Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[TESKATLIPOKA]

It will see a huge performance in all sorts of games. Just because AAA is mainly GPU limited does not mean there are no games that benefit.

Stuff like Stellaris and other paradox grand strategy, turn based 4x stuff like Civ, SIM stuff like ACC, iRacing etc will all benefit from v-cache even with a weaker GPU.

This total nonsense that v-cache helps when you have a top of the line GPU needs to die, it is entire down to what games you play.

Do you have a link to such a test or review? I would like to see how much It helps on weaker GPUs, for example 6700XT.

 

[randomhero]

Do you have a link to such a test or review? I would like to see how much It helps on weaker GPUs, for example 6700XT.

I can give you anecdotal evidence, for what is worth. I have played for years paradox grand strategies like HOI. It's never GPU bottleneck, believe me. Game likes CPU. And game keeps lots of data in flight so more cache the better. Game sessions can last for months, so that is lots of in flight data to handle. I saw one review somewhere of Stelaris for 5800x3d, late game difference was like 80,90 % IIRC.

 

[Timorous]

Do you have a link to such a test or review? I would like to see how much It helps on weaker GPUs, for example 6700XT.

ACC 4K results show massive advantages for the x3d chips. Factorio runs on a potato. Paradox grand strategy games are all pretty easy to run GPU wise.
The other factor is that for most of these types of title the important metric is turn time or tic rate not FPS.

 

[TESKATLIPOKA]

I can give you anecdotal evidence, for what is worth. I have played for years paradox grand strategies like HOI. It's never GPU bottleneck, believe me. Game likes CPU. And game keeps lots of data in flight so more cache the better. Game sessions can last for months, so that is lots of in flight data to handle. I saw one review somewhere of Stelaris for 5800x3d, late game difference was like 80,90 % IIRC.

ACC 4K results show massive advantages for the x3d chips. Factorio runs on a potato. Paradox grand strategy games are all pretty easy to run GPU wise.
The other factor is that for most of these types of title the important metric is turn time or tic rate not FPS.

I wanted some gaming charts with a weaker GPU and different CPUs being used.

Paradox games are ridiculously CPU heavy, but they don't need many cores or strong GPU.
The latest Victoria 3 has GTX 1660 or RX 590 as recommended, that's ~1/2 of RX7600 at 1080p.
A for CPU i5-6600K or 2600X. One is a 4core and the other 6core.
Strix Halo would be an overkill for It, even 8 Zen5 + 20CU IGP would be enough for It, just put 64-128MB V-case on top of that.

@Timorous I own some paradox games, although not the latest. I know very well what a slow fest It can be.

 

[Timorous]

I wanted some gaming charts with a weaker GPU and different CPUs being used.

Paradox games are ridiculously CPU heavy, but they don't need many cores or strong GPU.
The latest Victoria 3 has GTX 1660 or RX 590 as recommended, that's ~1/2 of RX7600 at 1080p.
A for CPU i5-6600K or 2600X. One is a 4core and the other 6core.
Strix Halo would be an overkill for It, even 8 Zen5 + 20CU IGP would be enough for It, just put 64-128MB V-case on top of that.

No. Not at all. You can play the game on a weaker CPU but it takes a lot lot longer because the tic rate is slow. If you only have a few hours a week to play it can be a huge difference because you can fit more games in per year.

Another genre that benefits are ARPGs, path of exile runs on a pretty weak GPU but late game maps really tax the CPU. Again though because it is an online title and running the same map twice is not that easy when you factor in material costs and the affixes and suffixes you can add it never really gets benched.

Unfortunately there probably are not reviews or charts of what you are looking for because nobody tests it. It is always this received wisdom that there is no point in pairing a strong CPU with a weak GPU for games because the GPU will be the bottleneck, this is true for new AAA titles but gaming is a lot lot broader than that.

 

[RnR_au]

Yeah the X3D sku's are awesome for some of the Paradox games. I used to play Stellaris, so discussions like in the thread below is very interesting for me.

https://www.reddit.com/r/Stellaris/comments/12mcpm8

Perhaps this is one of the disadvantages of having settled for a laptop with an egpu for the light gaming sessions I do nowadays is that no laptop have X3D sku available

 

[biostud]

Yeah the X3D sku's are awesome for some of the Paradox games. I used to play Stellaris, so discussions like in the thread below is very interesting for me.

https://www.reddit.com/r/Stellaris/comments/12mcpm8

Perhaps this is one of the disadvantages of having settled for a laptop with an egpu for the light gaming sessions I do nowadays is that no laptop have X3D sku available

There is one...

ROG Strix SCAR 17 X3D (2023) | Gaming Laptops | ROG – Republic of Gamers | ROG Global

Experience top-notch gaming with 2023 ROG Strix SCAR 17 X3D. A powerful AMD Ryzen 9 CPU, NVIDIA RTX 4090 GPU, and high-resolution QHD display lead the victory charge!

rog.asus.com

They could launch a 7745HX3D if they wanted, but apparently there isn't a market for it.

 

[igor_kavinski]

They could launch a 7745HX3D if they wanted, but apparently there isn't a market for it.

There is but AMD hates money, I guess.

I really wish one of the Taiwanese giants would release the 7800X3D in laptop form factor with some kind of heavy dock that doubles as a fan/heatsink. So people can enjoy the V-cache when on the move with reduced speeds and lower power consumption but when they get back home, they plug it into the dock and get the full desktop experience. Even the laptop water cooling thing would work but it gets pretty expensive at that point. The laptop + dock would still be a LOT more portable than the typical hulk of a desktop+monitor.

 

[eek2121]

At least you can show girls how big your...laptop is

Gaming laptops don’t need to be big/heavy.

My Alienware laptop is certainly thicker than a Macbook, but weighs less than 4lbs, and that is with a 3070 and a Cezanne chip. Don’t ask me about the power adapter, however. No idea what they were thinking.

 

[TESKATLIPOKA]

There is but AMD hates money, I guess.

AMD loves money. That's the reason for It's existence -> to make money.
I think the better question is If they actually could supply enough of them.
Just look at how many Dragon Range laptops you can choose from and It will be clear that there is not big enough supply from AMD.

 

[igor_kavinski]

My Alienware laptop is certainly thicker than a Macbook

The white ones I would buy based on looks alone if I had the money. Currently debating really hard on getting a Lenovo Glacier White gaming laptop. It looks awesome, comes with 6800H but GPU is pretty anemic 3050. Price is a bit over $700. One more problem is that I already have about 8 older laptops (one of them with Comet Lake/3070). Need to sell some first.

 

[moinmoin]

Just look at how many Dragon Range laptops you can choose from and It will be clear that there is not big enough supply from AMD.

That's not supply, that's availability from the manufacturers. Looking at Geizhals for the German market, there are 25 Dragon Range laptop configurations listed. 22 of them are in stock right now. So to me it looks like supply is fine, but you want more varied offers from the manufacturers.

 

[TESKATLIPOKA]

That's not supply, that's availability from the manufacturers. Looking at Geizhals for the German market, there are 25 Dragon Range laptop configurations listed. 22 of them are in stock right now. So to me it looks like supply is fine, but you want more varied offers from the manufacturers.

I checked It out, used Germany and Austria. Link
End result is 7 models and 17 configurations:

Asus: 3 models and 9 configurations
ASUS ROG Zephyrus Duo 16
ASUS ROG Strix Scar 17
ASUS ROG Strix G17

Lenovo: 2 models and 6 configurations
Lenovo Legion 7 Pro
Lenovo Legion 5 Pro

Msi: 1 model and 1 configuration
MSI Alpha 17 C7VG

Dell: 1 model and 1 configuration
Dell Alienware m16 R1

Why do you think there is not a higher variety and more manufacturers? Isn't It because AMD can't supply enough chips for more models to be released?

 

[igor_kavinski]

Surprised that HP doesn't have any Dragon Range models. Seems they are content putting out Intel gaming laptops with crappy 45% NTSC displays.

 

[moinmoin]

Why do you think there is not a higher variety and more manufacturers? Isn't It because AMD can't supply enough chips for more models to be released?

It may well be supply related in this case, but in the past manufacturers repeatedly showed an odd resistance of offering AMD based configurations even where it would make perfect sense and when the respective chips were reportedly abundant. So I tend to err on manufacturers not seriously trying to offer a higher variety. After all if models are available and not supply constrained supply is available.

Btw. I used geizhals.de which oddly enough results in a couple more entries compared to the .at version.

 

[Geddagod]

Are the CPU chiplets in AMD strix halo different than their CPU chiplets in their regular desktop and server line?
I'm guessing they are...

 

[Frenetic Pony]

It may well be supply related in this case, but in the past manufacturers repeatedly showed an odd resistance of offering AMD based configurations even where it would make perfect sense and when the respective chips were reportedly abundant. So I tend to err on manufacturers not seriously trying to offer a higher variety. After all if models are available and not supply constrained supply is available.

Btw. I used geizhals.de which oddly enough results in a couple more entries compared to the .at version.

Right now it's definitely part supply, one manufacturer is even suing AMD over not delivering contracted amount of chips on time: https://videocardz.com/newz/gpd-acc...t-supplying-enough-ryzen-7-7840u-apus-on-time

That being said, Intel has a lot of "bullying" power over laptops. They lock manufacturers into long term contracts and are just generally highly anti competitive there, so that could definitely be a factor.

 

[coercitiv]

Right now it's definitely part supply, one manufacturer is even suing AMD over not delivering contracted amount of chips on time: https://videocardz.com/newz/gpd-acc...t-supplying-enough-ryzen-7-7840u-apus-on-time

Apparently they got their CPUs.

 

[TESKATLIPOKA]

Right now it's definitely part supply, one manufacturer is even suing AMD over not delivering contracted amount of chips on time: https://videocardz.com/newz/gpd-acc...t-supplying-enough-ryzen-7-7840u-apus-on-time

GPD can't sue AMD because they have no contract with them.
They can sue their upstream supplier at most, but as @coercitiv posted, there is no longer a reason for that.

That being said, Intel has a lot of "bullying" power over laptops. They lock manufacturers into long term contracts and are just generally highly anti competitive there, so that could definitely be a factor.

AMD can't supply enough chips, that's the problem.
Manufacturers can do nothing else but use Intel, even If It's worse, but at least It can supply enough chips.
Your bullying story is exaggerated. Even If they sign a long term contract It has to have some incentives for OEMs too like lower prices, but you probably think that Intel simply threatens them into signing It or no chips for them, right?

 

[JoeRambo]

An interesting point from CB24 characterization by Chips And Cheese cough my eye:

Cinebench 2024: Reviewing the Benchmark

Maxon’s Cinebench is a perennial benchmark favorite. It’s free, easy to run, and scales across as many cores as you can give it. Its $0 cost allows the internet to provide plenty of res…

chipsandcheese.com

I think the fact that Z4 currently lacks "L0" to do zero bubble branching, pretty much ensures that Zen5 will reintroduce such capability.
So not that big of a deal in that slide as i have assumed before in this thread.

 

[Joe NYC]

Yea.

No, DG300 is a new TO.

Interesting. According to Dylan, it's partially new, he says the GMI links are still there in the silicon:

AMD MI300 – Taming The Hype – AI Performance, Volume Ramp, Customers, Cost, IO, Networking, Software

Amazing engineering, but what of the path to market?

www.semianalysis.com

 

[adroc_thurston]

According to Dylan, it's partially new, he says the GMI links are still there in the silicon:

Same layout doesn't mean it's the same TO.

 

[Kemano]

Hey

You seem to be an "insider" guy with great knowledge from the inside. I hope you have the answer.

We have had Apple's self-designed performance cores for a while which uses a somewhat different architectural approach with its large L1 caches and wide pipeline. Tell me if I'm wrong but I believe this gives it an edge on both IPC and efficiency. Not aiming for high frequency allows to spend transistors elsewhere and pack transistirs tighther giving a smaller overall dia area.

We've seen with Zen4c how much area can be spared by limiting the frequency target. On top of transistor (heat) density, the mere ability to reach 5-6Ghz costs transistors and probably contraints the size of the caches. Latency (that is not independent of target frequency) and capacity are probaby mutually exclusive. On the other hand merely larger cache hardly has any benefit without the proper architecture that can make proper use of the larger cache. It's truly a design choice to which one builds the architecture around.

But it seems to me that Apple's (and probably Qualcomm's) approach (big L1, wide, low frequency) gives better power efficiency.

While I can understand that AMD and Intel are racing for the satisfaction of their customers, people always say that the main target markets for which they create their designs and from which their most income/profit come are server and notebook. Both server and notebook are environment in which power-efficiency is a very important aspect, therefore frequency is somewhat limited.
I've always wondered why neither Intel nor AMD embraced Apple's approach of letting high frequency target go and going wide with big caches. Designing such an architecture, they would leverage IPC and efficiency gain in both their main (server and notebook) markets without having to suffer from the lowered frequency. Desktop ST would suffer only if the IPC gains fail to make up for the frequency regression.

So I still wonder why?

• because a CPU with x86 ISA (due to variable instruction length) is incapable of going that wide? Or is it much harder to do it?
• because frequency can be leveraged by existing software, but a completely different CPU architecture would require new software optimizations, recompiling, etc, and without the necessary control over the software such a move would be suicide?
• because AMD and Intel invested much into the knowledge of reaching high frequency over the years and they are bound by that (using existing knowledge always seems rewarding, or they simply lack the knowledge of building wide architecture.)

 

[itsmydamnation]

So the problem is small reads and writes , lots of server workloads etc are small reads and writes. Big L1D have trade offs , apple trade this off by having a min page size of 16kb vs 4kb of most arm and x86.

Everyone is going wide because transistor performance scaling has died. So in 2024 almost everyone is going to have cores just as wide as each other. But only x64 will be hitting 5let alone 6ghz, but they also pay in die size and power.

But having the fastest and by quite a margin 1T will allow you to set a price far in advance of what the extra transistors cost you.

 

[moinmoin]

I've always wondered why neither Intel nor AMD embraced Apple's approach of letting high frequency target go and going wide with big caches.

First of all both Intel and AMD are going wide with one of the coming gens, so it's already happening.

Then there are a couple of things to keep in mind for why that didn't happen earlier in x86 PC space:

• Mobile as in mobile phones has even stricter efficiency requirement than both servers and mobile as in non-DTR laptops.
• One has to keep in mind that completely new designs take around 5 years from their inception to their market launch. So companies have to correctly project what rivals are planning to prepare technology with which they could then react in a shorter time than that.
• Intel wasn't prepared at all to face the new competition it faced in resurging AMD and its unprecedented partnership with TSMC which suddenly brought a world class mobile tech foundry into the x86 PC market (that was in 2019, so 4 years ago). It failed to project that kind of competition, and as a result for years its products and foundry output reflected this.
• AMD was still rebuilding its business, with Zen 1 through 3 planned while it was close to bankruptcy, and Zen 4 building upon Zen 3. So Zen 5 for quite some time was already known to be the first Zen gen which an economically healthy AMD gets to work on from start to finish.