After the US slapped restrictions on the sale of AI accelerators into China, Nvidia – and later Intel – launched nerfed versions of their silicon that could be shipped to the Middle Kingdom without Uncle Sam blowing his stack. Now AMD is working on an export-compliant processor of its own to sell overseas.
“China is a very important market for us, certainly across our portfolio, as we think about certainly the accelerator market,” AMD CEO Lisa Su told analysts during the Ryzen maker’s Q2 earnings call with Wall Street analysts on Tuesday.
“Our plan is to of course be fully compliant with US export controls, but we do believe there’s an opportunity to develop a product for our customer set in China that is looking for AI solutions, and we’ll continue to work in that direction.”
What this accelerator might look like, Su didn’t specify. However AMD, like Intel, has more than a few options for addressing the Chinese market. The American chip designer’s current AI roadmap includes APUs, GPUs, FPGAs, and embedded accelerators like those found in Ryzen 7040 mobile processors.
What we do know is that US export restrictions implemented last year placed limits on the interconnect bandwidth of these cards at no more than 600GB/sec. This refers to the speed at which the accelerators can exchange data – an important factor when training parameter-heavy AI workloads like large language models.
The limits barred the sale of top-spec US-made GPUs and accelerators – including Nvidia’s A100, Intel’s Gaudi2, and AMD’s MI250X, into China.
But not for long. Within a month of the rules going into effect, Nvidia announced a borked version of its A100 that halved the memory and cut the interconnect bandwidth to two thirds. The graphics giant then launched the China-friendly H800, based on its Hopper architecture, earlier this year.
- Intel woos China with nerfed Habana Gaudi2 AI chips
- Make chips, not trade wars, says Semiconductor Industry Association
- Chips ahoy! US and China locked in self-destructive battle of trade restrictions
- AMD hopes for rebound in the second half of the year as Q2 profits plunge 94%
Intel joined the party in July with the launch of the Gaudi2 HL-225B, which was designed specifically for the Chinese market. To comply with US trade restrictions, Intel curbed the interconnect bandwidth by roughly 17 percent.
For AMD to sell its existing MI200-series parts in China, the biz would also need to cut back its GPU interconnect speed. Judging from the MI250X’s spec sheet, the house of Zen would likely need to chop off at least two of the chip’s eight 100GB/sec Infinity Fabric links.
However, the degree to which chipmakers may need to nerf their chips could end up a moving target, if the Biden administration enacts even stiffer restrictions on the sale of AI accelerators to China.
Last week, two House reps – Mike Gallagher (R-WI) and Raja Krishnamoorthi (D-IL) – both of whom serve on a select committee monitoring the Chinese Communist Party, made the case that existing trade restrictions didn’t go far enough.
The duo argued that the sale of US AI processors to other countries still poses a threat to American interests, and that the interconnect speed limits should be even lower.
Whether the White House moves forward with tighter restrictions is still up for debate – particularly with an election coming up. Several notable chip titans, including Intel and Qualcomm, recently traveled to Washington to discuss the impact of stricter rules on their businesses. ®
After the US slapped restrictions on the sale of AI accelerators into China, Nvidia – and later Intel – launched nerfed versions of their silicon that could be shipped to the Middle Kingdom without Uncle Sam blowing his stack. Now AMD is working on an export-compliant processor of its own to sell overseas.
“China is a very important market for us, certainly across our portfolio, as we think about certainly the accelerator market,” AMD CEO Lisa Su told analysts during the Ryzen maker’s Q2 earnings call with Wall Street analysts on Tuesday.
“Our plan is to of course be fully compliant with US export controls, but we do believe there’s an opportunity to develop a product for our customer set in China that is looking for AI solutions, and we’ll continue to work in that direction.”
What this accelerator might look like, Su didn’t specify. However AMD, like Intel, has more than a few options for addressing the Chinese market. The American chip designer’s current AI roadmap includes APUs, GPUs, FPGAs, and embedded accelerators like those found in Ryzen 7040 mobile processors.
What we do know is that US export restrictions implemented last year placed limits on the interconnect bandwidth of these cards at no more than 600GB/sec. This refers to the speed at which the accelerators can exchange data – an important factor when training parameter-heavy AI workloads like large language models.
The limits barred the sale of top-spec US-made GPUs and accelerators – including Nvidia’s A100, Intel’s Gaudi2, and AMD’s MI250X, into China.
But not for long. Within a month of the rules going into effect, Nvidia announced a borked version of its A100 that halved the memory and cut the interconnect bandwidth to two thirds. The graphics giant then launched the China-friendly H800, based on its Hopper architecture, earlier this year.
- Intel woos China with nerfed Habana Gaudi2 AI chips
- Make chips, not trade wars, says Semiconductor Industry Association
- Chips ahoy! US and China locked in self-destructive battle of trade restrictions
- AMD hopes for rebound in the second half of the year as Q2 profits plunge 94%
Intel joined the party in July with the launch of the Gaudi2 HL-225B, which was designed specifically for the Chinese market. To comply with US trade restrictions, Intel curbed the interconnect bandwidth by roughly 17 percent.
For AMD to sell its existing MI200-series parts in China, the biz would also need to cut back its GPU interconnect speed. Judging from the MI250X’s spec sheet, the house of Zen would likely need to chop off at least two of the chip’s eight 100GB/sec Infinity Fabric links.
However, the degree to which chipmakers may need to nerf their chips could end up a moving target, if the Biden administration enacts even stiffer restrictions on the sale of AI accelerators to China.
Last week, two House reps – Mike Gallagher (R-WI) and Raja Krishnamoorthi (D-IL) – both of whom serve on a select committee monitoring the Chinese Communist Party, made the case that existing trade restrictions didn’t go far enough.
The duo argued that the sale of US AI processors to other countries still poses a threat to American interests, and that the interconnect speed limits should be even lower.
Whether the White House moves forward with tighter restrictions is still up for debate – particularly with an election coming up. Several notable chip titans, including Intel and Qualcomm, recently traveled to Washington to discuss the impact of stricter rules on their businesses. ®
After the US slapped restrictions on the sale of AI accelerators into China, Nvidia – and later Intel – launched nerfed versions of their silicon that could be shipped to the Middle Kingdom without Uncle Sam blowing his stack. Now AMD is working on an export-compliant processor of its own to sell overseas.
“China is a very important market for us, certainly across our portfolio, as we think about certainly the accelerator market,” AMD CEO Lisa Su told analysts during the Ryzen maker’s Q2 earnings call with Wall Street analysts on Tuesday.
“Our plan is to of course be fully compliant with US export controls, but we do believe there’s an opportunity to develop a product for our customer set in China that is looking for AI solutions, and we’ll continue to work in that direction.”
What this accelerator might look like, Su didn’t specify. However AMD, like Intel, has more than a few options for addressing the Chinese market. The American chip designer’s current AI roadmap includes APUs, GPUs, FPGAs, and embedded accelerators like those found in Ryzen 7040 mobile processors.
What we do know is that US export restrictions implemented last year placed limits on the interconnect bandwidth of these cards at no more than 600GB/sec. This refers to the speed at which the accelerators can exchange data – an important factor when training parameter-heavy AI workloads like large language models.
The limits barred the sale of top-spec US-made GPUs and accelerators – including Nvidia’s A100, Intel’s Gaudi2, and AMD’s MI250X, into China.
But not for long. Within a month of the rules going into effect, Nvidia announced a borked version of its A100 that halved the memory and cut the interconnect bandwidth to two thirds. The graphics giant then launched the China-friendly H800, based on its Hopper architecture, earlier this year.
- Intel woos China with nerfed Habana Gaudi2 AI chips
- Make chips, not trade wars, says Semiconductor Industry Association
- Chips ahoy! US and China locked in self-destructive battle of trade restrictions
- AMD hopes for rebound in the second half of the year as Q2 profits plunge 94%
Intel joined the party in July with the launch of the Gaudi2 HL-225B, which was designed specifically for the Chinese market. To comply with US trade restrictions, Intel curbed the interconnect bandwidth by roughly 17 percent.
For AMD to sell its existing MI200-series parts in China, the biz would also need to cut back its GPU interconnect speed. Judging from the MI250X’s spec sheet, the house of Zen would likely need to chop off at least two of the chip’s eight 100GB/sec Infinity Fabric links.
However, the degree to which chipmakers may need to nerf their chips could end up a moving target, if the Biden administration enacts even stiffer restrictions on the sale of AI accelerators to China.
Last week, two House reps – Mike Gallagher (R-WI) and Raja Krishnamoorthi (D-IL) – both of whom serve on a select committee monitoring the Chinese Communist Party, made the case that existing trade restrictions didn’t go far enough.
The duo argued that the sale of US AI processors to other countries still poses a threat to American interests, and that the interconnect speed limits should be even lower.
Whether the White House moves forward with tighter restrictions is still up for debate – particularly with an election coming up. Several notable chip titans, including Intel and Qualcomm, recently traveled to Washington to discuss the impact of stricter rules on their businesses. ®
After the US slapped restrictions on the sale of AI accelerators into China, Nvidia – and later Intel – launched nerfed versions of their silicon that could be shipped to the Middle Kingdom without Uncle Sam blowing his stack. Now AMD is working on an export-compliant processor of its own to sell overseas.
“China is a very important market for us, certainly across our portfolio, as we think about certainly the accelerator market,” AMD CEO Lisa Su told analysts during the Ryzen maker’s Q2 earnings call with Wall Street analysts on Tuesday.
“Our plan is to of course be fully compliant with US export controls, but we do believe there’s an opportunity to develop a product for our customer set in China that is looking for AI solutions, and we’ll continue to work in that direction.”
What this accelerator might look like, Su didn’t specify. However AMD, like Intel, has more than a few options for addressing the Chinese market. The American chip designer’s current AI roadmap includes APUs, GPUs, FPGAs, and embedded accelerators like those found in Ryzen 7040 mobile processors.
What we do know is that US export restrictions implemented last year placed limits on the interconnect bandwidth of these cards at no more than 600GB/sec. This refers to the speed at which the accelerators can exchange data – an important factor when training parameter-heavy AI workloads like large language models.
The limits barred the sale of top-spec US-made GPUs and accelerators – including Nvidia’s A100, Intel’s Gaudi2, and AMD’s MI250X, into China.
But not for long. Within a month of the rules going into effect, Nvidia announced a borked version of its A100 that halved the memory and cut the interconnect bandwidth to two thirds. The graphics giant then launched the China-friendly H800, based on its Hopper architecture, earlier this year.
- Intel woos China with nerfed Habana Gaudi2 AI chips
- Make chips, not trade wars, says Semiconductor Industry Association
- Chips ahoy! US and China locked in self-destructive battle of trade restrictions
- AMD hopes for rebound in the second half of the year as Q2 profits plunge 94%
Intel joined the party in July with the launch of the Gaudi2 HL-225B, which was designed specifically for the Chinese market. To comply with US trade restrictions, Intel curbed the interconnect bandwidth by roughly 17 percent.
For AMD to sell its existing MI200-series parts in China, the biz would also need to cut back its GPU interconnect speed. Judging from the MI250X’s spec sheet, the house of Zen would likely need to chop off at least two of the chip’s eight 100GB/sec Infinity Fabric links.
However, the degree to which chipmakers may need to nerf their chips could end up a moving target, if the Biden administration enacts even stiffer restrictions on the sale of AI accelerators to China.
Last week, two House reps – Mike Gallagher (R-WI) and Raja Krishnamoorthi (D-IL) – both of whom serve on a select committee monitoring the Chinese Communist Party, made the case that existing trade restrictions didn’t go far enough.
The duo argued that the sale of US AI processors to other countries still poses a threat to American interests, and that the interconnect speed limits should be even lower.
Whether the White House moves forward with tighter restrictions is still up for debate – particularly with an election coming up. Several notable chip titans, including Intel and Qualcomm, recently traveled to Washington to discuss the impact of stricter rules on their businesses. ®