Discussion Intel current and future Lakes & Rapids thread

[majord]

AMD have likely picked all the low hanging fruit that Jim Keller left them for Ryzen. I highly doubt they'll manage more than a 5-10% IPC increase again, before going to a new architecture. Hopefully I'm wrong though, their pace of innovation to this point (using Jim Keller's designs) have been refreshing.

That's what people said about Zen 1 , then Zen 2...

PS .. Keller Didn't design the core. No one person did.

 

[itsmydamnation]

Kellar didn't have anything to do with corw design, if you listen to Kellar on podcasts etc you can find out what he is about.

I Iove the passage where he said for his first people management position he was worried so read 20 management books. Turned out he read 19 more then any of the MBA types.... Lol

 

[uzzi38]

AMD have likely picked all the low hanging fruit that Jim Keller left them for Ryzen. I highly doubt they'll manage more than a 5-10% IPC increase again, before going to a new architecture. Hopefully I'm wrong though, their pace of innovation to this point (using Jim Keller's designs) have been refreshing.

That's a bad take to have.

Zen 3 is a major redesign and new product family just so that AMD could create new low hanging fruit to take care of for Zen 4.

This is one of the things Jim Keller actually instilled into AMD. The best way to see generational leaps is to design a new architecture, optimise it and repeat the process.

 

[TheGiant]

eGPU gaming: Ice Lake-U processors offer nearly 50% better performance in some games than Comet Lake-H processors do

It turns out that Ice Lake laptops are better for eGPU gaming than their more powerful Comet Lake-H counterparts, Jarrod's Tech has discovered. Thunderbolt 3 controller placement seems to be at the heart of the difference in performance, which can be nearly 50% in some modern triple-A games...

www.notebookcheck.net

interesting
this should be a prediction of rocket lake improvement imo over sky/cometlake
it is ofc difficult to measure the TB3 implementation difference

 

[uzzi38]

eGPU gaming: Ice Lake-U processors offer nearly 50% better performance in some games than Comet Lake-H processors do

It turns out that Ice Lake laptops are better for eGPU gaming than their more powerful Comet Lake-H counterparts, Jarrod's Tech has discovered. Thunderbolt 3 controller placement seems to be at the heart of the difference in performance, which can be nearly 50% in some modern triple-A games...

www.notebookcheck.net

interesting
this should be a prediction of rocket lake improvement imo over sky/cometlake
it is ofc difficult to measure the TB3 implementation difference

No that's literally just the performance uplift created by having Thunderbolt on die instead of on PCH and data having to go across the OPIO.

 

[RTX]

Yes, but curiously, they're still only 4c parts. Intel has yet to market anything on any iteration of 10nm with more than 4c. 8c Tiger Lake-H is coming sometimes next year, albeit in unknown quantity.

Is the real reason for the delay for Tigerlake-H is just to sell Cometlake-H for 1 year first?

 

[Exist50]

No that's literally just the performance uplift created by having Thunderbolt on die instead of on PCH and data having to go across the OPIO.

Also 32Gb/s vs 22Gb/s of PCIe.

 

[uzzi38]

Is the real reason for the delay for Tigerlake-H is just to sell Cometlake-H for 1 year first?

No, probably production related. AFAIK TGL-H laptops are looking like H2 for at least one laptop vendor, though idk when actual launch will be.

 

[TheGiant]

No, probably production related. AFAIK TGL-H laptops are looking like H2 for at least one laptop vendor, though idk when actual launch will be.

IMO the real reason is rocket lake s
Intel cant release a H CPU which will be faster than anything else they have
And ofc they have tons of 14nm capacity so they must use it

 

[DrMrLordX]

Is the real reason for the delay for Tigerlake-H is just to sell Cometlake-H for 1 year first?

Officially, we don't know. Unofficially, I suspect yields are an issue. @TheGiant may have a point but I sincerely hope he's wrong.

 

[TheGiant]

No that's literally just the performance uplift created by having Thunderbolt on die instead of on PCH and data having to go across the OPIO.

are you really sure about it?
I can't find any data about this, the article itselft has the statement of latency but with no data support behind it
the min fps numbers are so low...
if it really is the latency issue (min fps really is latency based) we should have a measurement of like 10300H to 10980HK with no min fps increase because latency is the bottleneck
but I can't find such article
do you have anything? thanks

 

[Timorous]

Intel cant release a H CPU which will be faster than anything else they have

Yonah was faster than P4 in a lot of cases so that has not stopped Intel before.

 

[TheGiant]

Yonah was faster than P4 in a lot of cases so that has not stopped Intel before.

exactly, I remember dothan small form factor boards with 2.0GHz dothan ovecrlocked to 2.6GHz faster than the p4 emergency editions
but IMO they are trying to avoid than message now

 

[yuri69]

That's a bad take to have.

Zen 3 is a major redesign and new product family just so that AMD could create new low hanging fruit to take care of for Zen 4.

This is one of the things Jim Keller actually instilled into AMD. The best way to see generational leaps is to design a new architecture, optimise it and repeat the process.

True dat. Zen 3 can grow a lot on a new process. The structure increases were modest - OoO buffer size increases and no size increase for the ucache. Also Zen 3 still relies on the elder 12nm IOD which is both last-gen/slowish and eats the TDP budget.

So Zen 4 with pushed structure sizes and a up-to-date IOD tech can make a nice IPC boost.

 

[coercitiv]

Ryan was not impressed. (and tbh I was expecting a bit more factual information as well)

https://twitter.com/x/status/1326232760982368256

 

[Hitman928]

Ryan was not impressed. (and tbh I was expecting a bit more factual information as well)

https://twitter.com/x/status/1326232760982368256

Wrong thread.

 

[coercitiv]

Wrong thread.

It's Intel's Chief Performance Strategist commenting on marketing material from the competition. I think I'm in the right thread.

 

[Hitman928]

It's Intel's Chief Performance Strategist commenting on marketing material from the competition. I think I'm in the right thread.

Ok, I guess. You should at least give it context because no one was talking about the Apple event in this thread.

 

[blckgrffn]

Well, my Dad got impatient (I didn't help) and he ordered a 10850KA(!)/z490 combo from NewEgg last night at 1am (when all the best decisions regarding online purchases are made) along with some bling RGB ram kit that was on Shell Shocker. Woohoo. Less than $700 after taxes, and he was looking at the 5900x so probably marginally cheaper overall. I know that the 5900x over the long haul will be the much better CPU, but you can't actually buy it right now and that leads me to...

We are *probably* going down there for Thanksgiving and I can build it out then. Zen 3 supplies are probably super tight through the end of the year at least (I am thinking) and at this rate we may not see them for Christmas or into early next year. Got to do it now so he can enjoy the winter gaming time (he's a farmer, so it's really seasonal) with new hardware. The pandemic just keeps giving.

I never expected to be building a i9 based PC any time soon, but here we are. It should handily outperform his 5820k which was always a terrible clocker. Given the TDP on that thing, this is like a sidestep and I'll set PL1 to like 88W (haha, get it?) and PL2 to ~140W since he has a 150W BeQuiet! aircooler and and Corsair that is built for airflow.

Competition at least pushed better pricing and some more exotic configurations. Haha, 10C/20T Intel for ~$430 ($400 at MC)!?? And a gosh darned bundled game?

 

[IntelUser2000]

are you really sure about it?

On-die is almost always faster.

The speed-up is a combination of lower latency due to being on-die, faster throughput due to not going through the OPI, and faster throughput because its actually getting 32Gb/s versus 22Gb/s on the previous generation.

It's Intel's Chief Performance Strategist commenting on marketing material from the competition. I think I'm in the right thread.

The Apple M1 is an absolute killer. The CPU's single thread performance is Tigerlake or better, they are claiming 40% better overall performance so the MT performance might be on par with Renoir, and the GPU is a 2.6TFlop part that performs at least on par with a well-optimized + 28W Iris Xe(which doesn't exist on production laptops) and possibly even faster than that.

All that at 10W, and also with battery life advantage that won't be matched with any x86 device.

This is a combination of process advantage + the CPU team's ability to execute. AMD/Intel both stood still going nowhere for many years. Apple is the hypothetical AMD that didn't have Bulldozer in their history, and a hypothetical Intel that executed on every CPU and process since Skylake.

I also don't get how Intel is unable to get the PCH on-die, while AMD has done it since Carrizo I think?

 

[tamz_msc]

GPU is a 2.6TFlop part that performs at least on par with a well-optimized + 28W Iris Xe(which doesn't exist on production laptops) and possibly even faster than that.

Forget Iris Xe, M1 GPU is on par with mobile GTX 1050 Ti based on Tflops and fill rates.

 

[jeanlain]

That Ryan Shrout guy whose tweet was linked above should be happy that Apple didn't release precise benchmark numbers for the M1. Anandtech just did for the A14, and things don't look so rosy for TGL.
As for the GPU, the A12Z already beats the Intel Xe in 3DMark Wild life (some blame Windows Vulkan drivers, but I don't buy that theory). And consider that the M1 is probably quite a bit faster than the A12Z.

 

[jeanlain]

Forget Iris Xe, M1 GPU is on par with mobile GTX 1050 Ti based on Tflops and fill rates.

Yes, and I believe that the TBDR design of the Apple GPU compares quite favourably in fps/TFLOPS versus more traditional (IMR) rasterizers.

EDIT: the 1050ti for notebook yields a median score of about 16500 at 3DMark Wild life, which is 25% better than the A12Z score. The M1 should beat that.

 

[Antey]

Well, if A14 (4 cores) GPU is 30% faster than A12 (4 cores) GPU i think it wouldnt be a surprise if M1 GPU is 30% faster than A12Z. I think will see 16-17k scores for that benchmark. not a surprise, it's a 2,6 Tflops gpu. Intel xe has 96 execution units, i don't remember the exact die size but it was close to 40 mm2. A12Z GPU is 30-35mm2. I dont know about number of ALUs or frequencies but i doesnt seem to be like a super gpu. Vega 8 in the other hand is very small (like 15-20 mm2) and it has a very good performance and clocks very high with a low power draw (for an old arch).

 

[Antey]

for the M1, if the die size is 120mm2 as andrei says, for a total of 436665 pixels i count 87500 for the GPU, more or less 20% of the die area, more or less 24-25 mm2.

https://images.anandtech.com/doci/16226/M1.png

give me 16CU Navi2 Rembrandt / 2 GHz and LPDDR5/DDR5 as leaks claim and that M1 GPU wont be that interesting, it's just that amd has been using the same arch for 3 years... and it wouldnt get much bigger, not much further than 30-35 mm2 (@6nm, not even N5)