Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[Abwx]

Only 17% slower than a 7950X in MT!

View attachment 80279

14.5% slower, wich mean that the 7950X is 17% faster, to get the slowness you must put the latter as the 100% reference...

 

[DrMrLordX]

Only 17% slower than a 7950X in MT!

Also the 7950X has the advantage of faster/better RAM in that comparison.

 

[A///]

never a fan of that testing method using slow ram. I look forward to the gang land shooting amd will give intel next year with zen 5.

 

[eek2121]

The chip had power limits set to 90/125W according to the chart, so I wouldn’t expect a large difference.

I have not read the article, but I am curious as to how well it performs at 35-45W

 

[igor_kavinski]

I have not read the article, but I am curious as to how well it performs at 35-45W

Still performs admirably compared to Intel's power guzzlers.


With top 7950X as the reference.

Almost three times the power efficiency below:


Anyone disagree that this is nothing short of astounding?

AMD: The solution to Global Warming

Intel: We are going to Mars! (Let's trash this awful place)

 

[eek2121]

Still performs admirably compared to Intel's power guzzlers.

View attachment 80305
With top 7950X as the reference.

Almost three times the power efficiency below:

View attachment 80306
Anyone disagree that this is nothing short of astounding?

AMD: The solution to Global Warming

Intel: We are going to Mars! (Let's trash this awful place)

Oh I wasn’t disagreeing, Zen 4 can be extremely efficient. I was just commenting on the chart.

 

[DAPUNISHER]

Looking pretty beastly on Jarrod's chart

 

[igor_kavinski]

Intel needs another E-core cluster for a total of 36 threads to maybe beat that.

 

[RnR_au]

Looking pretty beastly on Jarrod's chart

Should be a very interesting video for those that want a beastly portable workstation;

https://twitter.com/x/status/1655699504920936449

 

[Geddagod]

Should be a very interesting video for those that want a beastly portable workstation;

https://twitter.com/x/status/1655699504920936449

I love how a lot more reviewers are adding in CB power scaling.
It's a shame though how with Intel adding little cores to their CPUs, comparing core-to-core comparisons for Intel vs AMD is way more rare. I would love to see how RPC alone compares in CB23 power scaling vs Zen 3 and Zen 4, but reviewers like HWUB don't delve that deep and scale by the entire chip instead.
I wonder if anyone can also compare the 7945HX power scaling vs the 7950x. The 7945HX would just be binned for better leakage vs worse scaling right? Since leakier chips are generally able to scale to higher clocks? Plus IIRC AMD heavily modifies the V/F curve of their mobile counterparts, so that would be interesting to see.

 

[eek2121]

I love how a lot more reviewers are adding in CB power scaling.
It's a shame though how with Intel adding little cores to their CPUs, comparing core-to-core comparisons for Intel vs AMD is way more rare. I would love to see how RPC alone compares in CB23 power scaling vs Zen 3 and Zen 4, but reviewers like HWUB don't delve that deep and scale by the entire chip instead.
I wonder if anyone can also compare the 7945HX power scaling vs the 7950x. The 7945HX would just be binned for better leakage vs worse scaling right? Since leakier chips are generally able to scale to higher clocks? Plus IIRC AMD heavily modifies the V/F curve of their mobile counterparts, so that would be interesting to see.

I don’t have insight into how AMD bins, but I suspect the process is a lot more complicated.

For example, a lot of folks assume EPYC gets the “best” bins. I suspect that is not the case. Something like this seems to make more sense to me:

Mobile: Efficiency - Optimized for low power and medium clocks
Desktop: Performance - Optimized for high clocks
EPYC: Balanced for clocks and power consumption
Threadripper - Same as Ryzen with power consumption a secondary weight

mobile gets the chiplets that can do 4.8-5.4ghz ghz within 45W tdp

Ryzen gets the chiplets that can do 5-6 ghz within 65-170w TDP.

Threadripper would get those that have a good combination of both (5.4ghz, PPT / chiplet count?) Possibly mobile chiplets needed for this?

EPYC would get the chiplets that have great (PPT / chiplet count?) power consumption at a 3- 4ghz.

So in the above scenario, nearly all chiplets qualify for epyc,most for desktop, fewer for mobile, and far fewer for Threadripper.

If this speculation is true it would also explain their release cycle.

Threadripper would require the best bins since it needs to hit 5+ ghz speeds at power levels that laptops would use, so technically it would get the best chiplets.

 

[Abwx]

Some leak for next gen TR release date :

Roadmap-Update: Noctua kündigt indirekt neue AMD Ryzen Threadripper an

Noctua hat wieder einmal seine Roadmap aktualisiert. Erneut gibt es Terminverschiebungen, aber auch neue Produkte und eine Streichung.

www.computerbase.de

 

[Geddagod]

Some leak for next gen TR release date :

Roadmap-Update: Noctua kündigt indirekt neue AMD Ryzen Threadripper an

Noctua hat wieder einmal seine Roadmap aktualisiert. Erneut gibt es Terminverschiebungen, aber auch neue Produkte und eine Streichung.

www.computerbase.de

Apparently Noctua roadmaps are notoriously bad.

 

[Hail The Brain Slug]

Apparently Noctua roadmaps are notoriously bad.

They perpetually delay things, I don't remember them ever meeting their first, or second, or even third timeline on their roadmap for a product.

 

[Abwx]

Apparently Noctua roadmaps are notoriously bad.

And.?..

So they wont be ready for TR launch, that s all you are saying, they dont decide of the SKU launch, isnt it, unless you think that AMD will delay it because Noctua wont be ready...

 

[A///]

not at all. gedda was commenting on their release schedule. we already knew tr was a 2h product but likely 3rd q a few weeks ago. i interpreted his post one of his more sane ones that you shouldn't expect noctua to have anything ready for tr when it comes out.

whether tr is good or not and isn't a fumble b amd is to be determined.

 

[Markfw]

Some leak for next gen TR release date :

Roadmap-Update: Noctua kündigt indirekt neue AMD Ryzen Threadripper an

Noctua hat wieder einmal seine Roadmap aktualisiert. Erneut gibt es Terminverschiebungen, aber auch neue Produkte und eine Streichung.

www.computerbase.de

I hope next gen TR coolers are SP5 socket. Then I can get some for my Genoa's !

 

[Geddagod]

And.?..

So they wont be ready for TR launch, that s all you are saying, they dont decide of the SKU launch, isnt it, unless you think that AMD will delay it because Noctua wont be ready...

Obviously.

 

[Geddagod]

not at all. gedda was commenting on their release schedule. we already knew tr was a 2h product but likely 3rd q a few weeks ago. i interpreted his post one of his more sane ones that you shouldn't expect noctua to have anything ready for tr when it comes out.

whether tr is good or not and isn't a fumble b amd is to be determined.

more sane?

 

[Abwx]

not at all. gedda was commenting on their release schedule. we already knew tr was a 2h product but likely 3rd q a few weeks ago. i interpreted his post one of his more sane ones that you shouldn't expect noctua to have anything ready for tr when it comes out.

whether tr is good or not and isn't a fumble b amd is to be determined.

That s up to Noctua to be ready on time, that s not difficult to use a piece of aluminium the same area and thickness as the SKU casing and screw 4, or more, 200W discrete transistors as heat source on a side and the tested cooler on the other one, set the current/voltage such that they dissipate a total of say 300W and then measure the cooling efficency with thermal probes...

I hope next gen TR coolers are SP5 socket. Then I can get some for my Genoa's !

It could work but 25% more power isnt trivial at these levels, it s not like we re going from 100W to 125W.

 

[Schmide]

Apparently Noctua roadmaps are notoriously bad.

So you're not a fan ? (i had to)

 

[Markfw]

That s up to Noctua to be ready on time, that s not difficult to use a piece of aluminium the same area and thickness as the SKU casing and screw 4, or more, 200W discrete transistors as heat source on a side and the tested cooler on the other one, set the current/voltage such that they dissipate a total of say 300W and then measure the cooling efficency with thermal probes...




It could work but 25% more power isnt trivial at these levels, it s not like we re going from 100W to 125W.

You know the new heatsinks will do 300-400 watt. My Genoa's are 400 watt, so should be no problem. You should see the crap I have to run now.Jet engines...

 

[Abwx]

You know the new heatsinks will do 300-400 watt. My Genoads are 400 watt, so should be no problem. You should see the crap I have to run now.Jet engines...

Akin to drive a Ferrari with badly worned tires...

 

[RnR_au]

Video is out.

 

[BorisTheBlade82]

Fritzchens Fritz posted some IR comparison images of a 7800X3D vs. a 7600X on Twitter.
Quite interestingly the cache die seems to overlap with the cores a bit. That wasn't the case on the 5800X. That should be the L2 area.

https://twitter.com/x/status/1659686660429459456