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!

 

[Det0x]

There is no 350W power limit. That was a result of someone playing with motherboard settings to get 40,000 pts in Cinebench R23 multithread.

Ok?

A more recent report coming from Hungarian publication ProHardver suggests that this PL override is actually a feature easily accessible on the upcoming Intel 700 series motherboards. It looks like Nvidia and now Intel are going to ignore all the climate change concerns advocating for optimized power consumption just to get ahead of AMD.

Only select Intel 700 motherboard models are rumored to include this power override feature that raises the PL cap to 350 W. Additionally, this “Extreme Performance Mode” is said to require top-of-the-line AIO cooling solutions, and most likely a power supply upgrade too. Surely, not many were expecting a CPU with TDPs that match those of a high-end GPU so soon. Nevertheless, Nvidia’s upcoming RTX 4000 series is expected to raise the bar once again, so why would Intel sit on its hands?

Tests performed on the aforementioned i9-13900K qualification sample running at 345 W showed only 15% gains in multi-thread loads over samples running in 253 W PL2 mode. Not exactly worth the additional 100 W plus the costs of a new cooling system, new premium motherboard and eventually a new PSU, but, depending on how Intel polishes things for the rumored October launch, final Raptor Lake CPU versions could show better results.

It is said that Intel Raptor Lake-S CPUs from the Core i9 line are to support a power limit of 350W, which is on top of the default (PL1/PL2) power limits going up to 241W. This feature is to become available on some Intel 700 motherboards, and it is reportedly showing better gains compared to Alder Lake power limit uncap.

As noted by ProHardver, with Alder Lake CPUs power uncapping beyond PL2 limit did not provide noticeable performance improvement, mainly because the CPU was not designed for such power. For Raptor Lake-S this concept will be expanded on supported platforms.

The 125W PL1 and 253W PL2 limits will remain as they are for Core i9-13900K series to ensure compatibility with Intel 600 series motherboards. However, for supported Intel 700 series motherboards and systems equipped with adequate cooling solutions, there will be an option to enable 350W mode. The site underlines the importance of the cooling, 350W mode will require ‘serious cooling solutions’.

The Intel Core i9-13900K CPUs or the Raptor Lake Core i9 "K" series in general, are going to get a new profile known as "Extreme Performance". This is also known as "Unlimited Power" setting on certain motherboards. What this profile essentially does is unlock Raptor Lake's power limit and let it reach up to a power-gobbling 350 Watts at stock loads. The higher power results in higher performance but at the cost of the aforementioned power budget and thermals.

This "Extreme Performance" mode will be amongst the several overclocking and tuning options coming to the Raptor Lake "K-series" CPU lineup. An interesting thing that the site states is that high-end Z790 motherboards will be required and only a certain will be qualified to support the feature. It is evident that a 350W power design will require a lot of cooling, not just for the CPU but also for the motherboard VRMs. Also, whether this feature remains exclusive to Z790 boards or also comes to Z690 is something only time will tell. The high-end Z690 boards are capable enough so some vendors may release the BIOS with "Extreme Performance" mode for current-gen boards too.

I wont bother searching more, but i know ive seen a few twitter posts from different sources that this is turned on by default on the motherboards intel plan to send out to reviewers..

But enough intel stuff in this Zen4 thread

 

[nicalandia]

Aren't you discounting Raptor Lake?

Raptor Lake is not relevant at this stage, I mean it will be released later, leaving Zen4 to Reign Supreme..

Haven't you seen this meme?



AMD is positioning themselves as the Gaming Leaders until late 2024 when Zen5 and if Intel is able to make Arrow Lake for desktop or something like that. We will be stuck with 13900K/S and 7950X/7950X3D for the next two years...

 

[deasd]

Thats the weird thing.. GB is not taking much advantage of the Crypto Score for Zen4, who knows why. Perhaps a beta bios or something.

It's not necessary to remove Crypto test. Because it can only give additional < 20% uplift to Crypto score when you use AVX512.

https://twitter.com/x/status/1439564626719506432

only 5% of this '<20%' add up to overall ST score when utilizing AVX512, finally +2% to the ST score. It's already a REMOVAL by Geekbench itself.

But the Crypto MT is strange to say the least.

 

[Abwx]

Games scale completely different to applications. 29% faster ST doesn't mean 29% faster in games.

At 70% scaling, wich is not a lot, roughly 20% would be left as real perf improvement.

According to Computerbase the 12900K has only 10% advantage over a 5600X in games for FPS and 16% for frametime, no wonder that AMD state a 7600X as being 5% faster than the 12900K.

CPU-Benchmark: Prozessor-Vergleich

Welche CPU für Games und Apps kaufen? ComputerBase leistet Kaufberatung mit Intel Core i-14000 und AMD Ryzen 7000(X3D) auf Basis von Tests.

www.computerbase.de

Looking at thoses average it looks like in the mind of a lot of people the 12900K perfs are vastly overated, hence they post accordingly, thinking that those 10% are actually 30 if not 40%...

But the Crypto MT is strange to say the least.

Methink that AES is processed by two specialised units and that it doesnt take a lot of cores to make full use of this hardware block, IIRC that was displayed on an old Zen slide.

 

[CakeMonster]

AMD is positioning themselves as the Gaming until 2024 when Zen5 and if Intel is able to make Arrow Lake for desktop or something like that. We will be stuck with 13900K/S and 7950X/7950X3D for the next two years...

That's a fun meme, but we don't even know if the 7950X3D will be a real thing, and if V-Cache is not available in 16c, then a lot of people are 'stuck' with the 7950X or 7800X3D and can't check all the boxes that we presume the 13900K will.

 

[tamz_msc]

Ok?




I wont bother searching more, but i know ive seen a few twitter posts from different sources that this is turned on by default on the motherboards intel plan to send out to reviewers..

But enough intel stuff in this Zen4 thread

The so-called 350W "Extreme OC" mode is a spin given by some rumor-mongers to a leak by Raichu, which shows the 13900K consuming 345 W when you set PL2 manually to 4096 W, which is the limit allowed in current motherboards.

Core i9-13900K breaks 40K points in Cinebench R23 with unlimited power and 5.8 GHz clock - VideoCardz.com

Intel Core i9-13900K compared with and without power limiting settings Leaker OneRaichu shares new benchmark results featuring the flagship Raptor Lake CPU. The 24-core and 32-thread Raptor Lake processor scores 2290 points in single-core Cinebench R23 test and 35693 points in multi-core. It...

videocardz.com

There is no basis to claim that such an extreme OC mode exists.

 

[tamz_msc]

At 70% scaling, wich is not a lot, roughly 20% would be left as real perf improvement.

According to Computerbase the 12900K has only 10% advantage over a 5600X in games for FPS and 16% for frametime, no wonder that AMD state a 7600X as being 5% faster than the 12900K.

CPU-Benchmark: Prozessor-Vergleich

Welche CPU für Games und Apps kaufen? ComputerBase leistet Kaufberatung mit Intel Core i-14000 und AMD Ryzen 7000(X3D) auf Basis von Tests.

www.computerbase.de

Looking at thoses average it looks like in the mind of a lot of people the 12900K perfs are vastly overated, hence they post accordingly, thinking that those 10% are actually 30 if not 40%...

All you do is post ComputerBase results to "validate" your claims. Do you even look at other sources for data?

 

[Markfw]

The so-called 350W "Extreme OC" mode is a spin given by some rumor-mongers to a leak by Raichu, which shows the 13900K consuming 345 W when you set PL2 manually to 4096 W, which is the limit allowed in current motherboards.

Core i9-13900K breaks 40K points in Cinebench R23 with unlimited power and 5.8 GHz clock - VideoCardz.com

Intel Core i9-13900K compared with and without power limiting settings Leaker OneRaichu shares new benchmark results featuring the flagship Raptor Lake CPU. The 24-core and 32-thread Raptor Lake processor scores 2290 points in single-core Cinebench R23 test and 35693 points in multi-core. It...

videocardz.com

There is no basis to claim that such an extreme OC mode exists.

So this page here: https://videocardz.com/newz/intel-c...ce-mode-on-some-intel-700-series-motherboards

Is another leaker saying the same thing. But you discount anything you do not like.

Now back to Zen 4 ......

 

[shady28]

Neither is Alder Lake rated for anything above DDR5-4800. Yet with the right kit and motherboard, it can do DDR5-6800, provided you won the silicon lottery.

I know that, as I said though, about 1 out of 5 Alder Lake apparently won't do DDR5-6000. Almost 100% Alder Lake can do DDR5-5600. 100% of Raptor Lake should do DDR5-5600 though, since it is rated for that speed. If it can't it's a defective RMA.

We have no idea what % of Zen 4 will do DDR5-6000 though.

Just as a reminder, how many of you Zen 3 owners are running DDR4-4000, hm?

 

[inf64]

It's not necessary to remove Crypto test. Because it can only give additional < 20% uplift to Crypto score when you use AVX512.

https://twitter.com/x/status/1439564626719506432

only 5% of this '<20%' add up to overall ST score when utilizing AVX512, finally +2% to the ST score. It's already a REMOVAL by Geekbench itself.

But the Crypto MT is strange to say the least.

The MT score for AES is indeed weird for all Ryzen 7000 parts, I wonder what is happening there. It's basically lowering the total MT score(avg). It could be memory related( BW limited?) but who knows at this point. We'll have to wait for final reviews and chips in the hands of users to figure it out.

 

[nicalandia]

All you do is post ComputerBase results to "validate" your claims. Do you even look at other sources for data?

It's true. Games don't necessarily scale with just speed, that is clearly the case with the 12900KS, except when the CPU uArch are design for such speed. For example Zen4 and Raptor Lake... Games will take advantage of their optimized designs(Optimized Memory Subsystems)

 

[tamz_msc]

So this page here: https://videocardz.com/newz/intel-c...ce-mode-on-some-intel-700-series-motherboards

Is another leaker saying the same thing. But you discount anything you do not like.

Now back to Zen 4 ......

Geez, do you read before you post? You and Det0x posted the exact same thing. I'd like to have a rational outlook based on evidence over rumor-mongering.

 

[shady28]

Retail 7950X vs Retail 13900K Geekbench Windows 11

Z690 Taichi vs ASUS System Product Name - Geekbench

browser.geekbench.com

View attachment 66820

That actually looks bad for the Zen 4 if you look at the sub-score comparison. All of its power is in AES. Should score very, very well in web benchmarks. That's not unimportant, it is important for many.

That helps explain the 1% cpu-z 1T IPC uplift and the 13% geekbench 1T uplift though.

Blue is Intel, grey is Zen 4.

 

[Det0x]

Just as a reminder, how many of you Zen 3 owners are running DDR4-4000, hm?

On single CCD's Zen3 you mean?
Many i would say.. I myself stopped at 4200 MT/s.

https://imgur.com/a/ajcu7Ay

Its with dual CCD's Zen3 most are stuck @ 3800MT/s..
I've had 5x 5950x cpus, they all did 3800MT/s easily when you know how to tweak Zen3. (correct voltages is key)

This is one of them @ 2000MT/s without whea errors

1 hour OCCT large dataset -> completed
2 hours y-cruncher torture test with all enabled -> completed
100% runmemtestpro -> completed
1 hour karhu ram test -> completed

Back to your cave

 

[Markfw]

Geez, do you read before you post? You and Det0x posted the exact same thing. I'd like to have a rational outlook based on evidence over rumor-mongering.

2 different leaks, and 2 different articles. BOTH cover the 350 watt. And the above post by shady28 is using the 350 watt 13900k.

Yes, we need to wait for real reviews of both but we have fairly good Zen 4 benches, but only extreme 13900k bench leaks.

 

[tamz_msc]

2 different leaks, and 2 different articles. BOTH cover the 350 watt. And the above post by shady28 is using the 350 watt 13900k.

Geekbench does not push CPUs to reach their max power consumption. And two articles mean nothing if they recycle the same leak.

Yes, we need to wait for real reviews of both but we have fairly good Zen 4 benches, but only extreme 13900k bench leaks.

Why not look at the leak directly?

https://twitter.com/x/status/1556288587985227777

 

[shady28]

On single CCD's Zen3 you mean?
Many i would say.. I myself stopped at 4200 MT/s.

https://imgur.com/a/ajcu7Ay

Its with dual CCD's Zen3 most are stuck @ 3800MT/s

I would bet there are more Alder Lakes able to do DDR5-6800 than there are Zen 3 that can do DDR4-4000.

But I think you already know that now don't you.

Maybe Intel should just present Raptor Lake running DDR5-7200, and say "This is the best, good luck!". Sort of like AMD does...

 

[TESKATLIPOKA]

Can anyone tell me how many points does 5950x score in CB R23 with TDP limited to 65W(88W PPT)? Thanks.

 

[nicalandia]

That actually looks bad for the Zen 4 if you look at the sub-score comparison.

Dude, stop it. Take the L with grace. Otherwise it looks like you are grasping at straws. Zen4 so far is turning out to be an impressive performance upgrade due to the combined IPC + Speed increase. We haven't see this type of performance boost since the big jump from Excavator to OG Zen1.

 

[tamz_msc]

Dude, stop it. Take the L with grace. Otherwise it looks like you are grasping at straws. Zen4 so far is turning out to be an impressive performance upgrade due to the combined IPC + Speed increase. We haven't see this type of performance boost since the big jump from Excavator to OG Zen1.

Against the 13900K, it loses in most of the single-threaded int workloads in Geekbench, while showing a strong performance in fp workloads. This has been the tradition for each generation of Zen. Zen 4 is no different.

 

[gdansk]

I'm more interested to see what 13900K vs 7950X look like when set to around 150W actual. Any higher than that and I suspect my fan will become louder.

I know people complain you're leaving performance on the floor but I'm OK with that until I can move my PC into another room

 

[Det0x]

Can anyone tell me how many points does 5950x score in CB R23 with TDP limited to 65W(88W PPT)? Thanks.

Without rebooting or anything:

 

[Zucker2k]

It's true. Games don't necessarily scale with just speed, that is clearly the case with the 12900KS, except when the CPU uArch are design for such speed. For example Zen4 and Raptor Lake... Games will take advantage of their optimized designs(Optimized Memory Subsystems)

How come people talk about game scaling as if the cpu alone is responsible for generating the fps? I believe the newer generation of gpus will reveal the real difference between a chip like the 5600x and 12900k.

 

[nicalandia]

Against the 13900K, it loses in most of the single-threaded int workloads in Geekbench, while showing a strong performance in fp workloads. This has been the tradition for each generation of Zen. Zen 4 is no different.

True and at ISO speed/Core and ISO Memory(DDR4 vs DDR4) Zen3 was a match in gaming to Alder Lake(12400K vs 5600X) due to better memory subsystem even if they were behind on IPC performance.

So my point has been, Look AMD has reach speed and Memory parity with Intel while their memory subsystem has been enhanced, this will translate to better gaming no doubt about it.

 

[tamz_msc]

Huh, so according to AMD's numbers the 7950X score @ 65 W TDP is greater than the 5950X score @ 105 W TDP. IIRC at default settings the 5950X scores ~26,000 pts.

EDIT: Then I wonder why TDP was increased to 170W, given that it doesn't scale much beyond 105 W according to AMD. I guess the power draw per core was targeted to increase per core clock speeds.