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!

 

[HurleyBird]

At most, they'd lose a tiny bit of IPC due to cache differences. In terms of the rest of the core, should be 100% identical.

They could also gain a bit of IPC from faster timings, since the core is not targeting high clocks and smaller caches tend to be faster. Vanilla Zen 4 increased L2 cycles and L3 cycles, even though the later didn't grow.

 

[PJVol]

Wow, that's even worse than what I thought!

I'm sure there were design challenges led to some compromizes, and I believe AMD engineers really fought this time for every saved bit, joule, °C or whatever could gain them PPW/PPA.
So I'd not rush to judgement on that

 

[Carfax83]

I'm sure there were design challenges led to some compromizes, and I believe AMD engineers really fought this time for every saved bit, joule, °C or whatever could gain them PPW/PPA.
So I'd not rush to judgement on that

Who said anything about judgment? I don't have a problem admitting I was wrong about this, I was just surprised that it was that low. What was the default FCLK for Zen 3?

Anyway, this means that Zen 4 shouldn't have any issues with higher DDR5 speeds.

 

[Det0x]

Stock JEDEC 5200MT/s:
1733 : 2600 : 2600

Sweet spot 6000MT/s:
2000 : 3000 :3000

Max tuned 6400MT/s ?
2133 : 3200 :3200

Now it only depends on at what speed they did measured that 63ns..
With theoretical perfect scaling you could get ~51ns @ 6400MT/s if that 63ns was done @ 5200MT/s
(and 54.5ns @ 6000MT/s)

 

[coercitiv]

Now it only depends on at what speed they did measured that 63ns..
With theoretical perfect scaling you could get ~51ns @ 6400MT/s if that 63ns was done @ 5200MT/s
(and 54.5ns @ 6000MT/s)

I had to look for a better slide deck source than Anandtech, luckily Techpowerup delivered: here's the AMD EXPO slide with readable indicators for footnotes, here's the footnote for the 63ns latency. Based on this I would say the 63ns figure was obtained @ 6000 MT/s.

The other interesting part is the footnote for the "up to 11%" gains. Turns out their info from the official press release is wrong, the slide deck claim is "up to 11%" gains in gaming on 7950X by using DDR5-6000C30 versus JEDEC 5200.

 

[Joe NYC]

Greymon55 on Twitter is hinting at which models get V-cache... Looks like it's the 7800X, 7900X, and 7950X.

If true.... *deep inhale*

Spoiler

View attachment 66876

https://twitter.com/x/status/1564807986497081346

https://twitter.com/x/status/1564808672282558465

I just hope my current PC and graphics card can last until then...

My PC and GPU are 9 human years old, which is some 54 in dog years...

 

[Yosar]

They obviously meant the 6000CL30 v 6400CL32 that that Shady guy (and a few others) keeps bringing up.

Also nope either way. 10% better avg and say, 15% better percentile in extreme stock vs ultra-tuned scenarios is something I would never classify as "massive", in my book at least.
I know this is mostly enthusiast talk, but words kinda lose their meaning that way.

Upgrading from a 3600 to a 5800X3D (w/ a GA102 or N21 GPU), now that's what I call massive differences.

They don't know what they talk about. They didn't deliver any proof, but it's enough to put in google phrase '6000 vs 6400 DDR5 benchmark' to get benchmarks of those 'massive' gains.
First one Tom's hardware.

https://www.tomshardware.com/reviews/gskill-trident-z5-rgb-ddr5-6400-c32-review

CPU Index Geometric Mean 6400 CL32 vs 6000 CL36 (yeah, much worse latency on 6000 kit)
Result: 2513.81 vs 2496.27 . Less than 1% despite 6400 better latency (what I was writing earlier about latency?)
The funny thing is when we take only games actually 6000 kit is better 116.25 vs 116.19.
Oh, and they even tried to optimize those 6400 settings,.

"The Trident Z5 RGB memory kit has a great CAS Latency (32); however, its other timings need some work. Sadly, there wasn't a considerable margin for optimization. Even at 1.45V, we only reduced the tRCD, tRP, and tRAS from 39-39-102 to 37-37-97. "

Yeah, they couldn't overclock much $450 kit.

Next one anandtech (what we here here).
Gaming.

Intel Alder Lake DDR5 Memory Scaling Analysis With G.Skill Trident Z5

www.anandtech.com

Applications.

Intel Alder Lake DDR5 Memory Scaling Analysis With G.Skill Trident Z5

www.anandtech.com

Again minimal differences. Less than 1% between 6000 and 6400.

The last one TechpowerUp, They were not even bothering to test 6400 at launch. So here's review of 6400 CL32 kit from G.Skill.

G.SKILL Trident Z5 DDR5-6400 CL32 2x 16 GB Review

G.SKILL is testing waters with the Trident Z5 DDR5-6400, an ultra-low latency kit. With impressive timings of 32-39-39-102, G.SKILL is pulling out all the stops. Tag along as we put the Trident Z5 DDR5-6400 through its paces!

www.techpowerup.com

And again. Less than 1%. And in some games it even loses to 6000 CL40 kit (sounds familiar?).

I'm sure the guys watched some anonymous video on youtube where there were 'massive' gains between 6000 and 6400 kit.
And I'm sure the guy who put this video demonetized it, because it was only honest review for the world. Not some marketing blah blah blah.

I'm not debating that there are some gains with faster memory. Of course there are. I'm debating there are some 'massive' gains for intel (and not for AMD) between 6000 kit (AMD chose for benchmarks) and 6400 kit (optimal for intel).

 

[Joe NYC]

Doesn’t seem like a very good naming scheme, but the usually aren’t.

Latest leak seems to be confirming this.

9000 socket SP5
8000 socket SP6 (Sienna)

Also in the leak:
- Bergamo will be 16 cores per CCD
- Bergamo CCD will have 32 MB of L3 (or 2 x 16?)
- Bergamo clock 2.05 GHz base, 3.2 GHz max
- There will be Genoa-X (but we knew that)

AMD EPYC Genoa-X & Bergamo CPU Lineup Leaked - Zen 4C & Zen 4 V-Cache Chips With Up To 128 Cores, 16 Cores Per CCD, 4 GHz+ Clocks & 400W TDP (wccftech.com)

 

[Saylick]

Might as well give the real source credit... That way, no misrepresentation of info like WCCFTech normally does due to typos and general lack of understanding.

I think the interesting thing about the two Bergamo SKUs is the core count vs TDP.

112 cores = 340W while 128 cores = 360W. Clocks are roughly the same.

That's 20W for an extra 16 cores, or roughly 1.25W per core. With 128 cores, that means the cores themselves use up: 128 cores x 1.25 W/core = 160W, leaving 200W for the IO?

Seems rather non-intuitive what's going on here. It surely can't be 200W for the IO, right?

https://twitter.com/x/status/1565184763777859584

 

[TESKATLIPOKA]

Might as well give the real source credit... That way, no misrepresentation of info like WCCFTech normally does due to typos and general lack of understanding.

I think the interesting thing about the two Bergamo SKUs is the core count vs TDP.

112 cores = 340W while 128 cores = 360W. Clocks are roughly the same.

That's 20W for an extra 16 cores, or roughly 1.25W per core. With 128 cores, that means the cores themselves use up: 128 cores x 1.25 W/core = 160W, leaving 200W for the IO?

Seems rather non-intuitive what's going on here. It surely can't be 200W for the IO, right?

https://twitter.com/x/status/1565184763777859584

View attachment 66932
View attachment 66933

I think It's a better binned SKU, with lower voltage needed for comparable clocks, not that a single core consumes only ~1.25W. I think It is 2-2.25W or 248-288W combined.
IO is power hungry, but not that power hungry.

 

[DrMrLordX]

Does that mean DDR6 is right around the corner?

No. JEDEC has a planned launch of DDR6 in 2025. Supposedly.

 

[SteinFG]

AMD EPYC 9004 "Genoa-X" and "Bergamo" CPUs SKU lists leaks out, EPYC 9754 features 128 Zen4c cores - VideoCardz.com

AMD goes EPYC with Genoa-X and Bergamo Hardware leaker YuuKi_AnS shares new information on AMD naming convention for Genoa-X and Bergamo CPUs. AMD EPYC Genoa-X/Bergamo Naming Convention, Source: YuuKi_AnS AMD is yet to officially launch its Genoa data-center CPU series. The company also...

videocardz.com

Hey, adored tv was right about 64 zen 4c cores on sp6. So sp6 maxes out at 32 normal zen 4 cores after all.

So 6 ddr5 channels on sp6 is also true I assume

 

[SteinFG]

Then I think threadripper 7000 will use sp6 socket for lower cost. 6 ddr5 channels is already ~1.5 times more bw than 8 ddr4 channels that threadripper pro 5000 has. Same with 96 PCIe 5.0 lanes instead of 128 PCIe 4.0 on tr pro 5000.

 

[AtenRa]

Ok a few thoughts for this October,

If we take as a given that 7600X @ $300 will be 2% to 5% faster vs $569 i9 12900K and $439 Ryzen 5800X3D ........

A lot of people will happily invest in B650 + 7600X + DDR5 leveraging the upgradability to ZEN5 later in 2024.
Both AM4 and Socket 1700 will be dead after Raptor Lake launch at the end of the month and although RL may bring more vfm (with DDR4) it may not get that much love from many.

 

[mikk]

it is reported that the AMD Ryzen 9 7950X will throttle below 5.0 GHz under intensive tasks as it comes with a 95C Thermal Threshold or TjMax and you'll need some beefy coolers to keep the chip running under that. At full load in stock configuration, the CPU is said to consume up to 230 Watts of power and operate at up to 95C. The Ryzen 5 7600X is also a similar scenario with the chip consuming up to 120W of power at full load and hitting temps of up to 90C.

AMD Ryzen 7000 CPUs Reportedly Run Hot, Ryzen 9 7950X Hits Up To 95C Thermal Threshold at 230W, Ryzen 5 7600X Up To 90C at 120W

AMD Ryzen 7000 CPUs including the Ryzen 9 7950X are going to run really hot and consume the full package power as reported in a rumor.

wccftech.com

Zen 4 is reportedly a hot running chip.

 

[eek2121]

Then I think threadripper 7000 will use sp6 socket for lower cost. 6 ddr5 channels is already ~1.5 times more bw than 8 ddr4 channels that threadripper pro 5000 has. Same with 96 PCIe 5.0 lanes instead of 128 PCIe 4.0 on tr pro 5000.

One can only hope it will be lower cost.

 

[Det0x]

AMD Ryzen 7000 CPUs Reportedly Run Hot, Ryzen 9 7950X Hits Up To 95C Thermal Threshold at 230W, Ryzen 5 7600X Up To 90C at 120W

AMD Ryzen 7000 CPUs including the Ryzen 9 7950X are going to run really hot and consume the full package power as reported in a rumor.

wccftech.com

Zen 4 is reportedly a hot running chip.

This will be be the intended/normal behavior for Zen4

...The smaller heatspreader area on Ryzen 7000 package design only makes things worse. We discussed the cooling difficulty and why the heatspreader is smaller in our Computex piece in May.

Due to the cooling difficulty and higher power limits on Ryzen 7000 Dekstop, the philosophy has changed to follow a more laptop-like approach. Instead of fixing a clockspeed or power limit, the processor is now temperature limited. This means maximizing performance for a given cooling capability. The processor will adjust to stay within the temperature limits. This also means a more significant performance difference that changes with cooling ability.

The CPU will try to boost until it hit maximum boost clock or intended tjmax which it seemingly aims for now.
(by adding more EDC/TDC instead of vcore i guess?)

 

[leoneazzurro]

View attachment 66913
Stock JEDEC 5200MT/s:
1733 : 2600 : 2600

Sweet spot 6000MT/s:
2000 : 3000 :3000

Max tuned 6400MT/s ?
2133 : 3200 :3200

Now it only depends on at what speed they did measured that 63ns..
With theoretical perfect scaling you could get ~51ns @ 6400MT/s if that 63ns was done @ 5200MT/s
(and 54.5ns @ 6000MT/s)

Those are the limits of organic substrate packaging of Zen4 (probably only a revision of the one used on Zen3). Thinking to get much higher speed out of such packaging was only wishful thinking. Now, Zen 5 could be very different in this aspect, and thus really improving the interconnect speed (costs - are another question).

 

[poke01]

Okay but Papermaster is such a cool last name!

 

[igor_kavinski]

Zen 4 is reportedly a hot running chip.

AMD/TSMC better start working on integrated cooling channels or something. Otherwise, the future for high speed CPUs doesn't look good.

One possibility is to have an additional IHS (with a cutout for connection with the mobo socket) at the bottom of the die so heat has more surface area to dissipate from. But that will surely require a redesigned socket.

 

[MarkPost]

AMD Ryzen 7000 CPUs Reportedly Run Hot, Ryzen 9 7950X Hits Up To 95C Thermal Threshold at 230W, Ryzen 5 7600X Up To 90C at 120W

AMD Ryzen 7000 CPUs including the Ryzen 9 7950X are going to run really hot and consume the full package power as reported in a rumor.

wccftech.com

Zen 4 is reportedly a hot running chip.

avx-512 ON or OFF in AIDA? thats the key here. I bet ON (default option)

 

[poke01]

AMD/TSMC better start working on integrated cooling channels or something. Otherwise, the future for high speed CPUs doesn't look good.

It maybe that AMD pushed Zen 4 clocks too high. M1 Ultra (16P + 4E) which is also a TSMC 5nm chip running at 3.2GHz has max temps of 50 celsius under 100% cpu load.

Still Zen 4 will be fine.

 

[uzzi38]

It maybe that AMD pushed Zen 4 clocks too high. M1 Ultra (16P + 4E) which is also a TSMC 5nm chip running at 3.2GHz has max temps of 50 celsius under 100% cpu load.

Still Zen 4 will be fine.

There's a lot more nuance than just clock speeds, core count and node when talking about temperatures.

 

[uzzi38]

AMD Ryzen 7000 CPUs Reportedly Run Hot, Ryzen 9 7950X Hits Up To 95C Thermal Threshold at 230W, Ryzen 5 7600X Up To 90C at 120W

AMD Ryzen 7000 CPUs including the Ryzen 9 7950X are going to run really hot and consume the full package power as reported in a rumor.

wccftech.com

Zen 4 is reportedly a hot running chip.

I know I've been saying that Zen 4 is very difficult to cool... I'd have my doubts on the specifics of this article.

 

[Timmah!]

avx-512 ON or OFF in AIDA? thats the key here. I bet ON (default option)

Lets hope so. The main allure of Zen4 over RPL is, aside of not being hybrid, the superior power efficiency - and to me personally, that comes with expectation of lower temps and less noise. Its actually more important that lower electricity bills (though that might change in the future given current energy shenanigans in Europe). All in all, its kind of pleasant to have relatively silent computer and not having to worry whether you are degrading your CPU everytime you put it under heavier load.