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!

 

[leoneazzurro]

I think you can remove the “allegedly”.

Yep, they tested the Zephirus Duo 16. Also, it seems the 4090 on this Asus is limited to 150W instead of 175W like in the other laptops based on the same GPU in their comparison. Quite odd IMHO from Asus.

 

[nicalandia]

Allegedly first results for Dragon Ridge

AMD strikes back: Ryzen 9 7945HX beats Intel Core i9-13980HX despite much lower power consumption - NotebookCheck.net News

Did AMD just win the mobile CPU Game? - AMD Ryzen 9 7945HX + Zephyrus Duo 16 Hands-On - YouTube

Edit: added the video from the same source

Wow, impressive performance.

As expected AMD will dominate in Mobile and Server since those are TDP restricted and Intel just can't go crazy like in Desktop

 

[Panino Manino]

Look IGP benchmarks: https://www.pcmag.com/reviews/amd-ryzen-9-7950x3d

Just 2CU!
Imagine what we could have if AMD launched a proper APU with 3DCache already.

 

[gdansk]

Look IGP benchmarks: https://www.pcmag.com/reviews/amd-ryzen-9-7950x3d

Just 2CU!
Imagine what we could have if AMD launched a proper APU with 3DCache already.

Has anyone else tested this?
2 RDNA2 CU should be about 2/3rd of 32 EU Xe. So I am a bit confused why their performance was so awful on the non-X3D parts.

 

[leoneazzurro]

There is something quite fishy with those results as the iGPU cannot access the additional cache directly, so there is literally no way they could achieve those results, except if the test they made is completely CPU limited (which is also strange in the comparison with the vanilla 7000 series).

 

[LightningZ71]

Has anyone else tested this?
2 RDNA2 CU should be about 2/3rd of 32 EU Xe. So I am a bit confused why their performance was so awful on the non-X3D parts.

I'm actually more suspicious of the 7950X numbers over the 7950X3d numbers. I suspect that the 7950X numbers may have been recorded using release day drivers instead of the current drivers that they used on the X3d. I don't have any evidence of this, but, it wouldn't shock me. They are at least consistent across the four games tested.

Overall, I would have expected the 7950X3d to have some sort of performance lead over the vanilla 7950X with it's IGP if only for the fact that the 7950X3d should have lower memory contention because of the existence of the 3D cache, but not to this degree. I don't see thermals or total power draw for the IOD being an issue enough to make that much of a difference.

That performance difference is just screaming for more investigation...

 

[gdansk]

I'm actually more suspicious of the 7950X numbers over the 7950X3d numbers. I suspect that the 7950X numbers may have been recorded using release day drivers instead of the current drivers that they used on the X3d.

Sorry I wasn't clear. That's what I find suspicious. The non-X3D numbers are abnormally low. The X3D numbers look to me about normal for the iGPU.

 

[Det0x]

Allegedly first results for Dragon Ridge

AMD strikes back: Ryzen 9 7945HX beats Intel Core i9-13980HX despite much lower power consumption - NotebookCheck.net News

Did AMD just win the mobile CPU Game? - AMD Ryzen 9 7945HX + Zephyrus Duo 16 Hands-On - YouTube

Edit: added the video from the same source

Intel's flagship consumed anywhere from 180-190W and even spiked beyond 200W with average temperatures hovering around 85C. Within a few seconds of operation, the CPU was throttling to sub 3.8 GHz clocks. Meanwhile, AMD's chip never broke past 120W which was its peak and temps were close to 80C but the chip remained steady with a 4.3-4.4 GHz clock speed.

 

[A///]

So just 1 other setup?

No that would be for around 30 setups. Plus or minus a few degrees on AC/room temp. Much like a normal heatsink but to a lower degree an AIO is still susceptible to a hot environment. If your room is hot, your case is even more hot. You won't get the same cooling as you would in a cooler environ. If you run AC but your case is stagnant and hot, your cooler won't perform well. The swing can be small or very large.If you're intake fans are at 100% and your rear exhaust fans are at 100% but your AIO fans are near idle, your temps will suffer.Your case ambient may be cooler, but there isn't enough lingering air + enough speed for the aio fans to suck in the cool air and expel it out thus cooling the fluid within through the fine channeling of the radiatior. On very hot days here where it'll reach high triple digits I'll slow down the exhaust fans, increase the baseline fan speed of the 4x 140 intakes to 65% and tie the rest to the GPU and CPU. The AIO will have similar baselines. What I get from this is while minimal air is expelled from the rest 2 exhaust fans the ambient air temp in the case drops significantly due to the ambient room temp due to AC hovering around 60-65f. The air inside the case is barely warm despite the system operating at more than 85% for hours at a time.

As I told @Markfw my home has an ac system but I installed a separate split air AC for this room (Mitsubishi) if I'm spending all day in my home office I don't need to cool the entire house down. My central air isn't fancy enough to have zoned cooling.

 

[Kepler_L2]

Look IGP benchmarks: https://www.pcmag.com/reviews/amd-ryzen-9-7950x3d

Just 2CU!
Imagine what we could have if AMD launched a proper APU with 3DCache already.

These results are wrong, iGPU can't access CPUs L3.

 

[Hail The Brain Slug]

I am not sure why AMD does not offer 7800X3D NOW, when it consumes just one chiplet instead of two. It must be easier for them to make in larger numbers.

The milking strategy of forcing people to buy more expensive stuff even if they do not really need it is not nice at all, hope it is not their main motivation.

My take is they are probably only making dual CCD packages, at least at first. 7950X3D and 7900X3D launch first because the defect rate for the chiplets and the packaging process is not bad at all. They are probably stockpiling enough failed packages to make a reasonable launch day yield for the 7800X3D.

 

[Thibsie]

My take is they are probably only making dual CCD packages, at least at first. 7950X3D and 7900X3D launch first because the defect rate for the chiplets and the packaging process is not bad at all. They are probably stockpiling enough failed packages to make a reasonable launch day yield for the 7800X3D.

But at a time there maybe be some 7800X3D as there are dual CCD 5600X too (deactivated second CCDs though, so probably some salvaged x900/x950.

 

[Heartbreaker]

My take is they are probably only making dual CCD packages, at least at first. 7950X3D and 7900X3D launch first because the defect rate for the chiplets and the packaging process is not bad at all. They are probably stockpiling enough failed packages to make a reasonable launch day yield for the 7800X3D.

Well, until they start making 7800X3D, they obviously must be making dual CCD packages.

But I would expect as soon as they start making 7800X3d, they will be single CCD. I also expect they are already making 7800X3d Single CCD parts and stockpiling them for launch.

There is no mystery why they are delaying 7800X3d: Mo Money!

 

[igor_kavinski]

The other issue is lack of competition. Intel is unable to challenge AMD in this space. They probably determined that there is no cost effective way for them to add a huge cache to their designs. It's sad since they were the actual innovators who made the legendary i7-5775C that served their owners remarkably well for years in games while everyone around them upgraded their PCs to newer gens of Core CPUs.

Until Intel has a real answer to V-cache, AMD will charge premium prices for these CPUs. Maybe Zen 5 will standardize V-cache on the entire product stack and then Intel's hand will be forced.

 

[Kocicak]

... They are probably stockpiling enough failed packages to make a reasonable launch day yield for the 7800X3D.

They need to test the bare chiplets for functionality before they mount the additional cash on it.

They are not building the CPUs from chiplets with unknown functionality and it would be a huge waste for them to sell a CPU requiring one chiplet with two chiplets in them.

 

[Hail The Brain Slug]

They need to test the bare chiplets for functionality before they mount the additional cash on it.

They are not building the CPUs from chiplets with unknown functionality and it would be a huge waste for them to sell a CPU requiring one chiplets with two chiplets in them.

Yet it's been observed with delidded 7600X and 7700X samples that some are dual chiplet packages. This implies there can be some failure in packaging that leads to necessitating the disable of one CCD.

 

[Joe NYC]

Until Intel has a real answer to V-cache, AMD will charge premium prices for these CPUs. Maybe Zen 5 will standardize V-cache on the entire product stack and then Intel's hand will be forced.

I just noticed that the price of 5800x3d at Micro Center dropped to $299, and all the stores are fully stocked.

This would indicate that AMD is starting to deliberately push this CPU, that the capacity is there for it now, and also, the cost structure of adding this stacked die is very good.

 

[Heartbreaker]

I just noticed that the price of 5800x3d at Micro Center dropped to $299, and all the stores are fully stocked.

This would indicate that AMD is starting to deliberately push this CPU, that the capacity is there for it now, and also, the cost structure of adding this stacked die is very good.

Or that they are clearing them out.

 

[Abwx]

Computerbase tested the 7950X3D iGPU, basically perfs are the same than the cacheless SKU with eventually 10% better perf in Cyberpunk, that s about all.

7950X3D vs. 7950X: Berichte über explodierende iGPU-Leistung sind falsch

Die iGPU des Ryzen 9 7950X3D soll gleich viermal schneller als jene des Ryzen 9 7950X arbeiten? Diese Mär kann ComputerBase widerlegen.

www.computerbase.de

 

[Hail The Brain Slug]

Computerbase tested the 7950X3D iGPU, basically perfs are the same than the cacheless SKU with eventually 10% better perf in Cyberpunk, that s about all.

7950X3D vs. 7950X: Berichte über explodierende iGPU-Leistung sind falsch

Die iGPU des Ryzen 9 7950X3D soll gleich viermal schneller als jene des Ryzen 9 7950X arbeiten? Diese Mär kann ComputerBase widerlegen.

www.computerbase.de

I had read speculation the previous testing was flawed. Interesting to see the truth here.

Allegedly the iGPU cannot access the L3 directly, so the L3 change should have made minimal difference.

 

[Kaluan]

I had read speculation the previous testing was flawed. Interesting to see the truth here.

Allegedly the iGPU cannot access the L3 directly, so the L3 change should have made minimal difference.

Sounds like something Strix Point and Granite Ridge/Granite Ridge-X designs may need to work on.

 

[Exist50]

Sounds like something Strix Point and Granite Ridge/Granite Ridge-X designs may need to work on.

It's such a small GPU, probably not worth bothering. Maybe if they add more IPs to the IO die (e.g. AIE) it would be worth having a small system cache there.

 

[Heartbreaker]

Computerbase tested the 7950X3D iGPU, basically perfs are the same than the cacheless SKU with eventually 10% better perf in Cyberpunk, that s about all.

7950X3D vs. 7950X: Berichte über explodierende iGPU-Leistung sind falsch

Die iGPU des Ryzen 9 7950X3D soll gleich viermal schneller als jene des Ryzen 9 7950X arbeiten? Diese Mär kann ComputerBase widerlegen.

www.computerbase.de

Thanks, that makes MUCH more sense.

 

[Hail The Brain Slug]

Sounds like something Strix Point and Granite Ridge/Granite Ridge-X designs may need to work on.

Not really sure the value in this when the iGPU lives on the IOD so it has to cross the fabric to touch L3 or RAM.

 

[Thibsie]

Not really sure the value in this when the iGPU lives on the IOD so it has to cross the fabric to touch L3 or RAM.

Indeed. But could L3 gets used as LLC before date goes thru IOD ? Zen5/Zen6 ?