AMD to unveil Zen 4 CPUs at CES 2022

[Twinz]

I hope AMD were just sandbagging...hard, with this announcement.

-otherwise the performance (inside customer's PC's) looks to be underwhelming.

 

[Dave2150]

The chipset is irrelevant to a consumer. Do you mean mass availability from release? If so there is literally no info on that and you've just picked random speculation.

In terms of single core performance. Indeed it seems lower than expected but the multicore is solid improvement still. Was expecting more but as a total platform it seems reasonably solid if you upgrading from a system 3yr + old.

Highly doubt anything "solid" about jumping on brand new AM5 on launch. AM4/First gen Ryzen was plagued with issues. AM5 with it's double chipset design and expensive DDR5 only design will likely be an unwise decision for many, at least at launch.

 

[Freddie1980]

What do we think the itx boards will be like, do you think they will fit two chipsets and also be the high-end x670e, I'm looking forward to an Asus itx impact board.

Going by the presentation I think you will be fine, x670 will only have 1 chipset so fine for ITX. The 2 chipset designs look like there exclusive on the X670 extreme range which supplies extra PCI-E lanes to the 2nd PCI-E slot.

 

[Freddie1980]

So I just watched HUB's summary of AMD's presentation. I'm not disappointed but I'm not blown away either. It seems a lot of the work AMD have been focusing on is developing AM5, redesigning the IOC. Their bringing DDR5, PCI-E 5.0 and a raft of new features and standards to the platform and it's clear that IPC improvements haven't been a priority (anyone who has bought a 5800X3D is probably feeling pretty good about their purchase right now). I guess it's success will depend on if AMD is willing to pledge support AM5 for multiple generations and how keenly it's priced in face of stiff competition from Intel.

 

[HACO]

This is likely just Zen 3 ported to TSMC's N5, resulting in gains in clock speed and efficiency, with DDR5 and updated IO.

Seems like the actual Zen 4 microarchitecture, the one which we were expecting a 15-20% IPC uplift, wasn't ready for release this year, so AMD is going to plan B. It remains to be seen whether that microarchitecture will be released next year or if AMD is skipping it entirely and prioritises getting back on track for the actual Zen 5 (i.e. the successor to the actual Zen 4, not this) for a combined 35-40% IPC uplift.

 

[ltron]

Why? if you look at retailers that publish sales comparisons its obvious that AMD are still outselling Intel in retail CPU's, and by significant margins.

To release a CPU that is better than the one currently outselling Intel in retail is not going to do that any harm is it?

I suspect they are sandbagging given what they have shown before. Intel has Raptorlake coming and AMD could lose that advantage if Intel starts winning consistently again. I remember the Bulldozer years (not that this is in any way close to that) and that relentless stagnation along with very high prices is what we get when Intel dominates. I don't want to go back to that or even begin to do so, at least until AMD have built up a huge war chest of cash so they can withstand it and compete against Intel on an equal footing (I am sure Intel will attempt to crush them next time, there is no more taking AMD lightly).

The trouble with past AMD successes is that they were short-lived (over years rather than decades), so they never got the strong foothold in the market that they needed (along with market abuse and dirty tricks from Intel for which they were never properly punished). However, I am confident with the brilliant Dr Su in charge and the great team around her that this recent success is very unlikely to be just another flash in the pan.

 

[Troezar]

This is likely just Zen 3 ported to TSMC's N5, resulting in gains in clock speed and efficiency, with DDR5 and updated IO.

Seems like the actual Zen 4 microarchitecture, the one which we were expecting a 15-20% IPC uplift, wasn't ready for release this year, so AMD is going to plan B. It remains to be seen whether that microarchitecture will be released next year or if AMD is skipping it entirely and prioritises getting back on track for the actual Zen 5 (i.e. the successor to the actual Zen 4, not this) for a combined 35-40% IPC uplift.

So really Zen3+ whilst getting the platform ready for Zen5?

 

[Robert896r1]

The chipset is irrelevant to a consumer. Do you mean mass availability from release? If so there is literally no info on that and you've just picked random speculation.

In terms of single core performance. Indeed it seems lower than expected but the multicore is solid improvement still. Was expecting more but as a total platform it seems reasonably solid if you upgrading from a system 3yr + old.

There’s 2 chipsets on the boards. That’s added cost that’s very real to the consumer.

 

[HACO]

So really Zen3+ whilst getting the platform ready for Zen5?

Lots of publications have likened it to a Zen 3+ so I guess that's fair to say.

 

[Orcvader]

Pure speculation on my end but might be connected how a future revision of Zen 4 will be used as efficiency cores (Zen 4D "densed" in some rumours). So maybe they wanted to keep the design simple enough for them to optimise for power usage later on while Zen 5 becomes the true performance core.

But yeah, clock speeds at least looks good, but IPC gains seems to be small and I suspect most of that comes from the higher IF speeds thanks to DDR5 RAM, assuming they're still linked.

 

[humbug]

I hate speculation about things we have absolutely nothing to go on.

Gaming performance is an unknown quantity, it could range anything a little better than Zen 3 and much better than Zen 3D, we have no idea what the impact of the larger L2 and DDR5 has, all we know for certain is it clocks 10 to 15% higher in games.

No one thought the 3D Cache would have an impact to the extent it has, MLID was wrong about that, just as he was wrong about Intel GPU's and he was wrong about Zen 4.

The Cinebench / productivity performance is what it is, they are not sandbagging, no point in looking for that, just as there is no point in getting depressed about what this all means for gaming performance, Zen 2 vs Zen 3 offered +20% in productivity but up to +50% in games, that's not to say Zen 3 vs Zen 4 will be like that but its just another example about how trying to make these predictions is daft.

 

[Troezar]

I hate speculation about things we have absolutely nothing to go on.

Gaming performance is an unknown quantity, it could range anything a little better than Zen 3 and much better than Zen 3D, we have no idea what the impact of the larger L2 and DDR5 has, all we know for certain is it clocks 10 to 15% higher in games.

No one thought the 3D Cache would have an impact to the extent it has, MLID was wrong about that, just as he was wrong about Intel GPU's and he was wrong about Zen 4.

The Cinebench / productivity performance is what it is, they are not sandbagging, no point in looking for that, just as there is no point in getting depressed about what this all means for gaming performance, Zen 2 vs Zen 3 offered +20% in productivity but up to +50% in games, that's not to say Zen 3 vs Zen 4 will be like that but its just another example about how trying to make these predictions is daft.

True but it is fun There's also not a lot going on whilst we're waiting for new CPU's and GPU's. That said, I feel quite happy with my 5800X3D so not all bad!

 

[humbug]

Speculation is fine, i guess what i'm saying is don't treat these equations like they are set in stone, as i said before, the Intel days of making these predictions are gone, we are in a new age of innovation.

 

[Orcvader]

True, it's still only a teaser, we haven't got more details on what the actual architectural changes are. V-cache has shown extra cache doesn't benefit Cinebench but the doubled L2 cache should be a good boost in gaming. We just need to wait until AMD reveals more.

 

[Dave2150]

I hate speculation about things we have absolutely nothing to go on.

Gaming performance is an unknown quantity, it could range anything a little better than Zen 3 and much better than Zen 3D, we have no idea what the impact of the larger L2 and DDR5 has, all we know for certain is it clocks 10 to 15% higher in games.

No one thought the 3D Cache would have an impact to the extent it has, MLID was wrong about that, just as he was wrong about Intel GPU's and he was wrong about Zen 4.

The Cinebench / productivity performance is what it is, they are not sandbagging, no point in looking for that, just as there is no point in getting depressed about what this all means for gaming performance, Zen 2 vs Zen 3 offered +20% in productivity but up to +50% in games, that's not to say Zen 3 vs Zen 4 will be like that but its just another example about how trying to make these predictions is daft.

15% ST performance increase is with high end, expensive DDR5, 6000Mhz C30. As per 99% of the official 12900k reviews, on launch day I imagine they'll use 4800Mhz/5200Mhz 'standard' DDR5, so could see less performance than this in reviews.

 

[humbug]

15% ST performance increase is with high end, expensive DDR5, 6000Mhz C30. As per 99% of the official 12900k reviews, on launch day I imagine they'll use 4800Mhz/5200Mhz 'standard' DDR5, so could see less performance than this in reviews.

Sure, AMD aren't topping the Cinebench threads with these CPU's, or at least not the ST part of them, Intel are still having to find near 3X their current performance to crack AMD's now 2 generation old CPU in MT.

AMD may lose ST by 10%, but Intel are still losing by 300% in MT.

 

[ltron]

Linus says the 12900K beats the 5950X by a very small margin in short Blender tests like the one shown in AMD's demo (even though the 5950X wins in longer tests):

That 31% Zen 4 performance increase over the 12900K is hugely impressive then.

 

[smilingcrow]

I seem to recall Mike Clark at AMD saying how excited he was for Zen 5, which did make me wonder if Zen 4 would be underwhelming.
It seemed an odd thing for the 'Chief Architect of Zen' to say about a year before Zen 4 would even hit the market.
Is Zen 5 due next year?
With the world as it is, maybe 2024 is more realistic!

 

[ltron]

I seem to recall Mike Clark at AMD saying how excited he was for Zen 5, which did make me wonder if Zen 4 would be underwhelming.
It seemed an odd thing for the 'Chief Architect of Zen' to say about a year before Zen 4 would even hit the market.
Is Zen 5 due next year?
With the world as it is, maybe 2024 is more realistic!

Zen 4 VCache is next year according to rumours, but AMD could surprise us and release Zen 5 that year too (although I agree it will probably be in 2024).

 

[humbug]

Linus says the 12900K beats the 5950X by a very small margin in short Blender tests like the one shown in AMD's demo (even though the 5950X wins in longer tests):

That 31% performance increase over the 12900K is hugely impressive then.

Yes because the 12900K can't sustain the higher boost clocks for long, this is like a positive way of saying it draws too much power and gets too hot once the cooler is heat soaked, its Ryan Shrout level framing, Linus.