Discussion Intel current and future Lakes & Rapids thread

[mikk]

Intel's 11th Gen Processor (Rocket Lake-S) Architecture Detailed

11th Gen Intel Core S-series desktop processors (code-named “Rocket Lake-S”) will bring more performance to gamers and PC enthusiasts.

newsroom.intel.com

https://newsroom.intel.com/wp-content/uploads/sites/11/2020/10/Intel-Rocket-Lake-S-Architecture.pdf

About the Architecture: The 11th Gen Intel Core S-series desktop processor will introduce a new desktop architecture, called Cypress Cove, designed to transform hardware and software efficiency and increase performance.

New Cypress Cove architecture featuring Ice Lake Core architecture and Tiger Lake Graphics architecture.

 

[jpiniero]

Feels unecessary to officially announce Rocket Lake at this point. Release isn't for another 5 months, right?

 

[inf64]

Intel's 11th Gen Processor (Rocket Lake-S) Architecture Detailed

11th Gen Intel Core S-series desktop processors (code-named “Rocket Lake-S”) will bring more performance to gamers and PC enthusiasts.

newsroom.intel.com

https://newsroom.intel.com/wp-content/uploads/sites/11/2020/10/Intel-Rocket-Lake-S-Architecture.pdf

It's interesting that they stated IceLake uarchitecture (which is ~18.5% faster according to intel) while Cypress Cove has only "double digits" listed, with no benchmark numbers at all. If IceLake is a product that is being back ported, why wouldn't they state ~18% IPC improvement? I'm not questioning that it is Icelake, I just find it interesting that no number is being stated. Double digit is 10% as well, very interesting.

 

[Bouowmx]

Intel's 11th Gen Processor (Rocket Lake-S) Architecture Detailed

11th Gen Intel Core S-series desktop processors (code-named “Rocket Lake-S”) will bring more performance to gamers and PC enthusiasts.

newsroom.intel.com

https://newsroom.intel.com/wp-content/uploads/sites/11/2020/10/Intel-Rocket-Lake-S-Architecture.pdf

Projected results are quite old: from August. Perhaps I can be surprised with Q1 meaning January/February.
Also, PDF says Rocket Lake graphics supports AV1 encode, which sounds unusual in that it is ahead of Tiger Lake in this area. It looks like a mistake: on Reddit, Intel lists AV1 decode only.

 

[jpiniero]

It's interesting that they stated IceLake uarchitecture (which is ~18.5% faster according to intel) while Cypress Cove has only "double digits" listed, with no benchmark numbers at all. If IceLake is a product that is being back ported, why wouldn't they state ~18% IPC improvement? Double digit is 10% as well, very interesting.

Pretty much confirms that it is a bastard version of Sunny Cove and not the real thing.

 

[mikk]

Projected results are quite old: from August. Perhaps I can be surprised with Q1 meaning January/February.
Also, PDF says Rocket Lake graphics supports AV1 encode, which sounds unusual in that it is ahead of Tiger Lake in this area. It looks like a mistake: on Reddit, Intel lists AV1 decode only.

It could be a mistake, I believe it is a mistake in this case. But in general the fixed function unit is separate and could differ from earlier platforms. Same for the display engine. The latest news rumour from notebookcheck claimed AV1 accelerated encode for ADL-S/ADL-P and it's still based on Xe LP.

 

[jpiniero]

It could be a mistake, I believe it is a mistake in this case. But in general the fixed function unit is separate and could differ from earlier platforms. Same for the display engine. The latest news rumour from notebookcheck claimed AV1 accelerated encode for ADL-S/ADL-P and it's still based on Xe LP.

Definately a typo for Rocket Lake, and Alder-S seems to have the exact same IGP as Rocket Lake. ADL-P, we'll have to see.

 

[IntelUser2000]

Ugh. So the gains are going to be closer to 10% than 20%, maybe closer to 10% than 15%.

I knew backporting would require sacrifices. This is why process is just as important as design, if not more. Good process doesn't guarantee a good product, but is a foundation for one.

Rocketlake is indeed this generations' Prescott. Hot, power hungry, and not always faster compared to the predecessor in an era where the main competition is executing at full speed.

Except its actually worse. Because they have many competitors now. It's a combination of Netburst screwups in the early 2000's combined with intense competition of the mid 90s.

 

[inf64]

Ugh. So the gains are going to be closer to 10% than 20%, maybe closer to 10% than 15%.

I knew backporting would require sacrifices. This is why process is just as important as design, if not more. Good process doesn't guarantee a good product, but is a foundation for one.

Rocketlake is indeed this generations' Prescott. Hot, power hungry, and not always faster compared to the predecessor in an era where the main competition is executing at full speed.

Except its actually worse. Because they have many competitors now. It's a combination of Netburst screwups in the early 2000's combined with intense competition of the mid 90s.

The sad thing in this whole mess is that IceLake is actually a great design and a first fully fledged tock from intel in like 7 years (being closer to 20% which is generally considered a tock uplift). Too bad that they are so behind in process tech that their roadmap is being butchered with Rocketlakes and hybrid designs on desktop instead of a full Zen3 like products they are generally capable of making.

 

[mikk]

Definately a typo for Rocket Lake, and Alder-S seems to have the exact same IGP as Rocket Lake. ADL-P, we'll have to see.

Looks like you didn't understand the posting. It's not about the GPU, it's about the fixed function unit (or display engine).

 

[Zucker2k]

What if the delay was to add this feature in order to make Rocketlake more attractive against the competition? Or, is that impossible from an engineering and time perspective? On desktop, power is not going to be a factor too much so the Xe graphics capabilities could easily be pushed to produce better results than mobile.

 

[Zucker2k]

More info here:

https://www.reddit.com/r/intel/comments/jkapv5

 

[IntelUser2000]

The sad thing in this whole mess is that IceLake is actually a great design and a first fully fledged tock from intel in like 7 years (being closer to 20% which is generally considered a tock uplift). Too bad that they are so behind in process tech that their roadmap is being butchered with Rocketlakes and hybrid designs on desktop instead of a full Zen3 like products they are generally capable of making.

I agree with everything, but I believe the hybrid design if done well is going to give us some real benefits.

The Bigger core is too large to put many of them, but it allows for both high perf/clock and high frequencies. The Big core is much much smaller which caps clock speeds, but it still will be a powerful core and you'll be able to put many more.

I originally imagined they would make the bigger core really really wide and large to have very high perf/clock with moderate clocks, but they seem to be using it for clockspeed instead. Perhaps clockspeed is still the easiest way to gain performance, and its easier to synchronize between the two the closer they are in features.

The new plan could be essentially making the smaller core Core-lite with perf/clock not much below the bigger core, and largely design methodologies and clock speed goals resulting in core size differences. Ideal may be not the perf/clock differences between the two cores being 40-50%, but 20-25%, with the smaller core still being 1/3rd the size and more power efficient.

If Alderlake was say, 7nm, it still could have been hybrid but much larger, such as being 8+16/16+16, 12+32, etc.

On desktop, power is not going to be a factor too much so the Xe graphics capabilities could easily be pushed to produce better results than mobile.

Yea except Rocketlake is coming with EUs cut down to 32EUs. 8 Sunny Cove class cores are physically large already. Even Alderlake S is going to stay at 32EUs.

I assume/hope the GPU has a chance to be significantly better still. 10nm ESF will allow higher GPU clocks, along with minor uarch improvements + LPDDR5.

 

[jpiniero]

Looks like you didn't understand the posting. It's not about the GPU, it's about the fixed function unit (or display engine).

Correct. Saying that Rocket Lake's IGP has AV1 encode is 100% a typo. DG2 might have AV1 encode but Alder-S is very unlikely to be anything other than identical to Rocket Lake's IGP.

 

[IntelUser2000]

Correct. Saying that Rocket Lake's IGP has AV1 encode is 100% a typo. DG2 might have AV1 encode but Alder-S is very unlikely to be anything other than identical to Rocket Lake's IGP.

Actually every generation that wasn't blocked by the process, the GPU improved in some way, whether in graphics architecture to improve performance, or better media.

Arrandale/Clarkdale's iGPU, Haswell, Kabylake all had notable and often significant improvements despite being on the same generation. It would be actually foolish for Alderlake to stay the same, especially when AMD and Nvidia are making big moves as well.

 

[jpiniero]

Actually every generation that wasn't blocked by the process, the GPU improved in some way, whether in graphics architecture to improve performance, or better media.

I think that's where the Alder Lake-S IGP being 14 nm comes in - it's the exact same as Rocket Lake's. Plus that might be part of why it's coming sooner than P.

Don't want to waste 10 nm or N6 wafers on desktop IGP if you don't need to.

 

[IntelUser2000]

Ok, maybe double digit still means close to 20%.

They claimed double digit for Tigerlake. We got 25-30%.

I hope they are seriously underselling Xe GPU gains. 50% over Gen 9, that's terrible. Yes, even with 32EUs. Even the Gen 11 with 32EUs is 20-30% faster than Gen 9.

 

[itsmydamnation]

The sad thing in this whole mess is that IceLake is actually a great design and a first fully fledged tock from intel in like 7 years (being closer to 20% which is generally considered a tock uplift). Too bad that they are so behind in process tech that their roadmap is being butchered with Rocketlakes and hybrid designs on desktop instead of a full Zen3 like products they are generally capable of making.

poor old cannonlake , its like it never existed

 

[SAAA]

Ok, maybe double digit still means close to 20%.

They claimed double digit for Tigerlake. We got 25-30%.

I hope they are seriously underselling Xe GPU gains. 50% over Gen 9, that's terrible. Yes, even with 32EUs. Even the Gen 11 with 32EUs is 20-30% faster than Gen 9.

I believe the description of Cypress Cove is clear enough from their words: it's a new arch with Sunny core and Tiger Graphics. So I expect pretty much Icelake on 14nm, with the benefit of not being restrained to max 4.1 GHz like mobile 10 nm parts before. 18% IPC and somewhat higher clocks than Comet lake is a decent jump.

Graphics is weird: performance should be closer to 2x over Gen9: 50% from EUs count alone, then all the improvements from Gen12/Xe.

If they are underselling increase on graphics like that the core could easily be 20% faster, stretching Icelake legs with desktop level memory and power.

 

[dullard]

I hope they are seriously underselling Xe GPU gains. 50% over Gen 9, that's terrible. Yes, even with 32EUs. Even the Gen 11 with 32EUs is 20-30% faster than Gen 9.

The tests used in their 50% gain projection (3DMark, 3DMark Fire Strike, and 3DMark Night Raid) are also CPU dependent. The end result is that the claimed ~50% gain is combination of the GPU gain and the CPU gain. We'll have to see actual GPU-only tests to know how much better the GPU is.

 

[Shivansps]

50% over Gen9 it would mean around 10% below 2200G Vega 8 at stock. But with the advantage of faster CPU cores it likely match it or be faster in a few games, thats not too bad actually, compared to what they used to have. Overall it is still terrible for a IGP in 2021.

 

[IntelUser2000]

AMD vs Intel Integrated Graphics: Can't We Go Any Faster?

Integrated graphics tested: ubiquitous and oh-so-slow!

www.tomshardware.com

50% increase over the iGPU that was considered slow back in 2015. Less than half of the performance of Ryzen 3000G.

@dullard See unlike the vague CPU numbers, they are quite specific about it. 3DMark numbers tend to be on the optimistic side for iGPUs. Also if you ever ran 3DMark tests on iGPUs, they run at single digit frames, explaining why the results are optimistic. It's saying if you run games at settings that make it unplayable, you'll get the 3DMark gains. Also it ignores the complex interaction that occurs in iGPU setups, where a good performing product carefully balances out the power distribution between the CPU and the GPU cores.

Can you imagine if they had Rocketlake-U instead of Tigerlake-U being everywhere? Going against Renoir!

 

[DrMrLordX]

poor old cannonlake , its like it never existed

It may as well not have.

That aside, 250W for an 8c seems a bit ridiculous. One of the potential problems I see here is that, in games, Rocket Lake-S may actually chew up more power than a 10900k since a 10900k may not use all its cores during gaming. Rocket Lake-S is more likely to do so. That increase in performance will come at a price.

 

[Dayman1225]

If anyone was wondering what an 1165G7 with a PL1 of 45w and PL2 of 65w looked like... here you go

https://twitter.com/x/status/1322006917523230720

https://twitter.com/x/status/1321939838233313285

 

[mikk]

Correct. Saying that Rocket Lake's IGP has AV1 encode is 100% a typo. DG2 might have AV1 encode but Alder-S is very unlikely to be anything other than identical to Rocket Lake's IGP.

It's not about RKL, it's about ADL. You told this:


Alder-S seems to have the exact same IGP as Rocket Lake

According to Notebookcheck ADL-S supports 3 display outputs and RKL-S supports only 3 display outputs. You have to understand that the display and media engine could differ even when both are Xe LP based.