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!

 

[Thibsie]

Do you think AM5 socket will support other generation beyond 7000 series? I mean, if AMD hypothetically releases 24 core Zen4+ or Zen5 in a year or 2, do you think they will fit these 670 boards?
If yes, thats one more thing to consider when choosing between Raptor Lake and Zen4.

It will

 

[gdansk]

My guess is AM5 will support Zen 4, Zen 4 2023 refresh, Zen 5 and Zen 5 2025 refresh. And some APUs but that might end earlier, like how AM4 won't get Rembrandt.

But we haven't heard anything official from AMD have we?

 

[carrotmania]

There's a lot of noise being added to this thread by (at least) 2 diehard anti-AMD peoples (who also claim to have HAVE AMD systems... sure). Believing every slight bit of speculative pro-Intel info, despite Intel overpromising and underdelivering on everything (yes, even AlderLake) in the last decade. But then being hard locked into just the numbers AMD gives out (eg the whole thing with 15% rather than >15%) is hilarious.

Comparing ADL to Zen3, with a year between them, but then refusing to do the same with ADL and Zen4, for instance... You understand the rest of us KNOW you're posting in bad faith, right?

Guys, if you don't want to be here, please just leave the thread, and stop making multiple posts between other people with more substantial content to add than yourselves. That would be lovely.

 

[maddie]

They will be able to scale the this new core for several generations - THEY ALREADY HAVE IT, while AMD is at the end of the road for Zen architecture. AMD is not guaranteed to succeed with Zen5 architecture, they might get delayed, might take a page too many from Bulldozer or K10 "success" stories. They are sure executing real good lately, but local circlejerkers were already overjoyed by "massive IPC increase of Zen4", so while unlikely it might happen.
The reality is that Zen3 is currently loosing in PPC to ADL very substantially and mediocre PPC gains ensure that Raptor Lake will also be in the lead versus Zen4. I'd be damned if AMD was not more efficient with full node advantage, but their days of performance leadership over 2013 year designs in 2020 are over.

Intel's future will be decided by their process, if they can come up with Intel 4 and move to Intel 3 on time, they will be just fine.




Common, even Anandtech with their JEDEC loving, but not for real world testing found:



Hard to imagine Intel not continuing to iterate on these things, they sure plan to use 32 E-Cores to make AMDs day miserable in throughput tests. The power "efficiency" of core pushed to 4Ghz by marketing on 10nm might not apply to chip on Intel 3 at correct voltage sipping power.

Just one point. You stated.

"I'd be damned if AMD was not more efficient with full node advantage, but their days of performance leadership over 2013 year designs in 2020 are over."

What is this node advantage?

 

[JoeRambo]

There's a lot of noise being added to this thread by (at least) 2 diehard anti-AMD peoples (who also claim to have HAVE AMD systems... sure).

Everyone wishing to join ADF team of circlejerkers (and before even considering posting here) should provide some proof of ever owning AMD system. Extra points given if You owned some junk produced by AMD, brand loyalty is highly rewarded here ( by $550 A64 X2 4200 usually, enjoy Your fan only pricing ).






Inflammatory posting just isn't allowed in the tech forums.

esquared
Anandtech Forum Director

 

[Timmah!]

Zen 4+/Zen 5 yes but what core counts it would have is still TBD.

Thanks. No doubt the core-counts of those series are unknown at this point, but given what we know about future Intel stuff (Meteor or Arrow lake to be 8P + 32E), i think AMDs 16C wont cut it, regardless of other improvements.
OFC those Intel core-counts are only rumors at this point and might turn out false.

 

[moinmoin]

that where in response in equally ridiculous pro-Intel post

You are aware, that you are in the Zen 4 thread not in the Intel one, aren't you? Troll elsewhere, pretty please.

 

[DisEnchantment]

I don't think there is need for sensationalism over a CPU uArch on either side.
These things takes years of design and implementation, would lose hair fairly quick getting riled up over such long spans of design time.

 

[JoeRambo]

I don't think there is need for sensationalism over a CPU uArch on either side.
These things takes years of design and implementation, would lose hair fairly quick getting riled up over such long spans of design time.

Fully agreed. Over long time I have especially noticed that such sensationalism always starts when when companies are neck to neck and not when one of them is clearly in the lead. We already had Dr. Sharikou era of "Intel Bankrupt in Q3 07", for sure we will survive this latest bout of sensationalism.

Still, out of all clearly pro-AMD people, i respect only @Markfw as he actually went out and bought 12700K to compare it in his very special usage case. Others just blabber around with Skylake in mind and Alder Lake equals 12900K SKU and 666W of power use to them. Usually they own 3rd hand Z1 2600, but love talking about servers and imminent Intel's doom ( where of course every single sale is top SKU of 64C ).

Welcome to Anandtech Forums, where people love to discuss hardware they don't own!

 

[inf64]

Fully agreed. Over long time I have especially noticed that such sensationalism always starts when when companies are neck to neck and not when one of them is clearly in the lead. We already had Dr. Sharikou era of "Intel Bankrupt in Q3 07", for sure we will survive this latest bout of sensationalism.

Still, out of all clearly pro-AMD people, i respect only @Markfw as he actually went out and bought 12700K to compare it in his very special usage case. Others just blabber around with Skylake in mind and Alder Lake equals 12900K SKU and 666W of power use to them. Usually they own 3rd hand Z1 2600, but love talking about servers and imminent Intel's doom ( where of course every single sale is top SKU of 64C ).

Welcome to Anandtech Forums, where people love to discuss hardware they don't own!

No offense, but you are talking just like a rabid intel fanboi in this post .

I for one still have my Haswell i5 system, it was a nice chip but doesn't hold a candle to my Pinnacleridge 12nm Zen1+. As for going nuts over PC hardware, you should take a look in the mirror as you claim that Zen4 is Zen3 shrink with AVX512 and more cache (which is 100% false). Last time I checked, die shrinks brought 1-3% IPC (if that much) and were pure ports to smaller nodes. Zen4 is a new core, not a brand new like Zen3, but still bringing a lot of changes. The real new core will be Zen5, and with more changes than what Zen3 brought.

And yes, intel is very much behind AMD in almost everything, especially server space. The only stronghold they have is pure ST desktop performance, and as we agreed, this will soon be a non issue (+-5% difference). They will be a whole node behind with atrociously bad perf/watt and much less threads in almost every segment of the market. The only saving grace could be lower price for lower end models in their stack, but I'll believe it when I see it.






Use of the word fanboy/fanboi or any other iteration isn't allowed in the CPU forum as per the rules.


esquared
Anandtech Forum Director

 

[maddie]

Fully agreed. Over long time I have especially noticed that such sensationalism always starts when when companies are neck to neck and not when one of them is clearly in the lead. We already had Dr. Sharikou era of "Intel Bankrupt in Q3 07", for sure we will survive this latest bout of sensationalism.

Still, out of all clearly pro-AMD people, i respect only @Markfw as he actually went out and bought 12700K to compare it in his very special usage case. Others just blabber around with Skylake in mind and Alder Lake equals 12900K SKU and 666W of power use to them. Usually they own 3rd hand Z1 2600, but love talking about servers and imminent Intel's doom ( where of course every single sale is top SKU of 64C ).

Welcome to Anandtech Forums, where people love to discuss hardware they don't own!

It would help if you use accurate info instead of fabricating " my facts". I will give you the benefit of the doubt and assume that you are ignorant of some factual data instead of using deliberately misleading arguments.

 

[Abwx]

News at eleven? Marketing SKU using 250W again, when pushed. AMD is pushing their own CPU too, i doubt V/F curve is that generous and they need to match 24 cores now, not 16.
so might be above 142W this time.

A 7950X@4GHz should consume 65W, that s the starting point, 20% higher frequency will get you at 130W@4.8GHz at worst.

That make power increasing as a 3.8 exponent vs frequency, not exactly what we can call a generous V/f curve, actually it should be slighly better but let s her on the cautious side of things, AMD stated a 3 exponent, that is, power as a cubic function of frequency, but that s surely for servers class frequencies.

 

[FangBLade]

AMD triggered permanently Intel fanboys, that's the best evidence of AMD success, and Intel is no longer considered best brand, most jokes and memes nowdays are about Intel. Some Intel boys here need to accept that AMD won't repeat bulldozer era, learn to live with it!




Use of the word fanboy/fanboi or any other iteration isn't allowed in the CPU forum as per the rules.


esquared
Anandtech Forum Director

 

[JoeRambo]

A 7950X@4GHz should consume 65W, that s the starting point, 20% higher frequency will get you at 130W@4.8GHz at worst.

Talking from my experience with 5950x i have two worries with highest SKU of Zen4:

1) The chip will have too high wattage requirement to hit region of 5.5+ghz or so. Talking about 25-30W per 2T loaded core, where 5950x was ~20W for 5ghz. So power efficiency at low core loads will be thrown under the bus, despite process and architecture being efficient. This is esp painful reaching PPT region, where you will have that 230 / (25-35W ) and efficiency will suffer big time.
2) Already discussed, but due to hunting higher clocks in MT, they will again push against PPT with high voltages and wattages in hockey stick V/F region. 5950x was very efficient at full load point, but clocks were 3.75 or so and hockey stick really started way above 4Ghz. Something like Intel has with E-Cores, except they clock/volt them in hockey stick blade tip region.

It's easily fixed with fixing clocks / undervolt, but efficiency of SKU @ stock suffers.


EDIT: reaching PPT region means graphically, so You already have it ~7-8C on 5950x for 4600mhz clocks and notice that nasty spike to 110W at 4c for 4725mhz

 

[Det0x]

1) The chip will have too high wattage requirement to hit region of 5.5+ghz or so. Talking about 25-30W per 2T loaded core, where 5950x was ~20W for 5ghz. So power efficiency at low core loads will be thrown under the bus, despite process and architecture being efficient. This is esp painful reaching PPT region, where you will have that 230 / (25-35W ) and efficiency will suffer big time.
It's easily fixed with fixing clocks / undervolt, but efficiency of SKU @ stock suffers.

I did some comparison runs for my 5950x and we seem to have some mismatch in the numbers.
In what workload did you measure 20w for a singlecore @ 5ghz ? For a comparison, i'm getting this in Cinebench r23:

~11.4watt for a static 5ghz singlecore Cinebench r23 run


~15.5watt for a sustained ~5050mhz PBO CO singlecore Cinebench r23 run


~14.8watt for a sustained ~4950mhz stock singlecore Cinebench r23 run (without PBO CO)

 

[JoeRambo]

I did some comparison runs for my 5950x and we seem to have some mismatch in the numbers.

It's from Anandtech's own testing in review, but i had very similar slopes and numbers ( just without HT "power tax") in my testing:
I think You are missing HT thread and obviously your PPT is ballistic as is due to OC which might impact things in some way.

I was focusing not a per core, but PPT/core as L3 caches and IF links light up even if there is single very busy thread ( something that probably happens at 4C Anandtech point, is that 2x CCDs are barely loaded and power usage goes up, having bad efficiency.

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

 

[Abwx]

Talking from my experience with 5950x i have two worries with highest SKU of Zen4:

1) The chip will have too high wattage requirement to hit region of 5.5+ghz or so. Talking about 25-30W per 2T loaded core, where 5950x was ~20W for 5ghz. So power efficiency at low core loads will be thrown under the bus, despite process and architecture being efficient. This is esp painful reaching PPT region, where you will have that 230 / (25-35W ) and efficiency will suffer big time.
2) Already discussed, but due to hunting higher clocks in MT, they will again push against PPT with high voltages and wattages in hockey stick V/F region. 5950x was very efficient at full load point, but clocks were 3.75 or so and hockey stick really started way above 4Ghz. Something like Intel has with E-Cores, except they clock/volt them in hockey stick blade tip region.

It's easily fixed with fixing clocks / undervolt, but efficiency of SKU @ stock suffers.


EDIT: reaching PPT region means graphically, so You already have it ~7-8C on 5950x for 4600mhz clocks and notice that nasty spike to 110W at 4c for 4725mhz

View attachment 63042

I said 4.8GHz on purpose since it should be or so the all cores frequency@130W in loads like Cinebench.

From Computerbase numbers the 5950X use 130W@4GHz in this bench at stock settings, that s almost 10% higher frequency than your estimation based at 3.7GHz.

4GHz MT stock frequency :

AMD Ryzen 5950X, 5900X, 5800X & 5600X im Test: Analysen zu IPC, Takt und Latenzen

AMD Ryzen 5000 im Test: Analysen zu IPC, Takt und Latenzen / AMD vergleicht die IPC jetzt anders / IPC-Zuwachs im Testparcours

www.computerbase.de

Power can be estimated here :

AMD Ryzen 5950X, 5900X, 5800X & 5600X im Test: Leistungsaufnahme und Temperatur

AMD Ryzen 5000 im Test: Leistungsaufnahme und Temperatur / Leistungsaufnahme von Leerlauf bis Volllast

www.computerbase.de

 

[Det0x]

It's from Anandtech's own testing in review, but i had very similar slopes and numbers ( just without HT "power tax") in my testing:
I think You are missing HT thread and obviously your PPT is ballistic as is due to OC which might impact things in some way.

I was focusing not a per core, but PPT/core as L3 caches and IF links light up even if there is single very busy thread ( something that probably happens at 4C Anandtech point, is that 2x CCDs are barely loaded and power usage goes up, having bad efficiency.

AMD Zen 3 Ryzen Deep Dive Review: 5950X, 5900X, 5800X and 5600X Tested

www.anandtech.com

Yes you are correct, difference was probably me running singlethread workload on a single core.. When i force two threads on a single core with affinity i'm getting the ~same power usage (but windows will not schedule work this way by default)

18.6watt for a sustained ~5000mhz PBO CO 2thread singlecore Cinebench r23 run

Do note i'm running AMD agesa 1.2.0.3c.. On all newer versions (1.2.0.4 upto current 1.2.0.7) AMD limits max set-vcore to 1425mv instead of 1500mv then running above 140 EDC, which will lower power usage again

 

[JoeRambo]

From Computerbase numbers the 5950X use 130W@4GHz in this bench at stock settings, that s almost 10% higher frequency than your estimation based at 3.7GHz.

4GHz MT stock frequency :

It's not really estimation, Anandtech had 3775 in review and 3.7 here by our member as well:

Page 234 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 234 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

The only estimations are Your numbers like 4.8Ghz

 

[DrMrLordX]

i respect only @Markfw

Really? Interesting.

Talking from my experience with 5950x

Do you really think a 7950X will use up to 30W core power in a strictly ST workload using default boost clocks?

Also why would the average user even care about that? Almost nobody complained about Vermeer's boost behavior or power consumption (except for the 5800X overheating).

 

[Abwx]

It's not really estimation, Anandtech had 3775 in review and 3.7 here by our member as well:

Page 234 - Discussion - Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

Page 234 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

The only estimations are Your numbers like 4.8Ghz

You have to use a corner case, Computerbase use mainly Cinebench and sometimes Blender for power measurement, both realistic loads and indicative of perfs,, while the AT page you linked is based on measurement made using OCCT, a CPU stress test, that s why there s no related perf graph...

 

[JoeRambo]

You have to use a corner case, Computerbase use mainly Cinebench and sometimes Blender for power measurement, both reaslistic loads and indicative of perfs,, while the AT page you linked is based on measurement made using OCCT, a CPU stress test, that s why there s no related perf graph...

It was ~3.9 in my own testing, so safe to say it depends on load, chip and motherboard ( and BIOS in the light of power cheats unvealed by Hwinfo64).

Do note i'm running AMD agesa 1.2.0.3c.. On all newer versions (1.2.0.4 upto current 1.2.0.7) AMD limits max set-vcore to 1425mv instead of 1500mv then running above 140 EDC, which will lower power usage again

That's good to know, unfortunately EDC comes late into play and will not impact low core count numbers.

Also why would the average user even care about that? Almost nobody complained about Vermeer's boost behavior or power consumption (except for the 5800X overheating).

At the risk of stating obviuos: from this discussion about modern turbos, PPT, voltages, V/F curves and frequency points it is well possible to design variuos SKUs. And depending on V/F curve, my worry is that such design @ 5.5+ghz might be a bridge too far and efficiency might suffer tremendously.

Intel gave 12900K 100W too many to chase 5950x, AMD might push 100-200mhz too far to chase 13900K. Clear enough?

 

[Markfw]

You have to use a corner case, Computerbase use mainly Cinebench and sometimes Blender for power measurement, both realistic loads and indicative of perfs,, while the AT page you linked is based on measurement made using OCCT, a CPU stress test, that s why there s no related perf graph...

I have 6 5950x, and in the DC arena, I can even count how many have one. And they are all the same, fully loaded, 142 watt as advertized, and 3.7 ghz. Most are on linux, so I don't have the luxury of Ryzen master, but the points generated are essentially all the same.

All on high end air, all stock. People usually don't buy a 5950x to let it sit idle.

 

[moinmoin]

Intel gave 12900K 100W too many to chase 5950x, AMD might push 100-200mhz too far to chase 13900K. Clear enough?

I'd say at this point you should worry more about Intel, in an Intel thread. With AMD chips ECO mode and tweaking PBO make it easy to tame any excessive power usage at stock.

 

[carrotmania]

No offense, but you are talking just like a rabid intel fanboi in this post.

There's little to no chance he was a subject of my post. /s