follows:

Afghanistan, Armenia, Azerbaijan, Bahrain, Belarus, Burma, Cambodia, Central African Republic, China, Democratic Republic of Congo, Cuba, Cyprus, Egypt, Eritrea, Georgia, Haiti, Iran, Iraq, Jordan, Kazakhstan, North Korea, Kuwait, Kyrgyzstan, Laos, Lebanon, Libya, Macau, Moldova, Mongolia, Oman, Pakistan, Qatar, Russia, Saudi Arabia, Somalia, South Sudan, Republic of Sudan, Syria, Tajikistan, Turkmenistan, United Arab Emirates, Uzbekistan, Venezuela, Vietnam, Yemen and Zimbabwe.