Discussion Intel current and future Lakes & Rapids thread

[inf64]

Intel certainly needs more than 8C if they want to compete with Zen 3 12-16C in MT workloads and HEDT is basically dead, at least for Intel.

It's obvious that intel completely conceded 12C+ market to AMD, they don't even talk about it anymore (as they have nothing to show for). The first good bet would be Sapphire Rapids HEDT platform if/when it arrives. Bad news is that intel will not be able to command the highest prices for "high" core counts any more, they have extremely difficult competition in even previous gen TR 3000. This is the reason why we don't hear anything about TR 4000, there is no point really.

 

[Zucker2k]

Phoronix Tiger Lake Linux Performance Test

 

[Markfw]

Phoronix Tiger Lake Linux Performance Test

From your link...

"So there are some healthy improvements to find with the Core i7 1165G7, but in many cases that still puts this high-end Tiger Lake notebook CPU behind the Ryzen 5 4500U and Ryzen 7 4700U. The context switching performance was also slower on Tiger Lake than Ice Lake. It would be interesting to see how it goes against the Ryzen 4800 series but unfortunately lack of hardware access at the moment and even in most cases the Ryzen 5 4500U with six physical cores was performing faster on Ubuntu Linux than the more expensive Dell XPS 13 9310. "

 

[Zucker2k]

It's obvious that intel completely conceded 12C+ market to AMD, they don't even talk about it anymore (as they have nothing to show for). The first good bet would be Sapphire Rapids HEDT platform if/when it arrives. Bad news is that intel will not be able to command the highest prices for "high" core counts any more, they have extremely difficult competition in even previous gen TR 3000. This is the reason why we don't hear anything about TR 4000, there is no point really.

In my book, chips priced above $500 do not belong on "mainstream consumer platform," or non-HEDT platform. 16 cores on this platform was just AMD trying to establish some decisive leadership in an area of desktop Intel could never compete in without introducing their own HEDT chips into the mix. I don't know why this seems oblivious to some - the 3950x is the successor to the 2950x. And now, we have a 5950x at $850, on desktop!
I really hope Intel doesn't follow AMD into this dark hole of throughput dominancy, or better yet hollow desktop supremacy because there's nothing mainstream about the price. At this rate, we'll be looking at $1,000 "mainstream" cpus in the near future. Heck, it might even be too late for that because if these chips are released in limited quantities I'll bet my last dollar scalpers are going to be selling them at a grand a pop.

 

[mikk]

Yea. 8+8 may compete with 12 core Zen, but Intel has no answer for 16 core.

You cannot know this. I can see ADL 8+8 winning against Zen 3 if Golden Cove brings a big IPC increase over Sunny Cove and clock speeds are high.

 

[Markfw]

In my book, chips priced above $500 do not belong on "mainstream consumer platform," or non-HEDT platform. 16 cores on this platform was just AMD trying to establish some decisive leadership in an area of desktop Intel could never compete in without introducing their own HEDT chips into the mix. I don't know why this seems oblivious to some - the 3950x is the successor to the 2950x. And now, we have a 5950x at $850, on desktop!
I really hope Intel doesn't follow AMD into this dark hole of throughput dominancy, or better yet hollow desktop supremacy because there's nothing mainstream about the price. At this rate, we'll be looking at $1,000 "mainstream" cpus in the near future. Heck, it might even be too late for that because if these chips are released in limited quantities I'll bet my last dollar scalpers are going to be selling them at a grand a pop.

Its $799, not $850. And the 2950x is HEDT, as it has 4 memory channels and more PCIE lanes.

 

[Zucker2k]

Its $799, not $850. And the 2950x is HEDT, as it has 4 memory channels and more PCIE lanes.

My bad on the pricing. I know the 3950x is dual channel. You can't run quad channel on AM4 though, can you?

 

[Markfw]

My bad on the pricing. I know the 3950x is dual channel. You can't run quad channel on AM4 though, can you?

No. It is a little confusing, I admit.2950 and up are HEDT 3950 and down are desktop.

Edit, and I think on topic.... 800 and down are desktop, 801 and up are HEDT. At least thes the AMD breakpoint. Not sure what Intel has, I think socket 2066 and 2011-3 are their hedt ??

 

[jj109]

Phoronix Tiger Lake Linux Performance Test

Looks like it's capped at ~3GHz on Kraken and Speedometer. That's a straight up busted power management firmware. No reason why it should be doing worse than Ice Lake.

Power numbers in Windows from another reviewer

15W sustained vs 24W for the Ice Lake version is also baffling. The chassis shouldn't have changed.
Perf/W at 15W isn't bad at all when accounting for actual sustained power of competitors; just needs more cores to scale up.

 

[TheGiant]

From your link...

"So there are some healthy improvements to find with the Core i7 1165G7, but in many cases that still puts this high-end Tiger Lake notebook CPU behind the Ryzen 5 4500U and Ryzen 7 4700U. The context switching performance was also slower on Tiger Lake than Ice Lake. It would be interesting to see how it goes against the Ryzen 4800 series but unfortunately lack of hardware access at the moment and even in most cases the Ryzen 5 4500U with six physical cores was performing faster on Ubuntu Linux than the more expensive Dell XPS 13 9310. "

the is something wrong with the laptop, look at the frequency max

like web browsing here

Intel Core i7 1165G7 "Tiger Lake" Linux Performance With The Dell XPS 13 9310 Review - Phoronix

www.phoronix.com

phoronix should't even publish this, looks like f..d up bios or what

 

[ondma]

O

You cannot know this. I can see ADL 8+8 winning against Zen 3 if Golden Cove brings a big IPC increase over Sunny Cove and clock speeds are high.

Obviously, we have to wait solid data. However, here is some off the cuff math. Assume (optimistically IMO), that Golden cove is 20% better IPC than Zen 3. So 8 big cores is 9.6 equivalent Zen cores. Add another 10% (optimistically) for clockspeed = 10.5 cores. Assume 8 atom cores is equivalent to 4 Zen cores (again optimistic IMO) and you get about 14 "zen equivalents". And that is also assuming tasks can be split up to perfectly utilize both big and small cores to the max at the same time. Since it seems unlikely all of the optimistic estimates will hold true, I am expecting 8+8 at best to be competitive with 12 core Zen 3, which actually would not be bad if it were out now, but it will probably be at least another year.

 

[eek2121]

O

Obviously, we have to wait solid data. However, here is some off the cuff math. Assume (optimistically IMO), that Golden cove is 20% better IPC than Zen 3. So 8 big cores is 9.6 equivalent Zen cores. Add another 10% (optimistically) for clockspeed = 10.5 cores. Assume 8 atom cores is equivalent to 4 Zen cores (again optimistic IMO) and you get about 14 "zen equivalents". And that is also assuming tasks can be split up to perfectly utilize both big and small cores to the max at the same time. Since it seems unlikely all of the optimistic estimates will hold true, I am expecting 8+8 at best to be competitive with 12 core Zen 3, which actually would not be bad if it were out now, but it will probably be at least another year.

While I do lean AMD these days I don’t buy that. I suspect that the Gracemont cores exceed Coffee Lake in IPC and the Golden Cove cores are something else...

 

[Gideon]

While I do lean AMD these days I don’t buy that. I suspect that the Gracemont cores exceed Coffee Lake in IPC and the Golden Cove cores are something else...

The problem with Gracemont in those workloads is that it lacks SMT (or even better, 16 atom cores instead of 8). In rendering workloads that immediatelly means ~40% perfomance loss.

In all likely-hood it also won't quite scale to ~4Ghz (as that costs die-area, which would be strange to waste for desktop-only frequency for little cores).

It would still be a good boost (like adding 2x i5 6600 at ~3.x Ghz to the score) but adding all of this together, even with IPC of around Skylake 8 gracemont cores will still be >=50% slower than 8 zen 3 cores in such extremely MT workloads.

Both of these mean that Golden Cove would have to be ridiculously good to make up for that.

 

[naukkis]

The problem with Gracemont in those workloads is that it lacks SMT (or even better, 16 atom cores instead of 8). In rendering workloads that immediatelly means ~40% perfomance loss.

In all likely-hood it also won't quite scale to ~4Ghz (as that costs die-area, which would be strange to waste for desktop-only frequency for little cores).

It would still be a good boost (like adding 2x i5 6600 at ~3.x Ghz to the score) but adding all of this together, even with IPC of around Skylake 8 gracemont cores will still be >=50% slower than 8 zen 3 cores in such extremely MT workloads.

In those exteme MT-loads Zen3 will operate at it's base clock ~3.4Ghz at TDP power level. It might be that Gracemont cores will operate at higher frequency at those tasks.

 

[coercitiv]

In those exteme MT-loads Zen3 will operate at it's base clock ~3.4Ghz at TDP power level. It might be that Gracemont cores will operate at higher frequency at those tasks.

In those extreme MT loads Alder Lake will need ~7% higher clocks to match Zen 3 throughput. It will likely need 10% higher clocks for a win, which means 3.8Ghz all-core or 4Ghz on the Golden Cove cluster and 3.5Ghz on the Gracemont cluster. That's assuming GC is 1.5X, Gracemont is 1x and Zen 3 is 1.25x, with 1.2x SMT scaling.

We're also 1 year away from Alder Lake and AMD will have something better by then. Even if it's just a minor 7% performance gain, that's another jump in clocks the Golden Cove cluster needs to do within TDP to match or beat the competition.

Gracemont being more efficient is a trap, it tempts us to ignore the fact that the Golden Cove cluster will also lose efficiency as it clocks higher.

 

[jpiniero]

Nvidia and EuroHPC Team for Four Supercomputers, Including Massive 'Leonardo' System

The EuroHPC Joint Undertaking (JU) serves as Europe’s concerted supercomputing play, currently comprising 32 member states and billions of euros in funding. In June 2019, EuroHPC selected eight supercomputing centers […]

www.hpcwire.com

FWIW nVidia announced a new HPC deal with the European Union that goes back to using an Intel CPU. Assuming it's Sapphire Rapids, that would mean that it would be available by the end of next year.

 

[mikk]

Assume 8 atom cores is equivalent to 4 Zen cores (again optimistic IMO) and you get about 14 "zen equivalents"

According to Raichu Gracemont has a little better IPC than Skylake, this is basically Zen 2 IPC. A big factor is Intels process node, it also depends on 10ESF how much it's improved over 10SF.

 

[mikk]

Power numbers in Windows from another reviewer

15W sustained vs 24W for the Ice Lake version is also baffling. The chassis shouldn't have changed.
Perf/W at 15W isn't bad at all when accounting for actual sustained power of competitors; just needs more cores to scale up.

This is super bad ouch, even the Firestrike scores are lower than on Acer Swift and even Asus Zenbook. This is super bad for a premium device like this. However even this review is better than Phoronix because it didn't boost over 3100 Mhz there. MSI Prestige will smash Dell XPS when it comes to performance.

 

[DrMrLordX]

When I was still at Intel, Sapphire Rapids was supposed to be a from-scratch design. The fact that it has become a small evolution is quite telling of Intel's design woes.

I'd like to chip in that I recall such a transition occurring from early Intel roadmaps featuring Sapphire Rapids to what we've heard about it lately.

Sapphire Rapids was, allegedly, going to be a clean-sheet departure from the Core series. This idea was posited before Intel began publicly referring to the Cove cores, or before anyone had suggested what Sapphire Rapids was just Willow Cove or Golden Cove plus some packaging technology to get core counts up (since Intel can't seem to make large monolithic dice on 10nm). It would've amounted to a complete departure from anything Intel had (or would have) sold on the desktop. That project seems to be dead.

 

[jpiniero]

Not saying that's what happened but I wouldn't be surprised if there was a project at Intel that Smeltdown basically made infeasible.

 

[yuri69]

Sapphire Rapids was, allegedly, going to be a clean-sheet departure from the Core series. This idea was posited before Intel began publicly referring to the Cove cores, or before anyone had suggested what Sapphire Rapids was just Willow Cove or Golden Cove plus some packaging technology to get core counts up (since Intel can't seem to make large monolithic dice on 10nm). It would've amounted to a complete departure from anything Intel had (or would have) sold on the desktop. That project seems to be dead.

Yea, there was that old leak revealing the existence of all the 10 and 7nm architectures (Sapphire, Granite, Diamond, and Meteor) some time in 2017(?). The "Rapids" being a departure from "Lakes" were hinted IIRC. Sapphire is gonna be the first one - it will feature advanced packaging tech given there is a SKU (a HPC one?) featuring HBM memory. How advanced will be the core architecture is ... yet unknown.

Later we got the Ocean Cove being a "new revolutionary" design targeted at 2023 time frame, just weeks after the new famous Senior Vice President of Intel Silicon Engineering Group was hired.

So one of those should probably bring a rather inventive/revolutionary/risky architectural approach.

 

[Exist50]

Later we got the Ocean Cove being a "new revolutionary" design targeted at 2023 time frame, just weeks after the new famous Senior Vice President of Intel Silicon Engineering Group was hired.

I've mentioned this before, but Ocean Cove (the revolutionary design) was cancelled and replaced with something far more evolutionary. That may be what he was referring to, though Sapphire Rapids would be the wrong context.

That also predates Keller.

 

[yuri69]

I've mentioned this before, but Ocean Cove (the revolutionary design) was cancelled and replaced with something far more evolutionary. That may be what he was referring to, though Sapphire Rapids would be the wrong context.

Sad to hear this. Revolutionary and ambitious designs are always a great topic, the evolutionary ones not so much :]

I'm still looking forward to both Golden Cove and Sapphire Rapids - these still *might* be revolutionary.

 

[IntelUser2000]

Sad to hear this. Revolutionary and ambitious designs are always a great topic, the evolutionary ones not so much :]

I'm still looking forward to both Golden Cove and Sapphire Rapids - these still *might* be revolutionary.

What does this "revolutionary" thing mean?

Because it doesn't mean faster. Actually the Pentium 4 was revolutionary. Too much change is actually not good.

20% gain(19%, bleh!) with Zen 3 is getting there. So we should be asking are we looking at Willow Cove gains or Sunny Cove/Zen 3 gains?

15W sustained vs 24W for the Ice Lake version is also baffling. The chassis shouldn't have changed.

The power numbers don't mean much without more detail on Tigerlake. Tigerlake can change TDP depending on scenario.

Although I agree the numbers could be a lot better at least in Cinebench 1T. It gets 470 while other TGL laptops get near 600.

 

[mikk]

Some Cinebench R20 power/performance numbers from the Dell XPS 13....

Dell XPS 13:

Dell XPS 13:

Dell XPS 13 2-in-1: https://youtu.be/XMNuwtLFYeo?t=592


The 2 in 1 version performs better and in ultra perfomance mode it runs with 35W sustained power for some time. But in general the limiting factor is cooling/heat on these devices.