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

[DisEnchantment]

Speculate at will

 

[FlameTail]

And this is a thread about Zen5 !

Ok ok. Chill.

I wanted to see the thoughts of the AMD people here regarding it.

If as rumoured it has 12 RDNA3 CUs (at a lower clock speed), the performance should be be pretty formidable.

 

[FlameTail]

As far as the higher than 5600 DDR5 speeds, I don't think any OEM uses it. Probably not worth the risk for them of reduced stability. But LPDDR5 is at 8533, moving on to 9600 with guaranteed stability...

LPDDR has to be soldered, and very close to the processor (?) at those speeds.

 

[TESKATLIPOKA]

Ok ok. Chill.

I wanted to see the thoughts of the AMD people here regarding it.

If as rumoured it has 12 RDNA3 CUs (at a lower clock speed), the performance should be be pretty formidable.

Then post It in the correct thread. I will happily talk about It.

 

[igor_kavinski]

But LPDDR5 is at 8533, moving on to 9600 with guaranteed stability...

I wouldn't mind LPDDR5 on mobos as long as they are reasonable with the pricing aspect of it.

16GB --- General use and office/home use by the average not very demanding user

24 to 48GB --- For gamers

64GB to 192GB --- Professional use

Price these mobos accordingly but not unreasonably (like max $1000 for a mobo with 192GB LPDDR5) and I could see this as a viable approach to satisfying the market demand.

 

[igor_kavinski]

LPDDR has to be soldered, and very close to the processor (?) at those speeds.

They can put it on the sides of the CPU socket and even on the backside of the mobo behind the CPU socket.

 

[Joe NYC]

LPDDR has to be soldered, and very close to the processor (?) at those speeds.

Well, yeah. So that would be a good reason to have a new platform - to accommodate LPDDR5 inside the package, have superior performance, superior stability and ultimately lower cost than dealing with exotic DIMMs.

 

[adroc_thurston]

I am not sure if there is any point of introducing a new DDR5 platform for Zen 6.

You can make a new platform every single gen.
You don't really need a reason per se.

It could make sense to have a new platform for Zen 6, that ditches external memory and moves to on package memory...

MOP is the most neverever thing for desktop possible.

moving on to 9600 with guaranteed stability...

That's POP-only for the foreseeable future.

They can put it on the sides of the CPU socket and even on the backside of the mobo behind the CPU socket.

Absolutely the last thing you should be doing on DT (or anywhere outside of 10W tablet chips).
Either way, LPCAMM says hello.

 

[Joe NYC]

I wouldn't mind LPDDR5 on mobos as long as they are reasonable with the pricing aspect of it.

16GB --- General use and office/home use by the average not very demanding user

24 to 48GB --- For gamers

64GB to 192GB --- Professional use

Price these mobos accordingly but not unreasonably (like max $1000 for a mobo with 192GB LPDDR5) and I could see this as a viable approach to satisfying the market demand.

Even better, inside CPU package, and then cheaper mobos.

Then, super cheap CPUs would be 8-16 GB
Mainstream tier 16-32 GB
High end 32-64 GB

(since a possibility of a new platform for Zen 6 and onward)

 

[Joe NYC]

You can make a new platform every single gen.
You don't really need a reason per se.

I think the problem with slow adoption of AM5 in 2022 is a good reason to not introduce new platform unless there is a strong reason for it.

MOP is the most neverever thing for desktop possible.

I was wondering if it might be possible to have a combination of the two. Have some memory channels on package, but still have a couple of channels for external memory.

Which would offer the ultimate flexibility from low cost fixed to higher cost extendable platform.

Most people what to have an option to upgrade their memory, but I would estimate that > 95% of users never do.

Which would actually make the cheap OEM system less of a garbage. If you get a low end OEM system, with 2x 8 GB sticks, to upgrade, you have to dump those 2 sticks.

If the base system had 16 GB from CPU package (for low end CPU), you could upgrade without losing anything.

 

[igor_kavinski]

High end 32-64 GB

On package creates the undesirable possibility of doing things the Apple way. $2500 Intel Extreme Edition Core 9 Ultra CPU with 64GB LPDDR5.

 

[adroc_thurston]

I think the problem with slow adoption of AM5 in 2022 is a good reason to not introduce new platform unless there is a strong reason for it.

That's not a platform issue but a DDR5 issue.
Solved itself, really.

I was wondering if it might be possible to have a combination of the two. Have some memory channels on package, but still have a couple of channels for external memory.

You are never, quite literally never ever ever ever getting tiered mem anywhere client.

 

[Timmah!]

I understand your pain. X670E is way, way overpriced for what it offers, especially when there are $150 mobos from Gigabyte/ASROCK that can run a 7950X without much issue.

Yes. But i had to pay 700 for Asus Hero, solely to have 2x PCIE 16x slots for 2 GPUs far enough from each other to be able host 2 new modern thick GPUs. There was no other board to provide the same at the time, bar Asus Extreme, which was even more expensive.

 

[Timmah!]

If It offers you more performance than 16C Zen5 then why not, but that price for 32C 7970x is not pretty at all.
I am also wondering If the difference between 16C Zen5 vs 24C 7960X will be big enough. Let's say that 16C Zen5 will really be 30-35% faster than 7950X, then this 7960X would be only 11-15% faster. In this case, I would probably forgo this extra performance and choose Zen5, which will be more powerful on anything below 16 threads and also cheaper.

This really depends on how much will it be faster. But those latest leaks did not exactly exude confidence, 2400 scores compared to 2200 for 7950x on CB24 MT, does not look like 30-35 percent at all.
One can only wish it was that case, but i am not holding my breath.

 

[Glo.]

I wouldn't mind LPDDR5 on mobos as long as they are reasonable with the pricing aspect of it.

16GB --- General use and office/home use by the average not very demanding user

24 to 48GB --- For gamers

64GB to 192GB --- Professional use

Price these mobos accordingly but not unreasonably (like max $1000 for a mobo with 192GB LPDDR5) and I could see this as a viable approach to satisfying the market demand.

Unified memory architectures will require much more memory than when you have split CPU and dGPU. 32 GB is bare minimum, when AI functions will be integrated to OSes - 64 GB is minimum.

Virtualization, Ray Tracing, AR, spacial computing - all of this will require stupid amounts of RAM.

Which is why everybody integrates CPUs and GPUs together, because that way they will make them obsolete much faster, if they will integrate not enough memory on package.

Think about this: MacOS on 8 GB MacBook Pro M3 uses after boot 5.39 GB, alone. Apple designed this way perfect planned obsolescence, and milking their customers, at the same time.

About the pricing. I laugh that people start to talk finally about buying MoBos with integrated RAM, when I was scoffed by this forum when I was writing that this will become reality very soon.

And no, if you will get soldered desktop products - you will get a Mobo with CPU soldered into it, and if package comes with RAM, you will get that as well, perfect planned obsolescence products.

Knowing how OEMs think - they will want to have similar price margin structures to Apple, with hefty price premiums on higher capacity configs.

 

[igor_kavinski]

Knowing how OEMs think

It's their business guys. We wipe them out, we make humanity more human.

Humanity could certainly do with less sleazeballs. The fewer the better.

I have no issue paying a premium for something that doesn't have a lot of volume. What I have a problem with is they price it three times higher than it needs to be and then laugh all the way to the bank. That's evil.

 

[Ajay]

You mean stop posting / speculating about Strix Halo for a year, until it ships?

You can't be serious.

No, that wasn’t it. Sh*t posting about an already doomed future product because *reasons*. Otherwise, please carry on.

 

[moinmoin]

Is MP1 aarch64/arm based as well?

FYI Cortex-A5 is the smallest 32bit Armv7 core available, originally intended to replace old ARM9 and ARM11 cores (so really low power microcontroller level cores).

And there appears to be no confirmation that MP1 (SMU) is ARM based like MP0 (PSP). But PSP is required for chain of trust in the platform initialization so I can imagine the SMUs to be ARM based as well, possibly even using the same physical core. One should be able to get confirmation of that either way by digging into any AGESA (or OpenSIL) firmware blob, but a quick search didn't bring up anything for me.

 

[FlameTail]

Is LPDDR more expensive than GDDR?

 

[adroc_thurston]

Is LPDDR more expensive than GDDR?

no.
You can always look up spot pricing on DRAMeXchange.

 

[FlameTail]

Well that's be good news for APUs.

You get higher VRAM quantity for the same price.

I wonder why big APUs like the upcoming Strix Halo haven't taken off yet. Did we have no idea until Apple showed us how to do it with their SoCs with massive iGPUs?

Instead of paying for two separate components (CPU+dGPU), Laptop OEMs will have to pay for only one component (big APU). This would give cost savings as well space savings in the laptop's internals.

 

[adroc_thurston]

Well that's be good news for APUs.

Kinda.

You get higher VRAM quantity for the same price.

Yeah, sexo, now AMD please support proviz stuff better (it's already OK but could be better).

I wonder why big APUs like the upcoming Strix Halo haven't taken off yet

It offers significantly less design flexibility than the usual CPU + dGP combo.

Did we have no idea until Apple showed us how to do it with their SoCs with massive iGPUs?

Intel shipped whole 4 gens of extended iGP parts (Haswell/Broadwell/Skylake/Coffee Lake GT3e, SKL even had a GT4e).
Those quite literally had zero market traction outside of Apple designs.
oh and Kaby Lake-G existed.

Instead of paying for two separate components (CPU+dGPU), Laptop OEMs will have to pay for only one component (big APU). This would give cost savings as well space savings in the laptop's internals.

Yeah you're basically retelling me STX-halo product pitch.

 

[igor_kavinski]

Intel shipped whole 4 gens of extended iGP parts (Haswell/Broadwell/Skylake/Coffee Lake GT3e, SKL even had a GT4e).
Those quite literally had zero market traction outside of Apple designs.
oh and Kaby Lake-G existed.

They never had a chance of gaining market traction coz they were not released in desktop PCs. The only serious iGPU that Intel ever released on desktop was in Core i7-5775C and even then their heart wasn't in it. No follow-up eDRAM product on desktop after that. Seems they couldn't figure out how to do eDRAM at high enough yields for it to be a mass market product.

 

[adroc_thurston]

They never had a chance of gaining market traction coz they were not released in desktop PCs

Bro big APUs will never, ever, quite literally not even in your dreams be on desktop.
Also, those all were in NUCs.

Seems they couldn't figure out how to do eDRAM at high enough yields for it to be a mass market product.

Wasn't a yield issue at all.

 

[igor_kavinski]

Bro big APUs will never, ever, quite literally not even in your dreams be on desktop.

So ARL also getting a gimped iGPU? No 720p Cyberpunk 2077 with raytracing for Intel iGPU peasants?

 

[adroc_thurston]

So ARL also getting a gimped iGPU?

Desktop?
Obviously lmao.

No 720p Cyberpunk 2077 with raytracing for Intel iGPU peasants?

Yeah use discrete graphics.