Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

You have to be extremely bias not to see that Intel 10nm is in full swing.

Intel isn't selling any 10nm Xeons
Intel isn't selling any 10nm desktop/workstation chips

That's a lot of silicon area still going to 14nm.

 

[trexfromouterspace]

Intel isn't selling any 10nm Xeons

That's blatantly false. They've moved dozens of units. Dozens!

 

[DrMrLordX]

That's blatantly false. They've moved dozens of units. Dozens!

Hey engineering samples of unreleased silicon don't count!

Or do they? Maybe if BK were still there, they would . . .

 

[shady28]

Intel isn't selling any 10nm Xeons
Intel isn't selling any 10nm desktop/workstation chips

That's a lot of silicon area still going to 14nm.

I would bet that Intel is selling more 10nm silicon than AMD is selling 7nm silicon.

Do the math. The number are out there. Intel=80% of the client PC market, 60%+ of that 80% are laptops, so 48% of client sales are laptop chips.

If half the laptop chips they sell are 10nm (and from what I see, it's a lot *more* than half) then 24% of all client PC sales are 10nm laptop chips. And as noted before, these chips also show up in AIO PCs a lot, so there's more.

24% is more than AMDs market share for all types of chips, desktop and laptop. They are around 20% right now in the client PC space. And Intel is doing that with 3 - 4 of their 16 fabs.

So the post I was responding to that said 10nm was giving them trouble and they had bad yields seems to be debunked, attempts to shift goal posts notwithstanding.

 

[ondma]

I would bet that Intel is selling more 10nm silicon than AMD is selling 7nm silicon.

Do the math. The number are out there. Intel=80% of the client PC market, 60%+ of that 80% are laptops, so 48% of client sales are laptop chips.

If half the laptop chips they sell are 10nm (and from what I see, it's a lot *more* than half) then 24% of all client PC sales are 10nm laptop chips. And as noted before, these chips also show up in AIO PCs a lot, so there's more.

24% is more than AMDs market share for all types of chips, desktop and laptop. They are around 20% right now in the client PC space. And Intel is doing that with 3 - 4 of their 16 fabs.

So the post I was responding to that said 10nm was giving them trouble and they had bad yields seems to be debunked, attempts to shift goal posts notwithstanding.

That may be true, but the problem is the performance of those 10nm chips is mediocre at best compared to AMD 7nm, and little if any improvement, even to Intel's own 14nm. Tiger Lake will help to remedy that somewhat, but even there, it seems performance is all over the place, depending on implementation.

 

[shady28]

That may be true, but the problem is the performance of those 10nm chips is mediocre at best compared to AMD 7nm, and little if any improvement, even to Intel's own 14nm. Tiger Lake will help to remedy that somewhat, but even there, it seems performance is all over the place, depending on implementation.

Maybe, maybe not, I wasn't addressiong yet another performance argument, I was addressing the comment that Intel's 10nm was still having problems and there was no supply. There is, it's everywhere in the largest parts of the market, just not big bulky desktops.

The new Fab 42 went fully operational in October and was the 3rd 10nm capable fab, so they've actually done that with 2 fabs as supply from Fab 42 is probably just now starting to enter the market. i.e. they likely put out more 10nm than AMD/TSMC put out 7nm, and they did it with 2 10nm fabs.

As far as performance, you're talking about quality of the laptop builds. EVO certified TGL laptops kick butt. They cost too. If you go cheap, you get cheap. Same thing happens to AMD but there's no easy way to tell the difference.

 

[Hulk]

That's where your logic fails.

Look at top selling Intel laptops, and even a bunch of AIO desktops. They're running Ice Lake, which is on 10nm. And that is coming from I believe 3 fabs, maybe 4 I forget. Intel has I believe 16 fabs total. And those 4 fabs are not just making Ice Lake / Tiger Lake chips - they're also doing HPC stuff and have probably been making Ice Lake server chips for some time.

Given that laptops sell significantly more than desktops, and that Intel has 80-85% of both markets, I think it would be a safe bet that there are more Intel 10nm parts for sale than AMD 7nm parts at this time. It looks to me like Intel's 10nm fabs are all going full tilt boogie.

I'm not following. It seems like if 10nm had been going well for Intel then 10 or more of those fabs would have already transitioned to 10nm and Rocket Lake would be coming at 10nm instead of 14nm?

Rocket Lake two months on high yield 10SF would most likely have allowed Intel to go higher than 8 core and with lower thermals thus reclaimed all of the Zen 3 sales they are currently losing out on.

 

[ondma]

Maybe, maybe not, I wasn't addressiong yet another performance argument, I was addressing the comment that Intel's 10nm was still having problems and there was no supply. There is, it's everywhere in the largest parts of the market, just not big bulky desktops.

The new Fab 42 went fully operational in October and was the 3rd 10nm capable fab, so they've actually done that with 2 fabs as supply from Fab 42 is probably just now starting to enter the market. i.e. they likely put out more 10nm than AMD/TSMC put out 7nm, and they did it with 2 10nm fabs.

As far as performance, you're talking about quality of the laptop builds. EVO certified TGL laptops kick butt. They cost too. If you go cheap, you get cheap. Same thing happens to AMD but there's no easy way to tell the difference.

I was talking about Ice Lake performance. Only 4 cores, and hardly any improvement in performance (except for graphics, and TBH who cares) or battery life. And still only 4 cores for Ice Lake or Tiger Lake, although at least 4 core TL is a decent improvement over 4 core CL/IL. (Yes, I know 8 core TL is on the roadmap, but want to see it widely available before I am convinced.)

 

[RTX]

What's the reason for having the stacked PCBs in Skylake-SP/Cascadelake-SP/Cooperlake-SP/Icelake-SP?

 

[DrMrLordX]

Do the math. The number are out there. Intel=80% of the client PC market, 60%+ of that 80% are laptops, so 48% of client sales are laptop chips.

That's just laptops. TSMC is pushing out 7nm wafers for AMD in server, workstation, desktop, and mobile. Intel is selling 14nm in all of those segments except laptops, and not all of Intel's laptop sales are 10nm. They're still selling a lot of 14nm there too.

 

[LightningZ71]

You also fail to note the massive volume at the low end with their atom based products, which are all on 14nm now.

 

[DrMrLordX]

You also fail to note the massive volume at the low end with their atom based products, which are all on 14nm now.

Not entirely true. Intel has been selling 10nm Atom for awhile now, in the form of comm equipment and network appliances. In what volumes I don't know.

 

[shady28]

That's just laptops. TSMC is pushing out 7nm wafers for AMD in server, workstation, desktop, and mobile. Intel is selling 14nm in all of those segments except laptops, and not all of Intel's laptop sales are 10nm. They're still selling a lot of 14nm there too.

Client compute for those numbers is desktop + laptop/mobile. There's no such differentiation between workstation / desktop / mobile. It's all contained in the 80/20.

Server is a different area yes, but AMD only has like 3-5% there. While Intel revenue for server is nearly that of client compute, the unit sales in server will be far lower.

 

[uzzi38]

Is there a reliable source for that?

He's reliable - I can assure you that.

 

[shady28]

I'm not following. It seems like if 10nm had been going well for Intel then 10 or more of those fabs would have already transitioned to 10nm and Rocket Lake would be coming at 10nm instead of 14nm?

Rocket Lake two months on high yield 10SF would most likely have allowed Intel to go higher than 8 core and with lower thermals thus reclaimed all of the Zen 3 sales they are currently losing out on.

You can't lose many sales to something that isn't there. Outside of review sites and forums, scalpers and bot watchers, there's no Zen 3 to speak of. And to that point, 80% of AMDs 7nm production was slated for XBox X/S and PS5. 85% of the market in client is via 7 OEMs and not one of them has a Zen 3 based system. Zen 3 will probably become a real threat Q2 2021, right now it's a marketing move.

When you see one of the major OEMs like Dell, Acer, Lenovo, Asus, HP selling Zen 3 boxes you'll know it's arrived. It is important for mind share, but I think Intel knows that, and have things timed accordingly. That's why RKL will be there when Zen 3 isn't paper anymore.

So the big question will be, how will RKL perform, and how far up and down their product stack will it penetrate. That mindshare thing is actually pretty important too.

 

[DrMrLordX]

Client compute for those numbers is desktop + laptop/mobile. There's no such differentiation between workstation / desktop / mobile. It's all contained in the 80/20.

Then your numbers are meaningless since most of the "client compute" segment is still served by 14nm.

 

[shady28]

Then your numbers are meaningless since most of the "client compute" segment is still served by 14nm.

Not the laptop segment, which is what I was focused on. You don't seem to grasp what it means to have 80% of the market, and for laptops to be 60%+ of that market. That means Intel laptop is ~48%+ of the client market.

This information is freely available, and if anything I have under-estimated Intel's 10nm market penetration and done so intentionally. For example :

"Global laptop sales totaled $101.7 billion in 2017, with 161.6 million units shipped worldwide. "

"In 2019, approximately 261.24 million PCs were shipped around the world "

"The publication's sources believe that AMD could hit 20% market share in the notebook market in Q1. That would be quite the growth, considering AMD's mobile CPU market share was 14.7% in Q3 2019 "

161.6M / 261.2 = 61.7%.

So basically what I've done, is assume that AMD actually hit 20% (if they haven't, then there are even more Intel laptops out there). I've also used the laptop sales numbers for 2017, but laptop sales have been growing and desktop sales shrinking, so I am under estimating the % of the market that is laptops.

Now we don't know what % of that is 10nm, which I stated. To get some kind of feel, I look at the top 10 best selling laptops from multiple vendors and look at the intel platforms, it is easy to see that far more than 50% of the Intel boxes (which we know is 80%+ of laptop sales) have Ice Lake. It's more like 65%.

So I could be wrong yes, but most likely wrong in the other direction - i.e. there's probably a lot more 10nm than I'm estimating. But I think I'll take my numbers vs your opinion every day of the week.

 

[Det0x]

Hey engineering samples of unreleased silicon don't count!

Or do they? Maybe if BK were still there, they would . . .

Intel 10nm Xeon



Coming to the test results, the Intel Ice Lake-SP Xeon CPU scores 553.1 points in a single-core and 10038.4 points in the multi-core test. The single-core results put the Xeon chip neck-to-neck with a Core i9-10900K which is Intel's fastest gaming chip that boosts up to 5.3 GHz. This is impressive for an engineering sample that's running at vastly lower clock speeds. In multi-core tests, the Xeon-SP ES CPU is about as fast as the Ryzen 9 3950X which is a 16 core and 32 thread part. This should put the Xeon-SP CPU in the same ballpark as the Ryzen 9 5900X which is a 12 core and 24 thread part but with much higher multi-threaded performance over its predecessor thanks to the Zen 3 core architecture.

However, there's one thing to note that the latest CPU-z version was used which makes full use of AVX-512 and that is a major contributor to Intel's performance here. Without AVX-512 acceleration, the Intel Ice Lake-SP Xeon ES CPU is much slower, scoring just 371.6 points in single-core and 6363 points in multi-threaded tests. Other benchmarks include Fritz Chess where the chip scores 19715 points which is just slightly faster than a first-generation 8 core and 16 thread Ryzen CPU.

Intel 3rd Gen Ice Lake-SP Xeon CPU With 14 Cores & 28 Threads Benchmarked

Quite a few benchmarks of Intel's 3rd Gen Ice Lake-SP Xeon CPU have leaked out which show a 14 core and 28 thread server part.

wccftech.com

Seems like Intel is putting all their eggs on AVX-512 going forward

 

[shady28]

However, there's one thing to note that the latest CPU-z version was used which makes full use of AVX-512 and that is a major contributor to Intel's performance here. Without AVX-512 acceleration, the Intel Ice Lake-SP Xeon ES CPU is much slower, scoring just 371.6 points in single-core and 6363 points in multi-threaded tests. Other benchmarks include Fritz Chess where the chip scores 19715 points which is just slightly faster than a first-generation 8 core and 16 thread Ryzen CPU.

Intel 3rd Gen Ice Lake-SP Xeon CPU With 14 Cores & 28 Threads Benchmarked

Quite a few benchmarks of Intel's 3rd Gen Ice Lake-SP Xeon CPU have leaked out which show a 14 core and 28 thread server part.

wccftech.com

Article also says :

"Leaker YuuKi_AnS states that the chip has a base clock of 2.0 GHz and a boost clock of 4.0 GHz however the actual boost operating range is between 1.8 to 2.0 GHz when running all cores."

So it's doing that at <=2Ghz. This is an engineering sample, and it is very common for them to be greatly under-clocked. They aren't meant for benchmarks, they're meant for operating / compatibility testing.

And to wit, the AIDA screenshot actually shows it running at 1 Ghz.

So what happens if it is running at 3Ghz in production, or 4Ghz?

 

[Det0x]

So what happens if it is running at 3Ghz in production, or 4Ghz?

The socket get melted ?

Na iam just kidding, have no idea, but if i were to guess:

Maybe score 15057 AVX512-points in the mulithread cpuz benchmark, while scoring 9544 points in the same multithread benchmark without AVX.
Singlethread scores stay the ~same

 

[uzzi38]

Article also says :

"Leaker YuuKi_AnS states that the chip has a base clock of 2.0 GHz and a boost clock of 4.0 GHz however the actual boost operating range is between 1.8 to 2.0 GHz when running all cores."

So it's doing that at compatibility testing.

And to wit, the AIDA screenshot actually shows it running at 1 Ghz.

So what happens if it is running at 3Ghz in production, or 4Ghz?

Just a reminder that Ice Lake-U had an all core boost of 3.5GHz and required around 40W to sustain that if memory serves me correctly.

Intel have already clarified that Ice Lake SP does not use Superfin.

Take both into consideration when trying to set your expectations.

 

[trexfromouterspace]

Take both into consideration when trying to set your expectations.

The only variable in any reasonable person's expectations is the exact manner in which Milan is going to murder Ice Lake SP.

Will it be drawn and quartered? Guillotined? Electric Chair? Any way you slice it Ice Lake SP is kill.

 

[Hulk]

You can't lose many sales to something that isn't there. Outside of review sites and forums, scalpers and bot watchers, there's no Zen 3 to speak of. And to that point, 80% of AMDs 7nm production was slated for XBox X/S and PS5. 85% of the market in client is via 7 OEMs and not one of them has a Zen 3 based system. Zen 3 will probably become a real threat Q2 2021, right now it's a marketing move.

When you see one of the major OEMs like Dell, Acer, Lenovo, Asus, HP selling Zen 3 boxes you'll know it's arrived. It is important for mind share, but I think Intel knows that, and have things timed accordingly. That's why RKL will be there when Zen 3 isn't paper anymore.

So the big question will be, how will RKL perform, and how far up and down their product stack will it penetrate. That mindshare thing is actually pretty important too.

Are you saying Intel isn't losing any significant sales to Zen 3? Seems like AMD is selling every Zen 3 part they can get to market and I don't think it's farfetched to believe many of those people would have bought Comet Lake if Zen 3 was not available. Not to mention all of the people upgrading to Zen 3 from Zen/2. I don't know the exact numbers but the reality and perception of these sales over Comet Lakes can't be a good thing for Intel.

It's like Intel has a man (or two) in the penalty box until RL arrives and they are certainly putting on the pressure.

 

[shady28]

Are you saying Intel isn't losing any significant sales to Zen 3? Seems like AMD is selling every Zen 3 part they can get to market and I don't think it's farfetched to believe many of those people would have bought Comet Lake if Zen 3 was not available. Not to mention all of the people upgrading to Zen 3 from Zen/2. I don't know the exact numbers but the reality and perception of these sales over Comet Lakes can't be a good thing for Intel.

It's like Intel has a man (or two) in the penalty box until RL arrives and they are certainly putting on the pressure.

Intel hasn't lost squat to Zen 3, because it isn't there in any quantity and won't be until sometime in Q1. Leftover Zen 2 is probably doing 10x the damage to Intel that Zen 3 is. You can't sell what you don't have.

 

[Hulk]

Intel hasn't lost squat to Zen 3, because it isn't there in any quantity and won't be until sometime in Q1. Leftover Zen 2 is probably doing 10x the damage to Intel that Zen 3 is. You can't sell what you don't have.

Okay then. We'll have to see where things are in 6 months but it does seem like AMD has a foothold in a place they haven't been before with Zen 3.