Unannounced AMD Instinct MI388X Accelerator Pops Up in SEC Filing

[T0@st]

AMD's Instinct family has welcomed a new addition—the MI388X AI accelerator—as discovered in a lengthy regulatory 10K filing (submitted to the SEC). The document reveals that the unannounced SKU—along with the MI250, MI300X and MI300A integrated circuits—cannot be sold to Chinese customers due to updated US trade regulations (new requirements were issued around October 2023). Versal VC2802 and VE2802 FPGA products are also mentioned in the same section. Earlier this month, AMD's Chinese market-specific Instinct MI309 package was deemed to be too powerful for purpose by the US Department of Commerce.

AMD has not published anything about the Instinct MI388X's official specification, and technical details have not emerged via leaks. The "X" tag likely implies that it has been designed for AI and HPC applications, akin to the recently launched MI300X accelerator. The designation of a higher model number could (naturally) point to a potentially more potent spec sheet, although Tom's Hardware posits that MI388X is a semi-custom spinoff of an existing model.



View at TechPowerUp Main Site | Source

 

[AnarchoPrimitiv]

I'm am still DESPERATELY awaiting performance figures and benchmarks for the MI300's...when are we going to see them?

 

[Wirko]

I'm am still DESPERATELY awaiting performance figures and benchmarks for the MI300's...when are we going to see them?

All the 300s AMD can produce go to a handful of hyper-large customers.

 

[Daven]

The full fledged CDNA3 architecture enables up to 320 CUs. Maybe this is the fully unlocked SKU with more and faster HBM.

 

[Denver]

Adding the area of the GCDs (115mm²) that makes up the mi300x, it must be almost half the size of the B200, I wonder if they could double the number of GCDs without too many problems.

View attachment 1000013024.webp

 

[wNotyarD]

The full fledged CDNA3 architecture enables up to 320 CUs. Maybe this is the fully unlocked SKU with more and faster HBM.

Isn't the 388X the stripped down SKU to comply with US restrictions so it can be sold to China?

 

[Denver]

Isn't the 388X the stripped down SKU to comply with US restrictions so it can be sold to China?

AMD Stalls on Instinct MI309 China AI Chip Launch Amid US Export Hurdles | TechPowerUp

Apparently the chip aimed at China was the one mentioned in the article above. Furthermore, the text in the Printscreen above mentions that the mi388X was also blocked by sanctions, so it already existed, it was not created afterwards

 

[wNotyarD]

AMD Stalls on Instinct MI309 China AI Chip Launch Amid US Export Hurdles | TechPowerUp

Apparently the chip aimed at China was the one mentioned in the article above. Furthermore, the text in the Printscreen above mentions that the mi388X was also blocked by sanctions, so it already existed, it was not created afterwards

Messed up the numbers then. My mistake.

 

[Vya Domus]

it must be almost half the size of the B200

B200 is actually 2 chips, both of those chips look to be close to the reticle limit, doubling the GDC means adding another chip.

 

[Denver]

B200 is actually 2 chips, both of those chips look to be close to the reticle limit, doubling the GDC means adding another chip.

I know, it's essentially twice the size of the mi300x. I was contemplating whether it would be feasible to double the number of GCDs. I believe it's achievable, albeit requiring a new interposer. Nonetheless, this approach would undoubtedly yield faster results compared to developing an AI accelerator based on a different architecture.