One of the more exciting segments of AMD’s Computex 2022 Keynote was the portion where AMD Chair and CEO Dr. Lisa Su brought out a demo of a mysterious Ryzen 7000 Series processor running Ghostwire: Tokyo at a very impressive clock of 5.5 GHz. Robert Hallock, Director of Technical Marketing at AMD, and Frank Azor, Chief Architect of Gaming Solutions, have now provided further insight regarding that presentation in a new interview with PC World’s The Full Nerd gang, revealing the interesting detail that the CPU was actually running at stock settings and without any sort of overclock. The obvious implication here is that owners of what is presumably AMD’s flagship Zen 4-based desktop processor will be able to experience amazing speed and performance right out of the box without serious tweaking. A majority of the CPU threads were reportedly able to maintain 5.5 GHz while running Tango Gameworks’ supernatural first-person adventure.
AMD's Robert Hallock on the 5.5 GHz Demo in @PCWorld stream:
— 𝐷𝑟. 𝐼𝑎𝑛 𝐶𝑢𝑡𝑟𝑒𝑠𝑠 (@IanCutress) May 24, 2022
▶ AMD Reference Mobo
▶ 280mm AIO cooler
▶ 16 core prototype from April
▶ Plugged in, no OC
Natural freq of that CPU
▶ Most threads around 5.5, depends on scene/game
▶ 5.2-5.5 on all threads was common on the game
It is not an overclock part. It’s just that’s the natural frequency of that particular prototype, so in the game we were running most of the threads around 5.5. That depends on the game load, depends on the scene. Of course clocks need to fluctuate up and down, so somewhere between 5.2 and 5.5 was pretty common on all the threads playing that game.
Source: PC World
The stuff I've seen more or less confirmed:
The CCDs jump to 5nm and 6nm on the I/O die.... and the move to DDR5 and PCI 5. A bump in L2 cache to 1MB. The AM5 socket allows for a higher TDP of 170W (up from 145W) - but that doesn't necessarily mean the new SKUs will all have a higher TDP. Higher boost clocks - up to 5.5Ghz. And an IGP on the I/O die!
There's a rumor that there could be low power CCDs and high performance CCDs - similar to little.BIG that Intel and Apple have moved to - but I can't find anything concrete about that.
There's rumor that V-cache from the 5800X3D will be present, but again, I can't find anything concrete.
Either AMD is sandbagging in terms of what the socket is capable of - and just look at those 22 stages of 90A power on the premiere motherboards! - or these are 'marketing' numbers. First crop of AM5 boards are more overbuilt than the top-end 12th-gen boards were when it comes to power delivery.
Alder Lake can do this on ambient cooling, assuming that it's very very good ambient cooling and one has a very very good sample (so the best of the 12900KS CPUs). What will be impressive is Zen 4 hitting that speed more often and doing so at less frightening voltage and subsequently power draw levels.
This has been a criticism of mine concerning Ryzen for a long time. AMD has leading graphics IP, and they just left that on the table. The APUs were also seriously neglected.
I'm not expecting competitive gaming performance here; I'd actually not even want that - but getting modern graphics IP in there is huge. AMD really needs a competitor to Quicksync, not just the hardware but also industry support, and this will help them get the marketshare to drive adoption.
I'm getting the "yes, but not yet" impression. It seems that, on paper, Zen 4 cores won't be as stout as the P-cores found in Alder Lake. Or maybe only just. Right now it does make sense for them to push their peak performance, as that's where they can really win mindshare; they fall behind there, and their marketing challenge grows.
And from what we've seen in the mobile space, AMD looks to be very close in terms of efficiency, and seemingly ahead when power limits drop to ultrabook levels.
I'm betting they're split on this. The v-cache limits thermal headroom which then puts a serious cap on raw compute throughput - so it's only useful when the CPU is cache limited. And there's only so many consumer applications where that's the case.
In other situations, the v-cache makes the CPU both more expensive to produce and slower, and from the enthusiast perspective, harder to cool for peak performance.
I can only point out that you and I are not representative of the majority of AMDs customers. And that while I prefer not to rely on iGPUs, I do find them useful in a pinch.
Basically this.
I'll reserve judgment until it is actually on the market and we can put it through it's paces.
I still remember the Pentium IV's inflated clocks.
;)
I'd never pay extra for the feature, but I'm usually glad it's there at some point or another in the lifespan of the build.
Intel IGP was also nice for QuickSync when I could get it working, but I would have issues with drivers and couldn't ever get it to work reliably for me.
Does anyone do this at all for the general market? Unless I'm missed something mobos with video are something from the past past, like pre 2000. As complicated as cpus have gotten, with the IO and plenty of stuff that also used to be on the mobo, I don't see how mobos with video make sense at this point. Unless you are making basically a powerful GPU into a mobo, basically a barebones console.