I suppose I should not be shocked to not get any actually juicy particulars about AMD’s imminent CPU and GPU generations from Monetary Analyst Day (FAD) 2022 (opens in new tab). In any case, it is extra about advertising the place AMD is correct now and the place it is going within the subsequent couple of years, so analysts can wax lyrical about why traders ought to throw cash at Dr. Lisa T Su et al.
A FAD then is just not a lot about what the next-generation of Ryzen CPU and Radeon GPU {hardware} goes to do, or the way it would possibly plan to implement completely different technological wizardry as a way to do it.
All we actually bought out of the completely different displays—from such AMD luminaries as Mark ‘The Papermaster’ Papermaster, Rick Bergman, and David Wang—was a number of advertising buzzwords and some extraneous performance-per-watt share good points.
Which makes it onerous to get too enthusiastic about what’s developing from the purple workforce. Although provided that the jaws of over-hyped expectation have bitten AMD within the collective posterior beforehand, possibly that is not such a foul factor.
There was, nonetheless, the affirmation we’re taking a look at an ~8% IPC uplift for the brand new Zen 4 desktop processors, which we might kinda found out from the mix of IPC and frequency making up the 15% gen-over-gen good points AMD touted at its Computex keynote final month.
We have been spoiled by earlier generational efficiency uplifts from AMD’s processor design groups, to the purpose the place a collective single thread acquire in Cinebench R23 of 35% has been largely ignored as a result of the precise Zen 4 architectural good points are anticipated to be lower than 10%. Perhaps the low-hanging fruit has already been picked from the Zen tree.
It is value remembering that, when issues have been stagnating over at Intel earlier than it bought its node recreation again on monitor, all you could possibly anticipate of what have been then the most effective new CPU generations was a ten% bump. So actually, any real-world improve over and above that determine, whether or not that is as a result of AMD’s new chips can fortunately chow down on extra energy, or as a result of they run at greater frequencies… properly, that is simply dandy.
I suppose I used to be foolishly hoping we might get one thing extra formally concrete from AMD about its next-gen GPUs than some imprecise noises about chiplet packaging (opens in new tab) and DisplayPort 2.0 help affirmation (opens in new tab). I need to be enthusiastic about AMD’s new playing cards, and never simply should divine the trickles of reality from inside the torrents of rumours, leaks, and hypothesis persistently cluttering up Twitter.
As a result of the GPU rumours sound probably very thrilling, however I now need the reality, rattling it. Yeah, I am impatient…
It is the chiplet factor which is probably the most intriguing, and one thing we have been talking about for years (opens in new tab), nevertheless it’s a disgrace there is not any meat on the advertising bones of those displays. Is AMD truly utilizing a chiplet design in the identical method because it has with Ryzen, and by that I imply is it going to have a number of precise GPU chiplets to actually pump up the shader counts?
All David Wang says in regards to the ‘superior chiplet packaging’ of the upcoming RDNA 3 GPU design is: “It permits us to proceed to scale efficiency aggressively with out the yield and the price considerations of a giant monolithic silicon.
“It permits us to ship the most effective efficiency on the proper value.”
From a monetary perspective that is what you need to hear, and is among the causes the chiplet-based Ryzen CPUs have been so profitable from each a technical and manufacturing standpoint.
However I spoke to Wang just a few years in the past a couple of potential shift to MCM designs, and he laid out the difficulties inherent in utilizing a number of GPU chiplets in gaming and visible phrases.
“To some extent you’re speaking about doing CrossFire on a single package deal. The problem is that until we make it invisible to the ISVs [independent software vendors] you’re going to see the identical type of reluctance.
“We’re happening that path on the CPU aspect,” Wang continues, “and I believe on the GPU we’re at all times taking a look at new concepts. However the GPU has distinctive constraints with one of these NUMA [non-uniform memory access] structure, and the way you mix options… The multithreaded CPU is a bit simpler to scale the workload. The NUMA is a part of the OS help so it’s a lot simpler to deal with this multi-die factor relative to the graphics sort of workload.”
You possibly can see that within the affirmation of 3D chiplet packaging for the CDNA 3 GPUs for the Radeon Intuition crew, that we’re prone to see a number of compute chiplets on that aspect of issues, however on the RDNA 3 aspect, I am not satisfied multi-GPU is prepared for prime time gaming in a chiplet kind.
So, what might this chiplet setup be? The newest rumours counsel, even for the highest two Navi 3x GPUs that are anticipated to be chiplet-based, these will solely include a single graphics compute die (GCD) with the precise chiplet packaging noises referencing the splitting off of the GPU’s cache onto separate multi cache dies (MCDs).
That may make issues easy from a straight gaming efficiency standpoint, and truly imply you possibly can focus many of the die-size of a GCD on the precise core graphics stuff, shaving off the area you’d usually have for the cache reminiscence. And if the IO and reminiscence bus interconnects are shifted off the GCD that additionally frees up numerous die area, too.
However once more we’re resorting to rumours to get engaged and excited in regards to the prospects. I am simply impatient; I am positive it will not be that lengthy till AMD truly offers us one thing chonky, architecture-wise, to get our tooth into. Perhaps.
Although with the anticipated RDNA 3 launch window to begin someday in October, and that solely being a mid-range kick off present, and a monolithic one at that, let’s simply say I am not holding my breath.