Wednesday, November 11th 2020
AMD Looks to Keep Performance, Efficiency Gains Momentum With Zen 4, RDNA 3, and Commitment to Threadripper
AMD's Executive Vice President Rick Bergman in an interview with The Street shed some light on the company's future plans for Zen 4 and RDNA 3, even as we are still reeling from (or coming in to) Zen 3 and RDNA 2's launches. Speaking on RDNA 3, Rick Bergman mentioned the company's commitment to achieve the same 50% performance-per-watt increase they achieved with RDNA 2, and had some interesting takes on the matter of why this is actually one of the most important metrics:
Finally, Rick Bergman had this to say regarding Threadripper, which has been absent from AMD's talks on Zen 3:
Sources:
The Street, via TechSpot
Rick BergmanIt just matters so much in many ways, because if your power is too high -- as we've seen from our competitors -- suddenly our potential users have to buy bigger power supplies, very advanced cooling solutions. And in a lot of ways, very importantly, it actually drives the [bill of materials] of the board up substantially. This is a desktop perspective. And invariably, that either means the retail price comes up, or your GPU cost has to come down. We focused on that for RDNA 2. It's a big focus on RDNA 3 as well.Rick Bergman then went on to talk about how exactly AMD's Infinity Cache played a crucial role in achieving these performance-per-watt improvements: it's basically a way to prevent power consumption that results from the need to address VRAM and achieving higher memory bandwidths exclusively from larger buses or faster VRAM speeds:
Rick BergmanOn Infinity Cache, it's somewhat linked to that as well, to a certain degree. If you've been in graphics for a long time, you realize there's a pretty good correlation between memory bandwidth and performance. And so typically, the way you do it is you jack up your memory speed and widen your bus to open up performance. Unfortunately, both of those things drive up power.On raytracing, Rick Bergman is confident on AMD's design goals, which were looking at the 1440p resolution as the performance target for high-quality raytracing effects - and doubles down on AMD's advantage of having its raytracing technology shipped with both next-generation consoles in the Xbox family and PS5.
Rick Bergman[1440p]…that was kind of the performance level that we targeted. Now it depends on particular games and everybody's systems and so on, but I think you'll find that we have very good raytracing performance overall. And the game support will be strong as we go through 2021, because again, we get that great leverage. It's just built in: You support raytracing on Microsoft or Sony [consoles], you're supporting AMD on the PC side as well."On AMD's response to NVIDIA's DLSS, which the company is calling FSR/FidelityFX Super Resolution (which will be joining AMD's Contrast Adaptive Sharpening), Bergman underlined AMD's approach to an open standard, in opposition to a proprietary one, which would only require ISV adoption:
Rick BergmanWe don't have a lot of details that we want to talk about. So we called [our solution] FSR — FidelityFX Super Resolution. But we are committed to getting that feature implemented, and we're working with ISVs at this point. I'll just say AMD's approach on these types of technologies is to make sure we have broad platform support, and not require proprietary solutions [to be supported by] the ISVs. And that's the approach that we're taking. So as we go through next year, you'll get a lot more details on it.Moving on to AMD's Zen 4, AMD is counting (naturally) on building upon the success of Zen 3, and Bergman thinks that the current state of x86 demands improvements in power efficiency and instructions per clock - which can be achieved across all areas of chip development, including core counts, frequency optimizations, and overall architecture revisions:
Rick Bergman[Given] the maturity of the x86 architecture now, the answer has to be, kind of, all of the above. If you looked at our technical document on Zen 3, it was this long list of things that we did to get that 19% [IPC uplift]. Zen 4 is going to have a similar long list of things, where you look at everything from the caches, to the branch prediction, [to] the number of gates in the execution pipeline. Everything is scrutinized to squeeze more performance out."Naturally, the change in process node from current TSMC's 7 nm to the company's 5 nm process with Zen 4 will also factor-in on the performance-per-watt improvement, and the amount of logic that AMD can cram into the same puny die space as their current Zen 3 CPUs require. However, Bergman didn't clarify whether AMD's RDNA3 would also benefit from the 5 nm process transition that is all but set in stone for Zen 4:
Rick BergmanNothing to disclose at this time. GPUs are increasingly complex, they're very logic-based, so they do take advantage of the advanced process nodes. But CPUs love [them] for the reasons we were just talking about --- for the IPC and the frequency [gains]. So we look at our product lineup and where the technology is, and how we want to manage risk and kind of pick the right product at the right time. For a lot of reasons I mentioned, [CPUs and GPUs] tend to line up at fairly similar timeframes at the end of the day, because we want to take advantage. We also look at when our foundry is really going to be ready for the type of volumes in [the] quality that we demand. And we don't land in two different timescales.It seems almost certain that AMD will prioritize the Zen 4 transition to a denser node rather than that of RDNA3, should they have to choose between one or the other. One need only look at CPU and GPU pricing and compare those with the overall die sizes to understand - put quite simply - that AMD achieves much higher margins on a n (estimated) 180 mm² - 250 mm² Ryzen 7 5800X for $449 (not to speak about the other, higher-priced CPU tiers) compared to the 536 mm² Radeon RX 6800 XT for $649.
Finally, Rick Bergman had this to say regarding Threadripper, which has been absent from AMD's talks on Zen 3:
Rick Bergman"I can't talk about unannounced products, but we're committed to the Threadripper family. And so you could certainly expect that we'll in the future continue to have products in that particular area. Absolutely."
28 Comments on AMD Looks to Keep Performance, Efficiency Gains Momentum With Zen 4, RDNA 3, and Commitment to Threadripper
Its cool to see its not just money that paves the way.
No other words needed really.
AMD are really kicking intel, even though intel is already down on the ground. :D Even more shivering?
I really do not understand about this uncontrolled eagerness for new CPUs and GPUs and some people wanting them right now...
And my original comment that AMD need to ship CPU's and not talk about the next generation while this one undergoes what amounts to severe supply shortages.... was true regardless of your post lol
Once again, moving along...........
People see something new/shiny and they have to have the new/shiny item now...right NOW!
BITCH! YOU BETTER BACK THE F OFF AND GIVE ME MY GODDAMN BRAND SPANKING F'ING SHINY PIECE OF HARDWARE OR SO HELP ME! ARGH!
Seriously though, I want my brand spanking new hardware, but I'm not in any rush to be the first to have it. Once AMD's 6xxx series launch I'm game for getting my new build going, but I'm a realist and understand the bots snagging things up or folks gobbling things up and trying to resell....or that initial inventory at the launch date might be limited....I'm okay waiting a few months. My life doesn't revolve around having the newest hardware to survive.
Im really excited to see zen 3 threadripper, but I really hope they still offer a "cheap" option for those of us willing to sacrifice cores to keep the lanes. The 2920X I have launched at $650 in 2018 for 12 core/24 threads. Original threadripper had the same, with the 1900x a year earlier with only 8 cores/16 thread but still the full 60x pcie lanes at $550. But we never got that in the threadripper 3900 series with zen2. Cheapest chip on there was the 24 core/48 thread 3960x at like $1400.
I would upgrade to the new zen 3 threadripper if they offer the right chip. Something like a Ryzen 9 5900x or 5950x with 60 pcie lanes would do it. The IPC gains with zen3 are huge compared to zen+ so I'd be happy staying with a "low core count" threadripper (which id consider sub 24 core). But if the entry level for the platform are all $1500+ chips suddenly epyc single socket (P) models are options instead.
What I really want is threadripper to be above AM4 ryzen 9, but top out at something like 24 or 32 cores. Then have threadripper pro chips all be above that and blur the lines with epyc like they do now.
Their offerings right now just seem out of phase and I hope they fix it.
I fail my sarcasm 2nd time within an hour... Do I need more practice?
Why do you care so much if people/users get their "promised" hardware parts really?
Things are moving slowly on entire 2020... get used to it, look around you whats happening and... "move along..."
Yeah its not good to announce and release a product with low availability and have scalpers on top of this in the middle of countries, corporations and people being in lock down...
But we need our promised hardware now! Oh wait... others not me. I fight for people's right to new stuff also.
screenshot taken today at "apple.com"
so just because they entered a 5nm node the previous phones are not upgraded to 5nm :)
Once you hit the power limit of the cooling system you can't go any faster. It doesn't matter what max voltage/current/frequency a single core can do if you can't keep the whole chip cool enough.
N12e => n12e.tsmc.com/english/dedicatedFoundry/technology/N12e.htm
Didn't come out till this year, 2020.
FinFETs are always high performance first, and low power last. Hence, the high current and high mobility support and then trying to control it at later node revisions.
TSMC N5 is HP, they have yet to announce the -e suffix version of it so, thus LP is no where to be seen.