Hallock answers questions about AVX, AM5, RDNA-2-IGP and more

The Ryzen 7000 announcement was promising, but also accompanied by some question marks. The maximum possible TDP of the new processors, for example, caused confusion. Now we know that it can be up to 170 watts TDP and up to 230 watts PPT. Meanwhile, AMD is doing further educational work via Robert Hallock (Director of Technical Marketing) with open questions. Already quite certain as a rumor, it is now confirmed that Ryzen 7000 is offered with up to 16 cores. As a result, both CCDs will be used in their full configuration in the top model.

And the two CCDs are now gold-plated thanks to a process called back-side-metallization. The coating is only a few hundred angstroms and is intended to both prevent oxidation and better hold the solder, which is used to dissipate heat to the heatspreader with now also less thermal resistance.

The dies were also measured immediately: the IO die from TSMC in 6 nm should therefore have 120 square millimeters; the two compute dies from TSMC in 5 nm each 72.5 square millimeters. The calculation is based on the fact that the AM5 or socket LGA1718 package measures 40 × 40 millimeters.

AI acceleration in Zen 4

One of the reasons why the TDP and PPT increase is reportedly due to the use of AVX-512 for AI acceleration. We already know from Intel that AVX loads quickly overwhelm and that needs to be covered. AMD’s Robert Hallock also confirms that the AI ​​acceleration uses VNNI and BLOAT16 instructions.

“More specifically, AVX 512 VNNI for neural networks and AVX 512 BLOAT16 for conclusions. Both are pretty good speed increases, we’re not using fixed function acceleration, that might be something we could do with our Xilinx acquisition. We’re beginning to see more uses for AI -Consumer workloads, such as video upscaling, which has grown exponentially over the last two years. I think there’s a general trend of the average enthusiast taking on more AI-style workloads.”

RDNA2 for all Ryzen 7000

Not much was initially revealed to the IGPs. Here, too, you go into more detail: All Ryzen 7000 get an RDNA-2 unit. AMD will probably not offer any “F models” either, i.e. Ryzen 7000 without IGP. It always sits in the IO die, which is manufactured in 6 nm at TSMC. In principle, it has everything you know from the Radeon RX 6000 – only in small format. Small probably means 4 compute units and the entire encode/decode technology including AV1.

“IGP is standard. It’s included in all 6-nanometer IO chips, which house a small number of processing units dedicated to video encoding and decoding and multiple display outputs. […] For enthusiasts, it is helpful to diagnose a bad graphics card to get the system up and running while still waiting for the GPU. The iGPU configuration [Spezifikationen] are uniform and all CPUs have them.”

AM5 and the longevity

They didn’t want to say how long AM5 will be used by AMD. Hallock says in an interview that he doesn’t know. AMD is switching to Land Grid Array and theoretically you could use AM5 for a long time, since DDR5 and PCI Express 5.0 are already taken into account and 170 watt TDP should leave possibilities. In practice, however, one does not know or does not want to say it.

“I don’t know yet – that’s the honest answer. We’re still in the early stages of developing AM5. It’s coming out in the fall, but that’s still a long way to go. One of the things we want to clarify is what that will look like. Our users expect transparency on this topic. But we just don’t have an answer yet.”

HEDT is also retained. The enthusiast platform was thought dead at AMD like Intel, but it’s coming back – possibly more with a workstation focus. So new Threadrippers are assured as they are “not going anywhere” – maybe on the rumored SP6 socket.

Sources: tech powerup, Crazy Tech Lab, hothardware, PC WorldTwitter (skyjuice 1, Skyjuice 2)