Discussion Intel current and future Lakes & Rapids thread

[Asterox]

https://twitter.com/x/status/1318539792733229057

This is 12/24 Zen 3 vs 8/16 Rocket Lake CPU Geekbench comparison. From this we can easily estimate, that in "CPU Multithread 8/16 Zen 3 will be significantly faster vs 8/16 Rocket Lake".

 

[DrMrLordX]

Zen3 wins single thread? Wow.

 

[dullard]

I swear every time I visit this thread you guys substract a few 100Mhz and % IPC from Rocket Lake. Now it's just a refresh of Skylake?

Mebiuw still claims 18% IPC in Spec2017 over Comet Lake.

https://twitter.com/x/status/1317641371906232320

Which makes sense if the core is from Willow Cove bolted onto the Sunny Cove cache configuration.

There are some major changes in Rocket Lake. But, no matter what, Intel is still up against a power wall. At this point, even if it was 18% better in all uses (not just Spec2017), that will still leave a lot of people disappointed. 18% gain in ~18 months is okay, but nothing spectacular.

 

[DooKey]

There are some major changes in Rocket Lake. But, no matter what, Intel is still up against a power wall. At this point, even if it was 18% better in all uses (not just Spec2017), that will still leave a lot of people disappointed. 18% gain in ~18 months is okay, but nothing spectacular.

Since when is 18% gain in 18 months not spectacular?

 

[IntelUser2000]

10 cores -> 8 is a requirement if you don't want these chips to fight nV's GA102 on power consumption, and the new Xe graphics, while much more capable and powerful, need a lot more area too.

You are wrong on the graphics. Xe doesn't make more space than Gen 11, and in Rocketlake its cut down to 32EUs.

This is 12/24 Zen 3 vs 8/16 Rocket Lake CPU Geekbench comparison. From this we can easily estimate, that in "CPU Multithread 8/16 Zen 3 will be significantly faster vs 8/16 Rocket Lake".

Rocketlake is terribly underperforming there, unless we're expecting slower performance in than 9900K.

9900K gets 9K in multi-thread or more.

 

[.vodka]

You are wrong on the graphics. Xe doesn't make more space than Gen 11, and in Rocketlake its cut down to 32EUs.

Good to know! I stand corrected.

Despite being pessimistic over what RKL could be as an end product for a 2021 release, I'm eager to see it.

 

[dullard]

Since when is 18% gain in 18 months not spectacular?

When you are more than 18% behind in many tests.

 

[Markfw]

When you are more than 18% behind in many tests.

Well, Intel is not catching up anytime soon.

And thanks for the down vote in another thread.... NOT

 

[dullard]

And thanks for the down vote in another thread.... NOT

You are welcome. Thanks for the constant reminder that you aren't here to provide unbiased information.

 

[Ajay]

Good to know! I stand corrected.

Despite being pessimistic over what RKL could be as an end product for a 2021 release, I'm eager to see it.

Yeah, I'd rather see the reviews. Just geekbench on and ES chip isn't telling me too much.
I'm sure Intel are going after gaming performance with RKL.

 

[Markfw]

You are welcome. Thanks for the constant reminder that you aren't here to provide unbiased information.

I do provide unbiased information. I go for the best at the time. Right now its AMD and 99% of this forum knows that, but YOU don't have a clue.

Back to talking about Intels trials and tribulations.

 

[LightningZ71]

A comet lake refresh for the i3 and below could easily be done on a small 4c die. Leave L1 and L2 untouched, but include a large L3 (double the existing Comet Lake i3 L3 size) and let it clock to current Comet Lake i7/i9 levels and you'll have celerons, Pentiums and i3 chips worth an 11th gen name. It's ok to not enable PCIe 4.0 on lower SKUs for Intel.

 

[jpiniero]

A comet lake refresh for the i3 and below could easily be done on a small 4c die. Leave L1 and L2 untouched, but include a large L3 (double the existing Comet Lake i3 L3 size) and let it clock to current Comet Lake i7/i9 levels and you'll have celerons, Pentiums and i3 chips worth an 11th gen name. It's ok to not enable PCIe 4.0 on lower SKUs for Intel.

The current higher end Comet Lake i3's come with 8 MB L3, so doubling would mean a new die. They aren't going to do that. They could do 4C8T with 12 MB L3.

 

[RTX2080]

I swear every time I visit this thread you guys substract a few 100Mhz and % IPC from Rocket Lake. Now it's just a refresh of Skylake?

Mebiuw still claims 18% IPC in Spec2017 over Comet Lake.

https://twitter.com/x/status/1317641371906232320

Which makes sense if the core is from Willow Cove bolted onto the Sunny Cove cache configuration.

I will take a grain of salt about 18% increase statement, Anand result showed 1185g7(max 4.8Ghz) has only 10% advantage against 10900k(max 5.3Ghz) in Spec2017(int), which makes me curious how they achieve another 8% increase since Rocketlake is a 14nm++++++ backport version of TGL/ICL. (OTOH TGL in Spec2017(fp) has 17% advantage against CML. )

Intel’s Tiger Lake 11th Gen Core i7-1185G7 Review and Deep Dive: Baskin’ for the Exotic

www.anandtech.com

(And just like I said before, this guy Mebiuw is a bit infamous at china, that's another reason why I keep doubt at all these statements which are from his mouth)

 

[IntelUser2000]

@cortexa99 I'm actually skeptical in an opposite way, that is about Anandtech conclusions.

The 9900K is significantly faster per MHz in SpecInt2017 compared to the 8550U, which uses the same Skylake core. The Zen 2 based 3900X is 15% faster per MHz than the 4700U.

Laptops lose perf/clock not just because its affected by power and thermal limits even in single thread, the memory clocks down, the storage clocks down, and to top it off, it takes longer to sleep and wake up from it.

I will not trust any numbers on Rocketlake-S short of the day where major review sites benchmark across a suite of applications. No ES leaks by twitter, no previews by a small foreign site. No, multiple sites, and multiple suites all clear and easily accessible with detailed information about the test system.

Despite being pessimistic over what RKL could be as an end product for a 2021 release, I'm eager to see it.

I'm also pessimistic about Rocketlake.

All the hype over a backport would have made sense if instead of the 9900K, we had Rocketlake. At the latest, 10900K should have been Rocketlake.

At this point I really wonder why they couldn't have used Tigerlake-H for just the K series SKUs. Just 3. I mean 11900K, 11700K, and 11600K. They really can't afford fab space? I call BS.

I see a lot of minis here. I mean the endless Skylakes are "mini-Netburst" with super high clocks, and super high power use. Rocketlake is mini-Prescott. It'll have scenarios where it wins against the predecessor, but also lose, and use ton of power to do it.

 

[jpiniero]

At this point I really wonder why they couldn't have used Tigerlake-H for just the K series SKUs. Just 3. I mean 11900K, 11700K, and 11600K. They really can't afford fab space? I call BS.

Yields obviously. Who knows how widespread the 8 core Tiger Lake-H will be. It's easier with Alder Lake because of chiplets but you have no such luck with Tiger Lake.

 

[DrMrLordX]

Yields obviously. Who knows how widespread the 8 core Tiger Lake-H will be. It's easier with Alder Lake because of chiplets but you have no such luck with Tiger Lake.

Well that and there's the issue of readiness. Tiger Lake-H won't hit the market until after Rocket Lake-S' launch quarter, for whatever reason. It may be an issue with yielding enough dice to ship product (which is related to your commentary).

 

[Asterox]

You are wrong on the graphics. Xe doesn't make more space than Gen 11, and in Rocketlake its cut down to 32EUs.



Rocketlake is terribly underperforming there, unless we're expecting slower performance in than 9900K.

9900K gets 9K in multi-thread or more.

No it is not, look at Singlethread score vs Zen 3 or it is kind of expected for 18% higher IPC.As far i now, in some situations Rocket Lake will have some IPC regression.

Intel Corporation Rocket Lake Client Platform vs Micro-Star International Co., Ltd. MS-7C34 - Geekbench Browser

In short, 8/16 RocketLake or RocketLake Desktop in general is kind of pointless launch+power hog champion.

It will be hilarious, in "Singlethread strong games" even R5 5600X will have the same performanse as 8/16 Rocket Lake.But power consumption numbers, this will be total debacle for Rocket Lake.

This comparison, but with new R5 5600 in gaming will be very close to 8/16 Rocket Lake power hog champion.

Backporting 10nm CPU to 14nm, this is bad idea or solution no doubt. But intel is in a situation without a good solution anyway.

 

[IntelUser2000]

@Asterox You can be stubborn as you want, but the reality is Geekbench is a user-submitted benchmark, which makes the results notoriously unreliable.

Based on Geekbench I can show you 1165G7 being slower than 1065G7, and Broadwell being faster than Zen 2.

 

[Gideon]

@Asterox You can be stubborn as you want, but the reality is Geekbench is a user-submitted benchmark, which makes the results notoriously unreliable.

Based on Geekbench I can show you 1165G7 being slower than 1065G7, and Broadwell being faster than Zen 2.

There were also early Tiger Lake geekbench scores there that topped out at 4.4 Ghz and were slower than 1065G7. I would definitely not take these Rocket Lake scores as fact yet.

 

[mikk]

Intel Lands A Hefty Tiger Lake Graphics Optimization - Phoronix

www.phoronix.com

Intel OpenGL/Vulkan Linux Drivers Strike Another Optimization For Tiger Lake - Phoronix

www.phoronix.com

By making use of the data cache for UBO pulls, there is generally up to another few percent improvements for various OpenGL and Vulkan games running on Linux -- either natively or through the likes of the DXVK layer.

 

[stebler]

There is a new codename in the Windows graphics driver: Goldwater Lake (GWL).

 

[DrMrLordX]

There were also early Tiger Lake geekbench scores there that topped out at 4.4 Ghz and were slower than 1065G7. I would definitely not take these Rocket Lake scores as fact yet.

https://browser.geekbench.com/v5/cpu/3037923.gb5

That Rocket Lake-S sample was running at nearly 5 GHz. I'm not a huge fan of Geekbench, mind you. But I don't think we can assume the Rocket Lake-S sample was running at a low clockspeed.

 

[Timorous]

There is a new codename in the Windows graphics driver: Goldwater Lake (GWL).

Is this Intel accepting the golden shower AMD are giving them at the moment?

 

[jpiniero]

Is this Intel accepting the golden shower AMD are giving them at the moment?

Intel does need a product released in 2022 that's not Meteor Lake. Perhaps that's what Goldwater Lake?