Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

Anantech speculated that it had 12 LLC/Ring Stops because when Intel Disabled one of the quad cluster in the 12700k the CPU lost 5 MiB of L2 Capacity instead of losing just 3

Little to do with technical details and more to do with positioning. 12900K is an i9. 12700K is an i7. Since quads of E cores all share one L2 cache it doesn't make sense it's more than one stop.

 

[nicalandia]

Little to do with technical details and more to do with positioning. 12900K is an i9. 12700K is an i7. Since quads of E cores all share one L2 cache it doesn't make sense it's more than one stop.

Sorry, I meant 5 MiB L3 Capacity was lost by disabling one of the cluster.

 

[nicalandia]

So DDR5 Optane is dead for the Upcoming Sapphire Rapids-X for High End Workstations?

 

[biostud]

Cache is the simplest product to validate on a new process. Just seems natural to scale cache products up to justify getting a process out the door at the earliest stage. Starts your path to the black a bit sooner than conventional development models.

What I hear you say is 3nm vcache for zen5 is already in production.

 

[witeken]

I'm clearly wasting my time responding, but as covered here several times over, TSMC never once publicly promised that N3 would be on the same schedule as N5, and the current timeline is perfect in keeping with what they have said.

That's a gross misrepresentation of what I actually said. My previous post contained only facts:

*TSMC announced N3 in 2020 for H2'22 HVM
*However, in 2021 during a Q&A TSMC more precisely specified its schedule as Q3'22
*This has since shifted to (late) Q4'22

Hence, the statement that TSMC is "[perfectly] keeping with what they have said" is wrong.

The changed Meteor Lake plans (its delay and seemingly even backport of the GPU tile from N3 to N4/5) further supports the N3 delay observation.

 

[Heartbreaker]

The changed Meteor Lake plans (its delay and seemingly even backport of the GPU tile from N3 to N4/5) further supports the N3 delay observation.

You are essentially stacking two sets of rumors with this one, to claim something has gone wrong.

Two wrongs (or Rumors) don't make a right.

Intel has said there are no Meteor Lake delays, so you really can't use rumors claiming that as some kind of "fact" that proves something going wrong behind the scenes.

 

[lobz]

You are essentially stacking two sets of rumors with this one, to claim something has gone wrong.

Two wrongs (or Rumors) don't make a right.

Intel has said there are no Meteor Lake delays, so you really can't use rumors claiming that as some kind of "fact" that proves something going wrong behind the scenes.

The moment, when basic mathematics lose its validity and reality+basic logic kicks in.

 

[dullard]

The changed Meteor Lake plans (its delay and seemingly even backport of the GPU tile from N3 to N4/5) further supports the N3 delay observation.

The key that you are leaving out is any shred of evidence that Meteor Lake was ever using N3. Care to share that? The closest thing that we have to being official from Intel is that the N3 orders in your link were for Arrow Lake pre-production chips (engineering samples and the like). Even your link says it isn't clear what chips would use N3.

 

[nicalandia]

I Was able to find this Jewell... A Release Sample of the Xeon Platinum 8470 On a 2S system(104C/208T Total) On SiSoftware Official Benchmark Ranker, I am surprised that Wccftech have not posted yet.

Details for Result ID 2x Intel(R) Xeon(R) Platinum 8470 CPU @2.00GHz (52C 104T 2GHz/3.8GHz, 52x 2MB L2, 105MB L3)

ranker.sisoftware.co.uk

It's a performance improvement over the previous 8380 Ice Lake

Details for Result ID 2x Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz (40C 80T 3GHz/3.4GHz, 2.4GHz IMC, 40x 1.25MB L2, 60MB L3)

ranker.sisoftware.co.uk

 

[shady28]

That's a gross misrepresentation of what I actually said. My previous post contained only facts:

*TSMC announced N3 in 2020 for H2'22 HVM
*However, in 2021 during a Q&A TSMC more precisely specified its schedule as Q3'22
*This has since shifted to (late) Q4'22

Hence, the statement that TSMC is "[perfectly] keeping with what they have said" is wrong.

The changed Meteor Lake plans (its delay and seemingly even backport of the GPU tile from N3 to N4/5) further supports the N3 delay observation.

Pretty much everyone knows that TSMC hosed its schedule, except people on forums at enthusiast sites It's that cognitive dissonance thing at work.

As it currently stands, Intel may well be the only x86 vendor to ship sub 5nm CPUs in 2023.

Intel has not had a measurable node advantage since 2018 when Zen 1+ was introduced on GloFlo 12nm (roughly equivalent to Intel 14nm, though I would argue Intel 14nm was marginally better).

Given TSMCs previous 3 year lead, this does not happen without them flubbing. Or at least, being slow.

This does not mean, of course, that Intel cannot screw it up again. We'll see.

 

[LightningZ71]

I Was able to find this Jewell... A Release Sample of the Xeon Platinum 8470 On a 2S system(104C/208T Total) On SiSoftware Official Benchmark Ranker, I am surprised that Wccftech have not posted yet.

View attachment 65657

Details for Result ID 2x Intel(R) Xeon(R) Platinum 8470 CPU @2.00GHz (52C 104T 2GHz/3.8GHz, 52x 2MB L2, 105MB L3)

ranker.sisoftware.co.uk

It's a performance improvement over the previous 8380 Ice Lake

View attachment 65660

Details for Result ID 2x Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz (40C 80T 3GHz/3.4GHz, 2.4GHz IMC, 40x 1.25MB L2, 60MB L3)

ranker.sisoftware.co.uk

So, if I'm understanding the results properly: The new Xeon, with a roughly 25% core advantage, a 10% clock speed advantage, almost double the cache, larger L2 and likely a better memory subsystem only managed to get roughly 10% more "GOPS"?

I'm not very impressed... I hope this is still not running it's "final production tune".

 

[nicalandia]

So, if I'm understanding the results properly: The new Xeon, with a roughly 25% core advantage, a 10% clock speed advantage, almost double the cache, larger L2 and likely a better memory subsystem only managed to get roughly 10% more "GOPS"?

I'm not very impressed... I hope this is still not running it's "final production tune".

The Ice Lake Xeons were boosting 45% Higher on the benchmark(2S System), The Average Speed Efficiency on the Sapphire Rapids much higher.

Ice Lake Xeons 8380



Sapphire Rapids Xeons 8470

 

[KompuKare]

I Was able to find this Jewell... A Release Sample of the Xeon Platinum 8470 On a 2S system(104C/208T Total) On SiSoftware Official Benchmark Ranker, I am surprised that Wccftech have not posted yet.

View attachment 65657

Details for Result ID 2x Intel(R) Xeon(R) Platinum 8470 CPU @2.00GHz (52C 104T 2GHz/3.8GHz, 52x 2MB L2, 105MB L3)

ranker.sisoftware.co.uk

It's a performance improvement over the previous 8380 Ice Lake

View attachment 65660

Details for Result ID 2x Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz (40C 80T 3GHz/3.4GHz, 2.4GHz IMC, 40x 1.25MB L2, 60MB L3)

ranker.sisoftware.co.uk

If I'm reading that correctly, isn't that aboost in FP but a regression in integer?

 

[jpiniero]

The key that you are leaving out is any shred of evidence that Meteor Lake was ever using N3. Care to share that? The closest thing that we have to being official from Intel is that the N3 orders in your link were for Arrow Lake pre-production chips (engineering samples and the like). Even your link says it isn't clear what chips would use N3.

Goes back to the rumor that the IGP was going to be 192 EUs.

Intel has said there are no Meteor Lake delays, so you really can't use rumors claiming that as some kind of "fact" that proves something going wrong behind the scenes.

Intel only really said that products would be in stores in 2023. You can see why they wouldn't want to use N3 for the IGP if Meteor Lake is going to end up being mostly Celerons.

 

[nicalandia]

If I'm reading that correctly, isn't that aboost in FP but a regression in integer?

Yes, Sunny Cove has better Dhrystone Long performance over Golden Cove.

 

[Heartbreaker]

Intel only really said that products would be in stores in 2023. You can see why they wouldn't want to use N3 for the IGP if Meteor Lake is going to end up being mostly Celerons.

The Trendforce rumor claimed Intel was delaying Meteor Lake and cancelling N3 orders, and TSMC was slowing its planned expansion.

Both Intel and TSMC, publicly debunked their respective sections of that rumor.

So why would you still believe any of it, and look for reasons to try and make it still work??

 

[Doug S]

As it currently stands, Intel may well be the only x86 vendor to ship sub 5nm CPUs in 2023.

AMD has not exactly been an early adopter of TSMC's processes, so I doubt many were expecting N3 CPUs from them in 2023 regardless of whether they started N3 mass production in Q3 or Q4 of this year.

 

[Joe NYC]

AMD has not exactly been an early adopter of TSMC's processes, so I doubt many were expecting N3 CPUs from them in 2023 regardless of whether they started N3 mass production in Q3 or Q4 of this year.

I think it is quite likely AMD is skipping N3 and jumping straight to N3E.

N3E timing is ~ H2 2023 production, which may align quite well with Zen 5 launch. TSMC said that they have a lot of "customer engagement" for that node, so it may include AMD, which is going to be 2nd largest customer of TSMC.

 

[nicalandia]

I found another Jewel(albeit an early ES sample)

Intel Xeon W9 HEDT Sapphire Rapids Preview & Benchmarks

Intel Xeon W9 HEDT Sapphire Rapids Preview & Benchmarks – SiSoftware

www.sisoftware.co.uk

Edit. This was just posted a few hours ago(Today) Final Release Product but it's the Xeon Platinum 8473 S1 not the Xeon W9 HEDT. But we are getting final results now.

Details for Result ID Intel(R) Xeon(R) Platinum 8473C (52C 104T 2.9GHz, 2.5GHz IMC, 52x 2MB L2, 105MB L3)

ranker.sisoftware.co.uk

 

[Markfw]

I found another Jewel(albeit an early ES sample)

Intel Xeon W9 HEDT Sapphire Rapids Preview & Benchmarks

Intel Xeon W9 HEDT Sapphire Rapids Preview & Benchmarks – SiSoftware

www.sisoftware.co.uk

While it is dated 27 July, wasn't the 5995wx out by then ? Thats only 2 weeks ago. They say in the notes "the latest tech" but the 5995wx easily beats the 3990wx. And the 3990wx for the most part beats the new X9 HEDT. The 5995wx could be a slaughter. (as way less wattage( 280 vs 500)

 

[pakotlar]

So, if I'm understanding the results properly: The new Xeon, with a roughly 25% core advantage, a 10% clock speed advantage, almost double the cache, larger L2 and likely a better memory subsystem only managed to get roughly 10% more "GOPS"?

I'm not very impressed... I hope this is still not running it's "final production tune".

Look at your statement. It is clearly not production performance.

 

[nicalandia]

Look at your statement. It is clearly not production performance.

While it is dated 27 July, wasn't the 5995wx out by then?

Yes, but it has not been tested on the Sisoft Benchmark.

I just update the post with just released today for the final 8473C which is a 52C/104T

Details for Result ID Intel(R) Xeon(R) Platinum 8473C (52C 104T 2.9GHz, 2.5GHz IMC, 52x 2MB L2, 105MB L3)

ranker.sisoftware.co.uk

So far these are the Arithmetic Native GOPS Points.

Threadripper PRO 3995WX: 1300

Xeon Platinum 8473C: 1370

Intel Xeon W9-3495X : 1470

 

[Doug S]

I think it is quite likely AMD is skipping N3 and jumping straight to N3E.

N3E timing is ~ H2 2023 production, which may align quite well with Zen 5 launch. TSMC said that they have a lot of "customer engagement" for that node, so it may include AMD, which is going to be 2nd largest customer of TSMC.

N3E looks to be lining up perfectly for Apple for next year's iPhone, so there aren't going to be very many wafers left over for others until late in the year. They always say they have a lot of "customer engagement" but that doesn't mean a lot of customers who will be taking large numbers of early production wafers.

 

[Joe NYC]

N3E looks to be lining up perfectly for Apple for next year's iPhone, so there aren't going to be very many wafers left over for others until late in the year. They always say they have a lot of "customer engagement" but that doesn't mean a lot of customers who will be taking large numbers of early production wafers.

H2 2023 launch of N3E would be too late for Apple 2023 model iPhone. There was some other article that mentioned that Apple will wait until the 2024 iPhone model to adopt this node.

So, without Apple using it for its major product, there should be capacity for HPC customers.

 

[DrMrLordX]

As it currently stands, Intel may well be the only x86 vendor to ship sub 5nm CPUs in 2023.

Since Intel 4 is a 7nm process (which Intel itself admitted when they called it 7nm), that would be a big "nope". It's not clear what is a sub-5nm process since Samsung and TSMC don't even use nm naming for their nodes anymore. But. Calling Intel 4 "sub 5nm" is patently false.