CPU’s weren’t the one factor mentioned at AMD’s Monetary Analyst Day. AMD positively desires to impress everybody with its upcoming Zen 4 CPUs (opens in new tab), however a gaming rig is ineffective and not using a succesful graphics card, and early indicators are that AMD’s subsequent era RDNA 3 graphics playing cards can be very aggressive certainly.
AMD hasn’t mentioned a lot about RDNA 3. The playing cards are nonetheless many months from seeing the sunshine of day, however as we get nearer, AMD felt it was time to provide a quick high-level overview of what we will count on. The factor that jumps out from the solitary slide under is a focused higher than 50% efficiency per watt uplift. That’s a really spectacular quantity certainly. Let’s assume a hypothetical RX 7900 XT is 50% sooner than a RX 6900 XT (opens in new tab) at 300W. What may it do at 400W, or 450W? Wowsers.
Subsequent up is the affirmation that AMD will migrate to a chiplet design. It has been a type of open secret (opens in new tab) that it could take this path, however that is the primary time we’ve got official affirmation. After all, we don’t but know particularly what AMD will construct ultimately. The calls for of a gaming GPU are very totally different from these of a CPU. It’s an enormous step to take and would require some inventive engineering to take care of the large quantities of low latency knowledge that must be moved to totally different dies. I sit up for seeing how AMD tackles this problem and if there are any bottlenecks that consequence.
RDNA 3 GPUs gained’t be the primary GPUs with chiplet designs although. AMD’s Intuition MI200 collection HPC accelerators (opens in new tab) function a multi-chip module design. These embrace ridiculously excessive bandwidth HBM2E stacks that may ship as much as 3.2 TB/s of bandwidth, a lot increased than the 512GB/s of a 6900 XT. So, they’re probably not comparable, however the MI200 playing cards and the efficiency of their die interconnects can have given AMD invaluable insights into how MCM based mostly designs could be utilized to gaming GPUs.
RDNA 3 playing cards can be constructed on TSMC’s 5nm node. That’s a full node bounce over the 7nm node utilized by RDNA 1 and RDNA 2 GPUs. With that alone, AMD can count on to have the ability to obtain increased clocks inside a given energy funds. It’s unlikely to stay to present ranges although. The times of 300W flagship GPUs are certainly over.
Apparently, AMD revealed that the RDNA 4 structure is within the works, although there are not any different particulars past that. It’s unlikely that we’ll see RDNA 4 earlier than late 2024.
AMD hasn’t offered a launch timeframe for its RDNA 3 playing cards, although the top of the 12 months is a fairly protected wager. Within the meantime, Nvidia is getting ready its RTX 40 collection of GPUs. They’re additionally rumoured to ship large leaps ahead in efficiency (opens in new tab). The subsequent battle between AMD and Nvidia is gearing up. Intel may need to sit down out this spherical. (opens in new tab)
