Speculation: Ryzen 4000 series/Zen 3

[Panino Manino]

Cezanne using Vega makes me think that AMD underestimated the new Intel GPU.
I know that Intel is exaggerating with the comparisons, but still, the advantage AMD used to have practically disappeared.

 

[inf64]

Is 15+% IPC a bad thing? Zen3 will rock, Rocketlake will be DOA.

 

[NostaSeronx]

Cezanne using Vega makes me think that AMD underestimated the new Intel GPU.

The use of Vega is because of end cost.

Standard APUs:
Llano = 228 mm2
Trinity = 246 mm2
Kaveri = 245 mm2
Carrizo = 244.62 mm2
Raven = 209.78 mm2

Budget APUs:
Stoney = 125 mm2
Raven2 = 150 mm2
Renoir = 156 mm2
Cezanne = similar area

Ultra-budget APUs:
Ontairo: 75 mm2
Kabini: 97 mm2 ~ 105 mm2

The competitor to Intel offerings aren't out yet, won't be till Xbox Series S gets announced. (X1S => Cato(a9-9820/rx-8125/rx-8120) / XSS => Van Gogh(???)) Of which, that processor is succeeded by Rembrandt, same spec as XSS but uses Family 19h core.

Cezanne remains on FP6/AM4.
Rembrandt goes with Van Gogh FP7/FM5.

 

[exquisitechar]

Cezanne using Vega makes me think that AMD underestimated the new Intel GPU.
I know that Intel is exaggerating with the comparisons, but still, the advantage AMD used to have practically disappeared.

They didn’t underestimate it, they just don’t care. Such a difference in iGPU performance is irrelevant. Cezanne is using Vega to minimize TTM, to have the best performing CPU core in laptops again as soon as possible - something much more important.

 

[maddie]

The use of Vega is because of end cost.

Standard APUs:
Llano = 228 mm2
Trinity = 246 mm2
Kaveri = 245 mm2
Carrizo = 244.62 mm2
Raven = 209.78 mm2

Budget APUs:
Stoney = 125 mm2
Raven2 = 150 mm2
Renoir = 156 mm2
Cezanne = similar area

Ultra-budget APUs:
Ontairo: 75 mm2
Kabini: 97 mm2 ~ 105 mm2

The competitor to Intel offerings aren't out yet, won't be till Xbox Series S gets announced. (X1S => Cato(a9-9820/rx-8125/rx-8120) / XSS => Van Gogh(???)) Of which, that processor is succeeded by Rembrandt, same spec as XSS but uses Family 19h core.

Cezanne remains on FP6/AM4.
Rembrandt goes with Van Gogh FP7/FM5.

Renoir is a budget APU? You have outdone yourself this time.

 

[eek2121]

Those emojis better than +15% higher IPC and 20% higher all-core boost clocks. Or better!

🤔🖕😉

Cezanne using Vega makes me think that AMD underestimated the new Intel GPU.
I know that Intel is exaggerating with the comparisons, but still, the advantage AMD used to have practically disappeared.

Vega has undergone a ton of refinement to the point where it is extremely power efficient. The GPU does not need to win benchmarks, it just needs to be in the game.

 

[NostaSeronx]

Renoir is a budget APU? You have outdone yourself this time.

150 mm2 is the budget area.
14nm ~1.0x => ~$50 is recommend MSRP
7nm ~2.0x => ~$100 is the price hike to get same margins.

14nm is stuck at a single phase at GloFo's F8.
7nm has three phases to pick from at TSMC's F15.

Implication set is that Renoir per wafer is overall cheaper than Raven2 per wafer.

Ryzen 7 2700X ($329 at launch) -> Ryzen 7 3700X ($329 at launch) -> Ryzen 7 47xxG ($309 at launch) ~= means huge margins for AMD relative to Raven2. Not bad for a budget APU.

Essentially, Renoir is in the FX-9590/FX-9370 $900 USD phase as time goes by it will reach that $390/350 USD phase. Becoming the all desired budget APU that scoffs at the i7-1185G7's $426 price tag.

 

[Panino Manino]

They didn’t underestimate it, they just don’t care. Such a difference in iGPU performance is irrelevant. Cezanne is using Vega to minimize TTM, to have the best performing CPU core in laptops again as soon as possible - something much more important.

If, Zen 3 proves to be as good as "expected".

 

[coercitiv]

If, Zen 3 proves to be as good as "expected".

Zen 3 will deliver exactly as expected. Based on the engineering sample IDs published by Igor's Lab and also the OEM slides that HardwareLuxx got their hands on we know Zen 3 clocks up to 4.8Ghz and comes with +15% IPC in integer workloads. Performance jump will likely be in the realm of 15-20% depending on SKU. And then comes the unified cache which is bound to have a positive effect on gaming. It's a solid upgrade.

Cezanne will bring gains exactly where Renoir needs them: ST performance and dGPU gaming. The sooner the better.

 

[turtile]

They didn’t underestimate it, they just don’t care. Such a difference in iGPU performance is irrelevant. Cezanne is using Vega to minimize TTM, to have the best performing CPU core in laptops again as soon as possible - something much more important.

Yes, AMD wanted to make the chip cost attractive to get design wins. I think the only thing they underestimated was the demand for the chips. I think they would have put more effort into them if they thought it would pay off so well.

 

[jamescox]

Zen 3 will deliver exactly as expected. Based on the engineering sample IDs published by Igor's Lab and also the OEM slides that HardwareLuxx got their hands on we know Zen 3 clocks up to 4.8Ghz and comes with +15% IPC in integer workloads. Performance jump will likely be in the realm of 15-20% depending on SKU. And then comes the unified cache which is bound to have a positive effect on gaming. It's a solid upgrade.

Cezanne will bring gains exactly where Renoir needs them: ST performance and dGPU gaming. The sooner the better.

I don’t know all of the code names at this point, so I may be confusing some things. I could use both a new desktop and a new laptop at the moment. I was going to try to hold out for Zen 3 based devices. I hope that they get the Zen 3 based mobile chips out sooner rather than later; I might not bother with a desktop if I can get Zen 3 in a laptop. It would be nice if their whole product stack was the same architecture rather than the spit between last generation mobile parts and the latest generation desktop parts under the same name. This hasn’t been that uncommon in AMD GPUs and such though. They often had the lower end filled out with previous generation derivatives.

I don’t know all of the part names, but I noticed that there is supposed to be a Van Gogh part for ultra low power that is actually Zen 2 based rather than Zen 3. That makes sense since Zen 3, with its larger L3 cache, may not scale down to low power as well as Zen 2. It is actually the same amount of cache for the desktop part, just 1x32MB rather than 2x16MB. The Zen 3 part will have a lot more interconnect though and that will burn more power. It would be great if they could get the Zen 3 desktop part, the Zen 3 mobile APU, and the Zen 2 based ultra low power part (Van Gogh) out in a similar time frame. It would be even better if Van Gogh is 5 nm. I don’t know if they have the resources to get them all out at once, but they have a lot more R&D money than they did previously and they have a lot of design reuse across many of the products.

That would make for some confusing naming conventions (nothing new) and possibly interesting choices. It would also probably hit intel mobile market share even harder than Zen 2 already is.

 

[jamescox]

🤔🖕😉

Vega has undergone a ton of refinement to the point where it is extremely power efficient. The GPU does not need to win benchmarks, it just needs to be in the game.

I was under the impression that even the older Vega based GPUs did extremely well in power consumption when under-volted.

 

[NostaSeronx]

It will be really awkward when 15FF(Vega1x/GDDR5) = 163F(Navi2x/GDDR6).

15FF (5xh) ~ 163F
15DD (10/11h) ~ 1636
15D8 (18h) ~ 1638
15D9 (2xh) ~ 164x

 

[Vattila]

These latency differences [of 6 clock cycles between near and far L3 slice, according to NostaSeronx's posted latency measurements] correspond well to the actual distance differences for a fully connected topology.

Although this is not my field of expertise, I very much doubt these differences are solely due to interconnect wire delay, as you suggest. Intuitively, from the little I know about this, it seems to me that a wire delay of 6 clock cycles is excessive. Assuming that "wires have an approximate propagation delay of 1 ns for every 6 inches (15 cm) of length" (Wikipedia), which equates to about half the speed of light, then signals travel 30 mm/cycle at 5 GHz. Hence, a signal can travel more than 180 mm in 6 cycles! The whole Zen L3 is 16 mm², about 2 to 1 rectangular; so around 6.3 mm diagonal, and 8.5 mm between opposite corners along the periphery.

Even if the actual L3 interconnect wire delay is twice that Wikipedia quote, the difference in wire lengths from any L2 to any L3 controller should not require extra cycles at all — the worst-case wire delay within the L3 should still be well within a single clock cycle (>15 mm).

PS. By the way, for those interested, here is a PhD thesis I found while reading up on interconnect delay: "Efficient High-Speed On-Chip Global Interconnects", Peter Caputa, 2006. It has a nice introduction to high-speed on-chip interconnect, and proposes an upper metal layer interconnect approaching lightspeed.

 

[leoneazzurro]

I was under the impression that even the older Vega based GPUs did extremely well in power consumption when under-volted.

Yes, it could reach fair values in power consumption but not at the clock targets we are seeing in Renoir. We are seeing now 8 CU @1,7GHz with power consumption well under 15W (reference: 4800U) while with Vega20 (still on 7nm but with a different process stepping) we had 60 CU at around the same clocks @295W (board power, but that anyway gives for 8 CU a net value around 25-30W). This is in the range of 3-5x improvement of power consumption figures per CU, at same clocks, on similar geometry. Nothing short of impressive.

 

[Nereus77]

Zen 3 will deliver exactly as expected. Based on the engineering sample IDs published by Igor's Lab and also the OEM slides that HardwareLuxx got their hands on we know Zen 3 clocks up to 4.8Ghz and comes with +15% IPC in integer workloads. Performance jump will likely be in the realm of 15-20% depending on SKU. And then comes the unified cache which is bound to have a positive effect on gaming. It's a solid upgrade.

Cezanne will bring gains exactly where Renoir needs them: ST performance and dGPU gaming. The sooner the better.

15-20% improvements is a great jump between generations, a worthy upgrade, particularly if you're jumping from Zen 1. I'd imagine that the only space Intel would win would be with AVX512 and AI improvements.

Will TigerLake only be a really big deal in the time between its launch and Zen 3's? If Zen 3 slaughters it (which it might) and AMD gains even more marketshare, Intel will be in some real trouble. A lot would rest on the shoulders of Alder Lake.

 

[A///]

Tigerlake mobile isn't a huge threat. Intel's benchmarks were a joke. The AI is cool stuff, but it's early days, and that's also integration of a product they bought, not something they developed. Naveen Rao, one of the former founders of a startup Intel bought, left the company a few weeks ago. Based on his curt statement it seems like he wasn't liking what Intel were doing or going with.

I've considered doing a write up on those benchmarks but I'm a nobody and no one will bother to read it.

 

[CakeMonster]

I guess the target for a lot of people is whether it rivals Intel at 5GHz per core. That seems the average OC goal from SL to current cores. If it matches or exceeds that, a lot of people will be tempted to get more cores without downgrading performance per thread.

 

[Nereus77]

Tigerlake mobile isn't a huge threat. Intel's benchmarks were a joke. The AI is cool stuff, but it's early days, and that's also integration of a product they bought, not something they developed. Naveen Rao, one of the former founders of a startup Intel bought, left the company a few weeks ago. Based on his curt statement it seems like he wasn't liking what Intel were doing or going with.

I've considered doing a write up on those benchmarks but I'm a nobody and no one will bother to read it.

Both AI and AVX512 are very niche, so.... Sounds like we may be witnessing a Bulldozer oopsie from Intel (far too early to tell though, perhaps in retrospect we can see). Intel is probably doing everything they can to prevent a Bulldozer moment, meaning that 2021 is shaping up to be a very interesting year for computing.


Hey man, I'm sure you'd have a few readers from here!

 

[jpiniero]

Tigerlake mobile isn't a huge threat. Intel's benchmarks were a joke. The AI is cool stuff, but it's early days, and that's also integration of a product they bought, not something they developed. Naveen Rao, one of the former founders of a startup Intel bought, left the company a few weeks ago. Based on his curt statement it seems like he wasn't liking what Intel were doing or going with.

They did spend $2B on Habana which basically made Nervana redundant.

 

[A///]

They did spend $2B on Habana which basically made Nervana redundant.

It wasn't the only two AI companies they bought. I'm not sure if Intel knows what they're doing anymore.

 

[A///]

Both AI and AVX512 are very niche, so.... Sounds like we may be witnessing a Bulldozer oopsie from Intel (far too early to tell though, perhaps in retrospect we can see). Intel is probably doing everything they can to prevent a Bulldozer moment, meaning that 2021 is shaping up to be a very interesting year for computing.


Hey man, I'm sure you'd have a few readers from here!

Possibly. Intel increased their purchase rate in specific fields of computing in the last 5 years. The AI isn't BS, it does help, but only to an extent. Without the AI in TGL, then their performance would be neutered. Intel speaks volumes without a hint of a word that their 10nm is still fairly troubled because they choose to park 4 cores on an CPU with an iGPU based on XE that's far larger.

Eh, I'll pass. I'm a very boring person. I fake the enthusiasm I have on here just so people don't think I'm an old miserable git.

 

[moonbogg]

I'm looking to get a PCI-E 4.0 board and CPU for the new GPUs, and it has to be Ryzen 4000 because Intel has exited the CPU game. I'm guessing "on track for 2020" means December 31st, 11:59pm?

 

[maddie]

I'm looking to get a PCI-E 4.0 board and CPU for the new GPUs, and it has to be Ryzen 4000 because Intel has exited the CPU game. I'm guessing "on track for 2020" means December 31st, 11:59pm?

11:59:59 pm to be precise.

 

[Kenmitch]

I'm looking to get a PCI-E 4.0 board and CPU for the new GPUs, and it has to be Ryzen 4000 because Intel has exited the CPU game. I'm guessing "on track for 2020" means December 31st, 11:59pm?

I'm thinking we'll hear something this month? Outing Zen 3 before Big Navi would be AMD's best option. Testing the new GPU's should be done with PCI 4.0 rigs. Why not seed some Zen 3's to a handful of reputable reviewers for the task?