Info Ryzen 4000 Mobile Chips Unveiled at CES

[guachi]

Acer Swift 3 with Ryzen 7 4700U
AMD Ryzen 4000 Mobile APUs

AMD debuted (finally!) their Ryzen 4000 mobile chips at CES that are confusingly Ryzen 2 CPUs that are analogous with their Ryzen 3000 desktop parts.

I like what I see, especially on the power front. I know many people said that 7nm Ryzen chips had the potential to be very power efficient and if AMD's slide deck is to be believed they have succeeded. Most of the power efficiency has come from the 7nm process. As well, the 7nm process looks to be allowing AMD to cram up to 8 cores onto a laptop chip.

Do you guys think AMD has a product that will be as competitive on the laptop as the 3000 series is on the desktop? I'm thinking that the 4600 will be the best buy like the 3600 is in the desktop space. The problem in the laptop space is AMD needs design wins. At least on the desktop I don't need some company to choose for me, I can just by the chip myself.

 

[kapulek]

Ryzen 9 4900u is probably going to be Van Gogh.

Why limit RDNA2 APU to just one SKU?

 

[NostaSeronx]

Why limit RDNA2 APU to just one SKU?

My theory is Renoir won't be the lead desktop chip, Van Gogh will be the one.

So, Ryzen 4200G and up will probably be Van Gogh rather than Renoir.
4200G = VGH B6
4400G = VGH B8
4600G = VGH B10
4800G = VGH B12 => Ryzen 9 4900u being a salvage of this. (Also, introduces post-FP6 socket for Cezanne)

One mobile SKU w/ 8c/16t and with a salvaged 8CU/4WGP iGPU in 15W. Much like the 3900 series, there is probably room for a 4900+ numbers in U-market.

 

[Shivansps]

My theory is Renoir won't be the lead desktop chip, Van Gogh will be the one.

So, Ryzen 4200G and up will probably be Van Gogh rather than Renoir.
4200G = VGH B6
4400G = VGH B8
4600G = VGH B10
4800G = VGH B12 => Ryzen 9 4900u being a salvage of this. (Also, introduces post-FP6 socket for Cezanne)

One mobile SKU w/ 8c/16t and with a salvaged 8CU/4WGP iGPU in 15W. Much like the 3900 series, there is probably room for a 4900+ numbers.

While it does make sence for desktop, specially after that 4500U review with lower game perf than a stock 2200G, i doubt it.

 

[Gideon]

First geekbench results for the 4800HS (35W TDP cpu), compared to my Desktop 3700X. Both run nearly at the same frequency (desktop aboot 100-150 Mhz faster), but the 4x larger caches obvioulsy help the desktop counterpart. Very close to the Core i9-9880H. MT score is probably also hindered by the TDP but also the cache hierarchy (4MB of L3 per 4 cores)

https://browser.geekbench.com/v5/cpu/compare/531276?baseline=1198171

 

[Zstream]

First geekbench results for the 4800HS (35W TDP cpu), compared to my Desktop 3700X. Both run nearly at the same frequency (desktop aboot 100-150 Mhz faster), but the 4x larger caches obvioulsy help the desktop counterpart. Very close to the Core i9-9880H. MT score is probably also hindered by the TDP but also the cache hierarchy (4MB of L3 per 4 cores)

https://browser.geekbench.com/v5/cpu/compare/531276?baseline=1198171

No one really cares about insane multi-core results on a laptop, as long as it is decent that’s a positive. It’s the single threaded result which is awesome. Finally a real AMD desktop class performer in a laptop, with good battery life as well.

 

[awesomedeluxe]

First geekbench results for the 4800HS (35W TDP cpu), compared to my Desktop 3700X. Both run nearly at the same frequency (desktop aboot 100-150 Mhz faster), but the 4x larger caches obvioulsy help the desktop counterpart. Very close to the Core i9-9880H. MT score is probably also hindered by the TDP but also the cache hierarchy (4MB of L3 per 4 cores)

https://browser.geekbench.com/v5/cpu/compare/531276?baseline=1198171

Do we have any idea what the differences are between this and the stock 4800H part? Any speculation on the trick ASUS is using here?

 

[guachi]

No one really cares about insane multi-core results on a laptop, as long as it is decent that’s a positive. It’s the single threaded result which is awesome. Finally a real AMD desktop class performer in a laptop, with good battery life as well.

The ability to do actual productivity work on the go in a laptop is major plus for many people. Good single core and good multi core is a major plus. I wonder if AMD will christen a HELT segment.

 

[Hitman928]

Do we have any idea what the differences are between this and the stock 4800H part? Any speculation on the trick ASUS is using here?

The 4800HS is 35 W, the 4800H is 45 W so the 4800H should perform better unless thermally constrained. With only pre-release leaks to go off of, it's hard to truly compare at this stage without knowing the cooling solutions, power plans, memory config, etc. that each unit is using to say why one is performing better or what the final performance will be.

 

[RetroZombie]

If amd would have been more ambitious with renoir i would see these mobile chips with just 35W beating all the desktop cpus that intel released two years ago, and making the three years old ones look like jokes (7700K for example).

 

[awesomedeluxe]

If amd would have been more ambitious with renoir i would see these mobile chips with just 35W beating all the desktop cpus that intel released two years ago, and making the three years old ones look like jokes (7700K for example).

I think they were as ambitious as they could afford to be. Yield on TSMC 7nm is not amazing so they were probably constrained by the size of the die. It sucks they were not able to fit more cache though.

 

[RetroZombie]

It sucks they were not able to fit more cache though.

Yeah i totally agree.
At least we will see how much cache dependent zen2 is, and if the 32mb was not just used has a space filler/increase.

 

[uzzi38]

I think they were as ambitious as they could afford to be. Yield on TSMC 7nm is not amazing so they were probably constrained by the size of the die. It sucks they were not able to fit more cache though.

Yield on N7 is perfectly fine, it's very nearly as good as N16.

Wafer cost however... not so much.

Not to mention, AMD want to be using as little wafer space as they can to maximise volumes. Mobile isn't a very performance focused market, so cutting down the cache there is relatively acceptable when compared to desktop/servers.

 

[HurleyBird]

I think they were as ambitious as they could afford to be. Yield on TSMC 7nm is not amazing so they were probably constrained by the size of the die. It sucks they were not able to fit more cache though.

Theoretically, they could have been ambitious in the vein of making an eight-core CCX which would effectively double the L3 size. Sort of a half way step between Zen 2 and Zen 3. That sort of makes sense if you look at Renoir coming in not too far off from Zen 3. And in that context so would a Navi-based GPU. But I think Renoir took AMD a fair bit longer to finish than they had first planned for.

 

[beginner99]

f amd would have been more ambitious with renoir

Actually in my opinion renoir is the first smart APU AMD has made. The ones before usually were too GPU heavy and especially the pre Zen ones lacked on the CPU side. Renoir is ideal for high end laptop. Small enough but a lot of CPU power and just enough GPU. If you want to game or do other GPU heavy stuff, you will need to dGPU anyway. Making the iGPU beefier would simply be a waste of die space for >90% of buyers. Look at intels shortages. IMHO they put way too much die space into iGPU. Problem with intel is, they need it because their uArch isn't efficient but die-space efficency will become more and more important.
Why are Apple products so expensive? Because they get their performance/watt at the cost of die space. There is no free lunch. Since intel and amd don't do products just for the luxury market, die space efficency will be a deciding factor in the years to come.

 

[lobz]

Do we have any idea what the differences are between this and the stock 4800H part? Any speculation on the trick ASUS is using here?

binned for 35W instead of 45W

 

[Topweasel]

binned for 35W instead of 45W

Well specifically its sponsored bin. Asus has temp exclusivity on binned 4800H's that supposedly hit all the 4800H's performance metrics at 35w. Whether that holds or we see some performance discrepancy when both are at limits remains to be seen. Considering that actual TDP limits are broken for limited turbos and such. Its probably not that hard get cherry dies for the HS.

 

[awesomedeluxe]

Yield on N7 is perfectly fine, it's very nearly as good as N16.

Wafer cost however... not so much.

Not to mention, AMD want to be using as little wafer space as they can to maximise volumes. Mobile isn't a very performance focused market, so cutting down the cache there is relatively acceptable when compared to desktop/servers.

Good point about the volume. Zen+ was on GloFo which was a weird arrangement. My memory is that AMD had forced them into a very favorable agreement where GloFo was actually reimbursing AMD for the bad dies to repay AMD for missing prior yield targets and deadlines. So using a Raven-Ridge size die would still be a lot less feasible after the breakup with GF.

Theoretically, they could have been ambitious in the vein of making an eight-core CCX which would effectively double the L3 size. Sort of a half way step between Zen 2 and Zen 3. That sort of makes sense if you look at Renoir coming in not too far off from Zen 3. And in that context so would a Navi-based GPU. But I think Renoir took AMD a fair bit longer to finish than they had first planned for.

Not too knowledgeable about where AMD is w/r to Zen 3, but what you're suggesting about memory would make a lot of sense in AMD's next APU (which--pure speculation--I suspect will be Zen 2 + RDNA2 derived from the PS5/Xbox). However, I don't think incorporating Navi was actually possible. Navi has tons of issues. They were able to squeeze a ton of extra performance out of Vega moving to 7nm, and I honestly think it's better than what they would have been able to get Navi to do in an APU at that point in time.

 

[maddie]

Good point about the volume. Zen+ was on GloFo which was a weird arrangement. My memory is that AMD had forced them into a very favorable agreement where GloFo was actually reimbursing AMD for the bad dies to repay AMD for missing prior yield targets and deadlines. So using a Raven-Ridge size die would still be a lot less feasible after the breakup with GF.



Not too knowledgeable about where AMD is w/r to Zen 3, but what you're suggesting about memory would make a lot of sense in AMD's next APU (which--pure speculation--I suspect will be Zen 2 + RDNA2 derived from the PS5/Xbox). However, I don't think incorporating Navi was actually possible. Navi has tons of issues. They were able to squeeze a ton of extra performance out of Vega moving to 7nm, and I honestly think it's better than what they would have been able to get Navi to do in an APU at that point in time.

On the other hand, AMD had to purchase wafers from them or pay a penalty if using another fab. In such a situation, I could see AMD saying, we only better pay for good die or in the extreme, we might end up stuck buying wafers with 100% defective die.

 

[RetroZombie]

On the other hand, AMD had to purchase wafers from them or pay a penalty if using another fab. In such a situation, I could see AMD saying, we only better pay for good die or in the extreme, we might end up stuck buying wafers with 100% defective die.

I think this will have to work both ways.

I could also design a mediocre chip, sent to manufacture and the performance is extremely bad or yield very poorly. Should the manufacturer be the blame here?

Just look at intel 10nm right now, mediocre foundries or mediocre designs? Or both?

 

[RetroZombie]

But I think Renoir took AMD a fair bit longer to finish than they had first planned for.

Vega 7nm was finished in June 2018.
Zen 2 was finished around april 2019.

Validating and gluing both takes it's time i would guess zen 2 was the one that delayed everything or navi really screwed their plans.

 

[moinmoin]

Not to mention, AMD want to be using as little wafer space as they can to maximise volumes. Mobile isn't a very performance focused market, so cutting down the cache there is relatively acceptable when compared to desktop/servers.

Cutting down the L3 cache also doubles as increasing power efficiency on mobile since less cache needs to be powered. And L3 cache is notoriously hard to power gate since it's shared between all cores, so while a core is power gated its own L3$ slice still can't be (unless it's known not to lose data that way).

Theoretically, they could have been ambitious in the vein of making an eight-core CCX which would effectively double the L3 size. Sort of a half way step between Zen 2 and Zen 3.

Can we already rule out that's already happening?
My guess so far is that much of the unified L3 cache per CCD is achieved through improved cache handling through the cache controller, which I assume resides on IOD currently but could be partly moved onto the CCD in Zen 3. In Renoir's case the die is a monolith anyway so even if the parts are Zen 2 the firmware could still use Zen 3 logic.

 

[uzzi38]

My theory is Renoir won't be the lead desktop chip, Van Gogh will be the one.

So, Ryzen 4200G and up will probably be Van Gogh rather than Renoir.
4200G = VGH B6
4400G = VGH B8
4600G = VGH B10
4800G = VGH B12 => Ryzen 9 4900u being a salvage of this. (Also, introduces post-FP6 socket for Cezanne)

One mobile SKU w/ 8c/16t and with a salvaged 8CU/4WGP iGPU in 15W. Much like the 3900 series, there is probably room for a 4900+ numbers in U-market.

Desktop APUs are not premium products.

Van Gogh is an extremely premium product.

It's very doubtful anything more than Renoir will hit the desktop if you ask me.

 

[awesomedeluxe]

Desktop APUs are not premium products.

Van Gogh is an extremely premium product.

It's very doubtful anything more than Renoir will hit the desktop if you ask me.

Do we know this for sure?

I mean, as far as I'm aware, Van Gogh could still be anything. There some pointers suggesting RDNA2, and it would certainly make sense given that AMD should be simultaneously finalizing work on RDNA2 APUs for the PS5/Xbox. But that's really it, right? Is there something I'm missing?

 

[joesiv]

Do we know this for sure?

I mean, as far as I'm aware, Van Gogh could still be anything. There some pointers suggesting RDNA2, and it would certainly make sense given that AMD should be simultaneously finalizing work on RDNA2 APUs for the PS5/Xbox. But that's really it, right? Is there something I'm missing?

I guess AMD could release a higher end APU on AM4, I'd welcome it. I think if we look historically APU's were always lower end though.

I wonder what they would price an 8c16t APU like Van Gogh at. it'd have to be more than a 3700x right? So.. $350, $400? more? At that price are people willing to buy that APU versus buying a 3700x + dedicated GPU, where they can upgrade the GPU later, and recoupe the old GPU by selling it? I'm not sure. It'd be an interesting experiment.

 

[moinmoin]

Do we know this for sure?

I mean, as far as I'm aware, Van Gogh could still be anything. There some pointers suggesting RDNA2, and it would certainly make sense given that AMD should be simultaneously finalizing work on RDNA2 APUs for the PS5/Xbox. But that's really it, right? Is there something I'm missing?

The work on PS5/Xbox is semi custom, the specs for Van Gogh look semi custom as well. Semi custom R&D doesn't usually feed back into consumer products by AMD right away. And AMD is just rolling out Renoir as their APUs for this round, it wouldn't make sense to put duplicitous effort on a consumer Van Gogh at the same time. So chances are high it's another console alike semi custom work.

I guess AMD could release a higher end APU on AM4, I'd welcome it. I think if we look historically APU's were always lower end though.

Keep in mind that the limiting factor is AM4's memory bandwidth anyway. DDR5 should change that (on AM5 or some such). But before that any more powerful iGPU than in 2400G/3400G/Renoir on AM4 is a pipe dream.