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!

 

[Exist50]

Genoa.... I will not bother to link, as its all in benchmarks in this thread.

I assumed it would have to be Bergamo, because Genoa doesn't match your claim. But I should know better by now than to humor them.

AMD EPYC 9754 Benchmarks For The 128-Core Bergamo Review - Phoronix

www.phoronix.com

 

[jpiniero]

You're referring to Bergamo, correct? So can you point me towards where I can buy a Bergamo server?

Dell, of all places. They have a model which you can configure the 128 core and the 112. The 128 core is an extra 16 grand per processor over the base.

 

[Exist50]

Dell, of all places. They have a model which you can configure the 128 core and the 112. The 128 core is an extra 16 grand per processor over the base.

Thanks. I didn't notice that they added it to their lineup. Example link for the curious. Can configure up to an Epyc 9754. https://www.dell.com/en-us/shop/ser...eredge-r7615-rack-server/spd/poweredge-r7615/

 

[Markfw]

I assumed it would have to be Bergamo, because Genoa doesn't match your claim. But I should know better by now than to humor them.

AMD EPYC 9754 Benchmarks For The 128-Core Bergamo Review - Phoronix

www.phoronix.com

Oh, I guess you need to specify decimal places to make you happy. 1.6 times the performance at 0.7 the power for Genoa. Bergamo is the one that 2x and 2x. I mean when its a matter of 5-10% then decimals are required,. At 30-50% or more its kind of not even relevant.

But you will argue any point that does not match your agenda.

If you were a data center manager and bought a CPU/system that was almost 1/2 the performance at even 30%-40% more power, you would not be doing your company a service.

 

[Hitman928]

Isn't it about revenue most of the times anyway?

Maybe Intel may change to units since that's the more favourable number for them.

Most market share reports from 3rd parties are for volume. Pretty sure the companies themselves care more about the revenue, just wanted to point out that what Lisa Su was talking about isn’t the same thing as what most market reports estimate.

 

[Exist50]

Oh, I guess you need to specify decimal places to make you happy. 1.6 times the performance at 0.7 the power for Genoa.

The 8490H 2P system scored 1027.21 at an average of 568.83W, vs the 9654 2P system with a score of 1580.79 at an average of 449.69W.

In other words, 1.54x the score, at 0.79x the power. So you claimed a 4x perf/watt difference, for what is really a 2.1x difference, and even your "correction" is wrong.

So tell me. Who has the agenda? The one lying about the data, or the one giving it? If you actually want to understand the market dynamics, you need to be working with real data, because that's what buyers use.

But as I said, I'm clearly wasting my time. This is not a new pattern from you.

 

[Abwx]

Isn't it about revenue most of the times anyway?

Maybe Intel may change to units since that's the more favourable number for them.

By stating that they sold 50% less in volume they just gave a hint of what is going on.

Since their DC revenue didnt shrink by 50% (-40% IIRC), and since AMD is taking marketshare mainly in the higher priced SKUs then this would had involved more than 50% lower DC revenue, yet their numbers are such that they seem to have an higher ASP since 50% lower volume reduced revenue by only 40%...

That s not logical at all, they should had seen way lower volumes in the segment where AMD is dominant, and eventually saved the day with lower priced SKUs, wich would had much reduced their ASP and induced way more than 40% lower DC revenue.

Methink that they flooded the market by selling their higher priced SKU at a big loss, say at -50%, but even at this slaughtered price this is still above their ASP, hence the revenue shrinking less than the volume %age wise.

 

[Abwx]

The 8490H 2P system scored 1027.21 at an average of 568.83W, vs the 9654 2P system with a score of 1580.79 at an average of 449.69W.

In other words, 1.54x the score, at 0.79x the power. So you claimed a 4x perf/watt difference, for what is really a 2.1x difference, and even your "correction" is wrong.

So tell me. Who has the agenda? The one lying about the data, or the one giving it? If you actually want to understand the market dynamics, you need to be working with real data, because that's what buyers use.

But as I said, I'm clearly wasting my time. This is not a new pattern from you.

9754 scored 1895 at 385W, why not using this exemple.?.

That s not 4x the perf/Watt but still 2.72x, wich is considerable, dunno how there can be people buying 8490H set ups, have you a logical explanation.?

Because personaly i have none, at least one that would be technicaly motivated...

 

[Geddagod]

That is only 33% more cores

sshhhh i cant math

 

[scannall]

Oh, and then comes your trusty OEM salesman and gets you a *nice price* on those new Intel models. "It's a deal, see ya next year".

Business is definitely not objective matter.

The thing is, that Intel sales person isn't selling parts. They are selling relationships. Same person goes to sell to the same company, and the same buyer. They talk about their families, maybe set up a fishing trip, how's life going. Looking good. Congrats on (insert achievement), oh and lets get this nasty 'ol paperwork out of the way. How many did you need again?

 

[Thibsie]

I have no evidence to support this, but I would guess he is talking about Intel virtually or literally giving away server CPU's to stop the loss of market share due to their server chips being so inferior at the moment. It is only logical, as AMD has proven in recent benchmarks to exceed Intel in performance by ~100% and use ~50% of the power in doing so. Why else would data centers keep buying CPUs that are inferior and cost them more to run, even at the same price !

You forgot idiocy.
AMD isn't even really compatible. Ya didn't that ?

 

[jpiniero]

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

TR Pro sighting - the 7985WX has 64 cores. There's probally a 7995WX with 96.

 

[igor_kavinski]

Is V-cache going to make its way into the TR Pro platform this time around?

 

[igor_kavinski]

They are selling relationships. Same person goes to sell to the same company, and the same buyer. They talk about their families, maybe set up a fishing trip, how's life going. Looking good. Congrats on (insert achievement), oh and lets get this nasty 'ol paperwork out of the way. How many did you need again?

That's how the Sacklers became billionaires.

 

[igor_kavinski]

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

8 DIMMs at DDR5-5600. Not bad!

 

[Ajay]

That's how the Sacklers became billionaires.

Even worse was done by the distributors. But as companies doing $100B US per year, even the justice department had to fold given the fleet of lawyer those companies would through at any court case. The USJD just didn't have the manpower to handle a case that large that would probably take 10 years to prosecute given the mountain of evidence and continuances, etc that would be thrown at them (even though they had a "smoking gun" memo).

 

[Abwx]

A few numbers at VDCZ for the iGPU less 7500F :

AMD Ryzen 5 7500F reviews are out, CPU to launch globally at $179 - VideoCardz.com

AMD 6-core Ryzen 5 7500F debuts without integrated graphics AMD has lifted the embargo on Ryzen 5 7500F processor that has been a subject of rumors for the past two weeks. AMD Ryzen 5 7500F, Source: Expreview Reviews going live on Sunday definitely sounds unusual, but this is the date set by...

videocardz.com

 

[jpiniero]

ASUS ROG laptop with AMD Ryzen 9 7945HX3D CPU has been listed by retailers - VideoCardz.com

AMD 3D V-Cache coming to laptops? AMD and ASUS might be working on a new system featuring Ryzen 7000X3D laptop series. It seems that ASUS is working on a new gaming laptop, and various sources have recently started mentioning it. The laptop is the ROG STRIX SCAR 17, identified by the product...

videocardz.com

Ooh, Zen 4D might be coming to gaming laptops.

 

[biostud]

ASUS ROG laptop with AMD Ryzen 9 7945HX3D CPU has been listed by retailers - VideoCardz.com

AMD 3D V-Cache coming to laptops? AMD and ASUS might be working on a new system featuring Ryzen 7000X3D laptop series. It seems that ASUS is working on a new gaming laptop, and various sources have recently started mentioning it. The laptop is the ROG STRIX SCAR 17, identified by the product...

videocardz.com

Ooh, Zen 4D might be coming to gaming laptops.

It would be weird if it didn’t at some point. It is low power and high performance.

 

[igor_kavinski]

It would be weird if it didn’t at some point. It is low power and high performance.

The really weird thing is the deafening silence from Intel on their version of V-cache. What's the point of having their own fabs if they can't come up with a competing solution? Raptor Lake with even slower eDRAM would be better than nothing. They can sell BGA versions to their mobo partners with only P-cores and eDRAM and push them as gaming optimized solutions.

 

[eek2121]

It would be weird if it didn’t at some point. It is low power and high performance.

I have a Cezanne based laptop, I would upgrade in a heartbeat if vendors can keep weight down while giving us a decent GPU with this X3D chip. My current laptop has a 3070 and is under 4 lbs (~1.8kg), but has a huge bulky charger. Get rid of the bulky charger (weighs probably almost as much as the laptop) and we will be in business.

Something like a 7945X3D/7900S with USB-PD charging and a high refresh rate 16” display would be awesome. Framework has the USB-PD situation under control, so I suspect other vendors will come around soon or maybe framework will step up the specs for their laptops.

 

[biostud]

The really weird thing is the deafening silence from Intel on their version of V-cache. What's the point of having their own fabs if they can't come up with a competing solution? Raptor Lake with even slower eDRAM would be better than nothing. They can sell BGA versions to their mobo partners with only P-cores and eDRAM and push them as gaming optimized solutions.

Maybe the benefit is less on the Intel platform?

 

[igor_kavinski]

Maybe the benefit is less on the Intel platform?

Maybe but the eDRAM access time on Crystal Well was slightly lower than that of DDR3 yet there were gains to be seen.

Now the gap between eDRAM latency and DDR5 latency has widened. I suspect there's benefits to be had but Intel must have rid themselves of eDRAM manufacturing capacity, shortsightedly thinking it had no future use. One of the many brilliant decisions that got them into their current mess of a situation.

 

[Exist50]

Maybe but the eDRAM access time on Crystal Well was slightly lower than that of DDR3 yet there were gains to be seen.

Now the gap between eDRAM latency and DDR5 latency has widened. I suspect there's benefits to be had but Intel must have rid themselves of eDRAM manufacturing capacity, shortsightedly thinking it had no future use. One of the many brilliant decisions that got them into their current mess of a situation.

Crystal Well eDRAM was in the range of 60ns of latency. Better than DDR4/5, maybe, but far from AMD's V-cache. And probably even harder to retrofit into a standard SKU. Doubt we'll see any non-stacked expansion cache from here on. Shame if ADM got canceled.

 

[BorisTheBlade82]

Maybe the benefit is less on the Intel platform?

I'd rather they simply don't have the engineering and advanced packaging capabilities needed.