Query – Hypothesis: RDNA3 + CDNA2 Architectures Thread

AMD Navi 31, the primary desktop chiplet-based GPU?
Navi 31 shaping into a real compute monster.

We have now heard rumors concerning the upcoming Navi 31 GPU for some time now. In truth, there have been rumors about Navi 41 already. The Navi 31 may be AMD’s first MCM (multi-chip module) design. NVIDIA is just too anticipated to take the identical route with its Hopper sequence, nevertheless, it stays unclear if the structure is supposed for gaming or compute workloads. However, AMD made it clear that RDNA3 has Radeon DNA and it’s for certain aiming on the gaming market.

The successor to Intuition MI100 (now not referred to as Radeon) based mostly on Arcturus GPU and CDNA structure will compete with NVIDIA’s Gx100 compute chips. The CDNA is greater than more likely to take the identical path with multi-chiplet design in some unspecified time in the future sooner or later – it’s merely simpler to synchronize easy compute workloads throughout a number of dies moderately than advanced graphics. Even Intel’s Xe-HP structure will probably be based mostly on ’tiles’, which may be the business’s first try at GPGPU chiplet design.

AMD Navi 31 had already appeared in a macOS11 leak. However now we have not heard a lot concerning the structure ever since. AMD truly did speak about RDNA 3 structure again in November, however the data supplied by AMD’s EVP Rick Bergman was solely associated to at least one matter – energy effectivity. In an interview with The Road again in November, Rick Bergman confirmed that AMD is dedicated to delivering the identical efficiency per watt enchancment over RDNA 2.

Bergman: “Let’s step again and speak about the advantages of each. So why did we goal, fairly aggressively, efficiency per watt [improvements for] our RDNA 2 [GPUs]. After which sure, now we have the identical dedication on RDNA 3.”
“So [there are] truly a whole lot of efficiencies…in case you can enhance your perf-per-watt considerably. On the pocket book aspect, that’s in fact much more apparent, since you’re in a really constrained area, you may simply deliver extra efficiency to that platform once more with out some unique cooling options…We targeted on that on RDNA 2. It’s an enormous deal with RDNA 3 as nicely.”
— AMD EVP, Rick Bergman, through The Road

A brand new rumor had appeared on January 1st. We have now solely observed it as a result of it was retweeted by 3DCenter.org. Whereas it’s inconceivable to substantiate these revelations, I believed it would nonetheless be attention-grabbing to our readers. In line with Twitter consumer @Kepler_L2, it’s rumored that Navi 31 has a twin chiplet design with 80 Compute Models every. Which means that the GPU may provide as much as 160 CUs in complete, twice as many as Navi 21
AMD RDNA 3 can also be anticipated to deliver a noticeable efficiency improve in ray tracing. Primarily based on evaluations, the RDNA 2 implementation of ray tracing {hardware} acceleration is clearly inferior to NVIDIA 2nd Gen RT core, so AMD has a whole lot of work to catch up. We’re nonetheless ready for extra particulars on AMD’s implementation of Deep Studying Tremendous Sampling for AI-based super-resolution expertise. These, nevertheless, could be based mostly on Tensor-like compute cores and AMD has to date not carried out such cores into their GPUs. Would Navi 31 provide such a kind of core? It’s unknown but.

Patent-hunting Twitter consumer @Underfox3 has not too long ago found that AMD has is already growing a expertise to synchronize workload between MCM based mostly GPUs and a brand new command processor orchestrating ray tracing pipeline for next-gen GPUs. These two could seem in RDNA 3 architectures.

AMD has by no means confirmed when precisely may RDNA 3 structure be introduced, however based mostly on the roadmaps, AMD is anticipated to speak concerning the structure by 2022. It might seem that AMD would possibly elevate a curtain on RDNA3 someplace later this 12 months.

Recent Articles


Related Stories

Leave A Reply

Please enter your comment!
Please enter your name here

Stay on op - Ge the daily news in your inbox