Discussion Intel current and future Lakes & Rapids thread

Page 750 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
With your pts charts (which I appreciate) you really ought to add clearer notes of what we are actually looking at.

So "a" is 7950X (16c/32t) and the other is actually 2x 8490H (so 120c/240t). :oops: Honestly find that hard to believe, both the difference between the two chips as well as the difference between AMX and no-AMX. Are Intel chips so bad in ONNX?
The Benchmark does not extend past 16 Threads.
 
  • Like
Reactions: moinmoin
Jan 12, 2021
30
64
61
Not within the same ecosystem they aren't. Gracemont currently occupies the position held by cores like A510 in phone SoCs. ARM just isn't throwing around as many fat cores (yet) as x86. Give it time, you'll see.
The Gracemont is far above any A5XX, it is much wider and faster, comparable to the A7XX. The position held by cores like A510 in phone SoCs should be similar to LPE-core present in meteor lake SOC-die
 

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
How many of you believe that 8490H Sapphire Rapids would be faster if 9654 Genoa would be restricted to only 6 Channel?
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136

That's why we have to talk about "throughput" or "work rate" to include factors outside of architecture like memory subsystem. Otherwise we end up so far into the weeds we can't even compare generations as you try and "subtract" out memory performance and things like that. I just wanted to more real world percentage gain from gen to gen.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
Oooo. Me likey. Do you have plan on adding Golden Cove to the mix?

I don't have AMD data. I will eventually add Golden Cove data. It takes a bit of time to compile good benches, determine clocks that test were run at etc... do some corroborating to make sure things make sense. I actually enjoy the work but with family and kids it's hard to find the time.

I also want to do the Skylake to Sunny Cove comparison again using Comet Lake and Rocket Lake as the comparison platforms.

I'll get around to it eventually and post the results.
 

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
I have read the E cores are actually "Cinebench accelerators." While funny, there is some truth to that statement as there aren't a lot of applications that will fully utilize all 16 E's in a 13900 series part.

So I guess the first 8 E's are actual general purpose CPU's but the last two clusters are more "Cinebench accelerators" as they are only activated in Cinebench and other benchmarks!

This is of course meant to be humorous but in all seriousness in real world applications 13700K is generally very close in performance to the 13900K. The P's are clocked virtually the same and not many applications go "deeper" than 8 E's.
 
  • Like
Reactions: Joe NYC and adamge

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
No way. Genoa on 2 channel screams. (Zen 4)
True, I have the Numbers. 9654 restricted to 6 Channels is still faster than the 8490H. Also, the Recent Phoronix AVX512 article did not include the 9554 64C/128T CPU on the AVX512 Benchmarks, but I also have the Numbers and the 9554 beats the 8490H in AVX512 Benchmarks.
 
  • Like
Reactions: Joe NYC

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
I'll respond to these assuming they're made in good faith. That is a very tenuous assumption.
Uh, the Tremont "Ridge" SoCs are not ASICs.
They are, by definition. They're highly targeted at networking applications, including a boatload of specialized hardware. But I'm not going to debate semantics.
Personally I think Intel will sell it to a few select customers and that's it.
It will target the same markets as Bergamo. Namely, cloud providers. That is a huge market.
They could have cores as good as Zen4.
Twice the PPA of Golden Cove, iso-process, would be far beyond Zen 4. So I ask again, what magic do you claim they're able to pull out to eliminate the inherent gap between big and small cores? And again, even AMD's acknowledged that.
Not within the same ecosystem they aren't.
What on earth is that supposed to mean? The A5XX is designed for very light, background workloads, but Atom is more of a throughput processor. But let's ignore all that. Raw performance, Atom and N series are directly comparable. So why claim there's a market for one but not the other?
ARM just isn't throwing around as many fat cores (yet) as x86. Give it time, you'll see.
No, instead, most of the market right now is using their N series, which as I said, are very comparable to Atom in both their performance and place in the lineup.
 

Cardyak

Member
Sep 12, 2018
72
159
106
Canceled more like 6 years back, iirc. And it was to be far more radical than just a new uarch. Probably the biggest change in CPUs since the proliferation of OoO computing.

Do we have any idea what Ocean Cove actually entailed? I know it was a hugely ambitious project that was cancelled, but I still can’t confirm whether it was evolutionary (I.E: Intel’s standard approach of “Wider, Deeper, Smarter”) or whether it was a more revolutionary departure from the standard Core family that had brand new ideas.

I also don’t know where/if the ideas of Ocean Cove died and where the ideas for Royal Core originated from, are the lines blurred here? Is it possible Royal took some of the ideas from Ocean Cove and is essentially its spiritual successor? Or are they 2 completely different approaches to processor design and there is limited overlap between the two?
 

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
Do we have any idea what Ocean Cove actually entailed? I know it was a hugely ambitious project that was cancelled, but I still can’t confirm whether it was evolutionary (I.E: Intel’s standard approach of “Wider, Deeper, Smarter”) or whether it was a more revolutionary departure from the standard Core family that had brand new ideas.

I also don’t know where/if the ideas of Ocean Cove died and where the ideas for Royal Core originated from, are the lines blurred here? Is it possible Royal took some of the ideas from Ocean Cove and is essentially its spiritual successor? Or are they 2 completely different approaches to processor design and there is limited overlap between the two?
Hopefully they are different. It would have been years between the development between the two next gen architecture products, so I'm assuming new techniques would have arisen, more internal testing would have been done, and new solutions would have been figured out.
For example, core fusion was investigated in Intel after they bought out soft machines? was it? and eventually got mothballed internally after testing for a bit.
 
  • Like
Reactions: Cardyak

Cardyak

Member
Sep 12, 2018
72
159
106
For example, core fusion was investigated in Intel after they bought out soft machines? was it? and eventually got mothballed internally after testing for a bit.

I’m still adamant this hasn’t been mothballed.

I have no concrete evidence of this. But I’m convinced core fusion is on the table sooner rather than later, and I wouldn’t be surprised if that’s what Royal Core’s “impressive changes” pertains to either.

I’ve had conversations with people that have second hand information regarding projects at Intel - and they insist core fusion and merging is being worked on to some degree - what that means for the product line-up is a real mystery.
 

Geddagod

Golden Member
Dec 28, 2021
1,147
1,003
106
I’m still adamant this hasn’t been mothballed.

I have no concrete evidence of this. But I’m convinced core fusion is on the table sooner rather than later, and I wouldn’t be surprised if that’s what Royal Core’s “impressive changes” pertains to either.

I’ve had conversations with people that have second hand information regarding projects at Intel - and they insist core fusion and merging is being worked on to some degree - what that means for the product line-up is a real mystery.
I would love that to be true, I believe it was Ian Cutress who claimed Intel didn't look like it was working on core fusion anymore. I'm too lazy to look up the actual post from which I'm recalling this from though, maybe once I finish some hw I'll edit my comment and link it down here lol
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,683
1,218
136
I would love that to be true, I believe it was Ian Cutress who claimed Intel didn't look like it was working on core fusion anymore. I'm too lazy to look up the actual post from which I'm recalling this from though, maybe once I finish some hw I'll edit my comment and link it down here lol

It should be noted that the timeline was:
1. Softmachines bought out, everything was scrapped but 2^x physical cores to one virtual core.
2. Get Softmachines used to using Intel Foundries (TSMC -> Intel)
[2016->2019--BKrza]
3. Softmachines started work on an American P-core/Core(tm) Project
4. All American P-core/Core(tm) projects killed
[2019->2020--BSwan]
5. The leftovers from the reckoning were moved to an Atom/E-core Project.
[2021+--PGels]

Pay attention to the distances in Intel3 and Intel20A, if they clear it for launch it will probably be 18A:
cresmontstuff.jpg

I believe this is from the P-core project(started at Intel was assisted by SoftMachines):
US10437590-20191008-D00000.png [Memory-Math-2xAVX512(4x256-bit;;in Client)-Math-Memory::2 P-corefusion]
 
Last edited:

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Do we have any idea what Ocean Cove actually entailed? I know it was a hugely ambitious project that was cancelled, but I still can’t confirm whether it was evolutionary (I.E: Intel’s standard approach of “Wider, Deeper, Smarter”) or whether it was a more revolutionary departure from the standard Core family that had brand new ideas.
Definitely the latter. At one point, it was called "Next Generation Core". Unfortunately, I doubt they'll ever acknowledge it publicly, much less talk about the one fundamental idea at its heart. And public leaks from that team seem non-existent.
I also don’t know where/if the ideas of Ocean Cove died and where the ideas for Royal Core originated from, are the lines blurred here? Is it possible Royal took some of the ideas from Ocean Cove and is essentially its spiritual successor? Or are they 2 completely different approaches to processor design and there is limited overlap between the two?
I think Royal took some learnings from Ocean Cove, or perhaps the Ocean Cove design process. A decent number of the original Royal team seem to have come from that project. That said, my understanding is that the main idea behind Ocean Cove was not carried over.

Ah, it will be a very fun day on these forums when Royal eventually hits the market.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136

It should be noted that the timeline was:
1. Softmachines bought out, everything was scrapped but 2^x physical cores to one virtual core.
2. Get Softmachines used to using Intel Foundries (TSMC -> Intel)
[2016->2019--BKrza]
3. Softmachines started work on an American P-core/Core(tm) Project
4. All American P-core/Core(tm) projects killed
[2019->2020--BSwan]
5. The leftovers from the reckoning were moved to an Atom/E-core Project.
[2021+--PGels]

Pay attention to the distances in Intel3 and Intel20A, if they clear it for launch it will probably be 18A:
View attachment 76442

I believe this is from the P-core project(started at Intel was assisted by SoftMachines):
View attachment 76443 [Math-Memory-2xAVX512(4x256-bit;;in Client)-Memory-Math::2 P-corefusion]
Goes without saying, but this "timeline" is nonsense.
 
  • Like
Reactions: IntelUser2000

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I'll respond to these assuming they're made in good faith. That is a very tenuous assumption.

Back atcha. Actually no, I'm just gonna cut it off here. You're a -mont fan and I'm not. You have faith in Intel's ability to execute with massive debt, shrinking revenue, and shrinking cash reserves. I don't. Good luck.

And just because some of the "Ridge" SoCs have accelerators doesn't mean they're primarily ASICs. They still have x86 cores in them which to the best of my knowledge still define the primary functions for those CPUs. If we ever see Sierra Forest-SP or -AP, we'll probably see accelerators in those products as well, just as we're seeing in Sapphire Rapids.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
Back atcha. Actually no, I'm just gonna cut it off here. You're a -mont fan and I'm not. You have faith in Intel's ability to execute with massive debt, shrinking revenue, and shrinking cash reserves. I don't. Good luck.

And just because some of the "Ridge" SoCs have accelerators doesn't mean they're primarily ASICs. They still have x86 cores in them which to the best of my knowledge still define the primary functions for those CPUs. If we ever see Sierra Forest-SP or -AP, we'll probably see accelerators in those products as well, just as we're seeing in Sapphire Rapids.
This is a matter of basic math and data, not feels. If you're fundamentally unwilling to acknowledge either, you're not arguing in good faith.
 
Last edited:

diediealldie

Member
May 9, 2020
77
68
61


I believe this is from the P-core project(started at Intel was assisted by SoftMachines):
View attachment 76443 [Memory-Math-2xAVX512(4x256-bit;;in Client)-Math-Memory::2 P-corefusion]

It looks AMD CMT processors with additional Inter-core resources to me. Will it really perform well in ST workloads compared to vanilla BF OOO cores? I agree if they are planning to mitigate ST degradation due to small core sizes, but not sure if it's really the way to boost ST performances.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
This is a matter of basic math and data, not feels. If you're fundamentally unwilling to acknowledge either, you're not arguing in good faith.

Again, BACK ATCHA. You're ignoring Intel's recent inability to execute and leaning on products that don't even exist yet, and may never on the broader market! You keep pumping up e-cores in contrast to cores that are known to have area and power-efficiency problems. There is no "arguing in bad faith".