AMD - Zen chitchat

[kompukare]

Nice. Now do I wait for them to support my Vega 56 (which currently boot loops back into the BIOS if I enable this) or is it an excuse to get a 6000 series gpu?

Well, aside from the dodgy attempt from AMD marketing to make this sound like Zen3-only (which seems to have backfired at them), the other main reason they didn't enable this universally must be validation. Might need a big(ish) change to the drivers which they would need to validate if enabled on older GPUs.
6000 series are still far to scarce to consider unless you want to pay scalpers. The eventual 6700 and 6600 parts should be a bit better value when they show up.

[kompukare]

Does this work on the older Nvidia cards?

Simple answer: no.
Complex answer: probably not. The drivers have to be changed, and we don't know how complex any changes are, then if eventually Nvidia validate these changes, it's likely they'd only do so for current cards.
Like with position with old AMD GPUs, it is always possible that someone will find out where the driver checks for support and that might be hackable (not sure how Windows driver signature and hacking play out).
That would be totally unsupported, of course.

[Essexboy]

Yeah that is what i thought we know how Nvidia like to force us to get new gear

[DanceswithUnix]

, the other main reason they didn't enable this universally must be validation.

That has always been my expectation on this. Development happens on new product, and gets eventually back ported to old stuff if it isn't too difficult.

In the case of Vega, well the APUs which currently use Vega cores store graphics data in CPU ram because that is all they have, so they have *always* had full access to the GPU memory. OFC that might muddy the waters rather than make discrete GPU driver changes easier.

But the fact that my motherboard doesn't get anywhere near booting Windows if the SAM feature is enabled is a bit of a worry. Perhaps the BIOS isn't right, perhaps my early 3700X is buggy, perhaps the Vega 56 doesn't properly support it. *shrug*, who knows.

[k9wazere]

Anyone starting to wonder if AMD will *not* release a £200 6-core part this time around?

With the 5800 and 5900 being OEM only, and seemingly no plans for a 5700X, it seems AMD might be playing hard-ball. "Pony up or go without."

Great for shareholders, sucks for budget-minded consumers.

[DanceswithUnix]

Anyone starting to wonder if AMD will *not* release a £200 6-core part this time around?

With the 5800 and 5900 being OEM only, and seemingly no plans for a 5700X, it seems AMD might be playing hard-ball. "Pony up or go without."

Great for shareholders, sucks for budget-minded consumers.

We are still missing all the non OEM APU range to replace the 3400G, which isn't compatible with B550 motherboards so that's a massive hole in the lineup. I wonder if now APUs go up to 8 cores, a budget 6 core CPU might be seen as too much overlap. I would be fine with that, if they actually allowed me to buy a 4000 series pro APU (and I would have bought at least one for work already).

[Terbinator]

Anyone starting to wonder if AMD will *not* release a £200 6-core part this time around?

With the 5800 and 5900 being OEM only, and seemingly no plans for a 5700X, it seems AMD might be playing hard-ball. "Pony up or go without."

Great for shareholders, sucks for budget-minded consumers.

I suspect we will see them once Intel's new stuff drops, or, a price shift to essentially fall in to that product bracket.

[kompukare]

First Cezanne review.
Apparently sometimes AT can get reviews out before anyone else:
https://www.anandtech.com/show/16446...-mobile-tested
Pretty good result.
That Asus laptop has a dGPU but Ian ran iGPU benchmarks; might have to revise my comment on the Intel discrete Xe GPUs thread as Iris Xe is still a lot further behind Vega 8 than I would have though.

[kompukare]

Well, this is where AMD's TSMC wafers have gone:
Sony sold 4.5 million PS5 already since November. At a loss of course:
https://www.computerbase.de/2021-02/...ellungskosten/
Yields are hardly 100% that means they've used a lot more wafer than that.
And Sony were meant to have had yield problems (probably because they decided to run the PS5 faster than their initial spec). According to some reports really bad yield problems.
But let's go with the oft-quoted 0.09 defects per mm (from some AT article about 5nm and 7nm yields) and put those into a yield calculator tool:

The 3.3 million Xbox Series X is an estimate from Statista
https://www.statista.com/statistics/...s-x-worldwide/
Couldn't find any info on Xbox S series.
Anyway, that's roughly 70,000 wafers. Possibly a lot worse if Sony had to scrap many for not being able to hit the clock speed they wanted (wouldn't mind some of those for NUC build!).

[Corky34]

Thought I'd post this just because it made me go OMG..

AMD Genoa with Zen4 core architecture would feature up to 96 cores, 192 threads, and 12-channel DDR5 memory. The leaker confirmed that Genoa would utilize DDR5-5200 memory and PCI Gen5.

Plenty of salt required but if true that's a stonking amount of cores and silly fast memory.

[DanceswithUnix]

Thought I'd post this just because it made me go OMG..

Plenty of salt required but if true that's a stonking amount of cores and silly fast memory.

I guess 5nm will have a lower tdp, and in a server that means more cores.

12 channel ram could make for an interesting Threadripper platform!

[CAT-THE-FIFTH]

Interesting:
https://videocardz.com/newz/amd-prep...-up-to-5-0-ghz

Current Zen3 is a B0 stepping and now its going to be B2 stepping? I wonder if that means Zen3 has had some significant errata?

[[GSV]Trig]

Well there's bound to be some sort of changelog somewhere surely?

[DanceswithUnix]

Current Zen3 is a B0 stepping and now its going to be B2 stepping? I wonder if that means Zen3 has had some significant errata?

All chips have significant errata.
Or it might be they ran new masks for the supposedly design rule compatible 6nm process, so this might be Warhol:

https://www.extremetech.com/computin...oming-in-zen-4

[CAT-THE-FIFTH]

If it was a half node shrink,then the stepping would reset for that node?? There was noise the Zen2 XT series was on a slightly improved TSMC 7NM,which was later used for Zen3.

But going from B0 to B2,means one whole stepping has been ditched. I have been hearing of certain Zen3 SKUs having stability issues at stock,especially the Ryzen 7 5800X. The Ryzen 5 5600 B2 has the same clockspeeds too as the B0 stepping. It wouldn't surprise me one bit due to the pandemic AMD pushed out what it had,especially as they had very little spare 7NM volume due to the consoles. So as production volume frees up,they are probably making fixes - I have noticed Zen3 seems somewhat easier to get worldwide now!

[kompukare]

Well, AMD released a statement about the B2 revisions:

As part of our continued effort to expand our manufacturing and logistics capabilities, AMD will gradually move AMD Ryzen 5000 Series Desktop Processors to B2 Revision over the next 6 months. The revision does not bring improvements in terms of functionality or performance, furthermore, no BIOS update will be required.

As reported by
https://www.computerbase.de/2021-05/...-bis-zu-5-ghz/
who based this on a story by
https://www.benchmark.pl/aktualnosci...ewizji-b2.html
So maybe there will be some more overclocking headroom but no new launch, binning or price increases.