Discussion Intel current and future Lakes & Rapids thread

[Markfw]

Sapphire Rapids might end up being 80-100% faster than the predecessor Icelake-SP. ICL has issues clocking high and it regressed noticeably compared to Cascade Lake. SPR will fix this plus clock higher than Cascade Lake.

20% uarch gains, 40% more cores, 10-15% higher clocks is going to be a substantial gain.

It's interesting how if you look from that perspective, it only takes a year before the competitive landscape changes. If they got this last year Intel would have had a competitive Xeon as the above should result in beating Milan by a bit in quite a few applications.

It'll be quite a contest in consumer desktop. Raptorlake versus Zen 4. 8 additional E cores should result in 20% gain, with faster P cores resulting in an overall 25% gain in MT. Golden Cove is what, ~10% faster than Zen 3 at equal clocks? Zen 4 might be faster per clock by a bit but Intel clocks higher so that cancels out.

V-Cache and HBM is going to be a situational thing. Neither will be widely available nor implemented across the board.

Milan-X is already shipping, and Genoa is sampling I think. Sapphire Rapids is not even sampling yet, how could you possibly think they can compete ? and on desktop, ADL is barely competitive, and not in perf/watt, so what makes you think is will be interesting there either ? Especially when Vermeer-X (or whatever) comes out ? Exist50 even agrees.

 

[Exist50]

Milan-X is already shipping, and Genoa is sampling I think. Sapphire Rapids is not even sampling yet, how could you possibly think they can compete ? and on desktop, ADL is barely competitive, and not in perf/watt, so what makes you think is will be interesting there either ? Especially when Vermeer-X (or whatever) comes out ? Exist50 even agrees.

SPR must have been sampling for some time now. We've seen ES in the wild before. And didn't realize I'd been put in a camp...

 

[Markfw]

SPR must have been sampling for some time now. We've seen ES in the wild before. And didn't realize I'd been put in a camp...

I had not seen that, but I still think SPR can't touch Milan-X let alone Genoa. And I did not put you in a camp, neither am I, just agreeing with:
"Meanwhile, Zen 4 will have a very hefty process advantage in addition to a step forward in uarch. I think it'll be a pretty clean sweep in AMD's favor. "

 

[Exist50]

I had not seen that, but I still think SPR can't touch Milan-X let alone Genoa. And I did not put you in a camp, neither am I, just agreeing with:
"Meanwhile, Zen 4 will have a very hefty process advantage in addition to a step forward in uarch. I think it'll be a pretty clean sweep in AMD's favor. "

Sure, agreed with SPR vs Genoa, though Milan X will be relatively low volume. Hope we can get to the point where it becomes common, however, but last I heard, the equipment hits volume somewhere in the 2023-ish timeframe.

 

[tomatosummit]

I had not seen that, but I still think SPR can't touch Milan-X let alone Genoa. And I did not put you in a camp, neither am I, just agreeing with:
"Meanwhile, Zen 4 will have a very hefty process advantage in addition to a step forward in uarch. I think it'll be a pretty clean sweep in AMD's favor. "

STH showed a few sapphire systems in their coverage of the super computing summit. I suspect all the usual suspects have spr already, My guess would be hyperscalers are swinging more towards amd for early samples but intel will still have a lot of clout with the channel vendors so can have a nice reveal with 400 different systems available on launch.

I don't think spr+hbm has been seen yet?
It looks like it has a different package which should mean a different socket, I can only see that being trouble.

 

[IntelUser2000]

Sapphire Rapids will lose significantly to Genoa but AMD can not rest either. The difference is merely a year. I suspect they'll have about a quarter advantage before Genoa.

With Raptor Lake that's why I said 20% from the E cores, rather than the 30%+ since it'll have to clock lower. Although I don't think Raptor Lake will remain just with more E cores plus a modified Golden Cove core. Even refreshes get slight boosts.

Circuitry refinements and additional year from Intel 7 will allow them to squeeze little more out of it. I could attribute it to DLVR, but I personally believe it's a mobile thing.

Don't underestimate them, and also know that any "entity" that doesn't die, be it a person or a company comes back. Things are at the lowest point before it swings back.

 

[Accord99]

And I think you're either overestimating Raptor Lake, or underestimating Zen 4. The bonus from +8 E cores will be blunted by power limits, and who knows what we'll actually see from what's effectively a Golden Cove refresh.

You get the watts from running the 8P cores one bin lower, a couple of percent loss in MT performance more than made up by the 8E cores.

 

[IntelUser2000]

I assume there's also the fact that with Raptor Lake they would design it with desktops in mind as well, unlike Alder Lake which seems like a last minute approach.

The naming is funny.

Raptor-Lake
Meteor-Lake

Back when I was hanging around Intel investor sites one guy made a comment that they have humor when they went from SILVERmont to GOLDmont.

 

[DrMrLordX]

The bonus from +8 E cores will be blunted by power limits,

They'll be blunted by Amdahl's Law.

Meanwhile, Zen 4 will have a very hefty process advantage in addition to a step forward in uarch. I think it'll be a pretty clean sweep in AMD's favor.

Agreed, Raptor Lake will struggle against Raphael.

Sapphire Rapids will lose significantly to Genoa but AMD can not rest either.

Ehh sadly that may not be true. Intel has no clear followup after Sapphire Rapids. We're talking about a completely unknown design (Granite Rapids) on a new node with no track record. The only things we know about Intel 4/7nm are:

1). Ponte Vecchio was pulled from this node. Note that the Intel 4 chiplets that were meant to go into Ponte Vecchio were supposed to be ready by now. Intel has delayed a major supercomputer contract over this debacle.

2). Loihi 2 is the pipecleaner.

3). Wafer capacity may be less than 1/3rd of the capacity for Intel 7/10ESF.

Not really inspiring a lot of confidence yet, is it? If Intel had this many delays with Sapphire Rapids on 10ESF, what makes anyone think Granite Rapids will ship in a timely fashion? It will not hit the market at the same time as Raptor Lake. Probably 6-12 months later.

The difference is merely a year. I suspect they'll have about a quarter advantage before Genoa.

On paper? Maybe. See the Milan launch. Genoa will be in many hands long before we see it released to the public. Intel will be pushing Sapphire Rapids out the door ASAP unless they finally bury it like IceLake-SP.

 

[Exist50]

Sapphire Rapids will lose significantly to Genoa but AMD can not rest either. The difference is merely a year. I suspect they'll have about a quarter advantage before Genoa.

With Raptor Lake that's why I said 20% from the E cores, rather than the 30%+ since it'll have to clock lower. Although I don't think Raptor Lake will remain just with more E cores plus a modified Golden Cove core. Even refreshes get slight boosts.

Circuitry refinements and additional year from Intel 7 will allow them to squeeze little more out of it. I could attribute it to DLVR, but I personally believe it's a mobile thing.

Don't underestimate them, and also know that any "entity" that doesn't die, be it a person or a company comes back. Things are at the lowest point before it swings back.

The problem with Raptor Lake, and Emerald Rapids, for that matter, is that it's a stopgap gen because of delays to Meteor Lake/Granite Rapids. Stopgap gens never yield anything noteworthy.

 

[DrMrLordX]

Emerald Rapids

First I'm hearing of that. Is that like a Sapphire Rapids refresh?

 

[igor_kavinski]

First I'm hearing of that. Is that like a Sapphire Rapids refresh?

Intel's next-gen Emerald Rapids CPU: 64 cores, 80 PCIe 5.0 lanes, more

Intel Emerald Rapids (EMR) early leak: 64 cores at 350W, 80 PCIe 5.0 lanes, DDR5-5600 support, rumored launch of Q1 2023.

www.tweaktown.com

Intel's next-gen Emerald Rapids would launch with up to 64 cores at up to a whopping 350W and it'll be made on Intel's upcoming Intel 7 node. But more surprisingly, Emerald Lake would roll out with an insane 80 PCIe 5.0 lanes and support for DDR5-5600 memory.

350W?! They gonna do liquid cooling in the data center too?

 

[DrMrLordX]

That article is confusing. Intel 7 isn't an "upcoming" node: it's just 10ESF. Also, I think Sapphire Rapids will have SKUs in the 350W range.

 

[andermans]

And rumors are than AMDs Genoa will also have a cTDP of up to 400W, so it seems like everyone is increasing their power envelope. Given that PCIe cards and OAM around that TDP already exist with air cooling I'd bet it isn't really an issue.

 

[Markfw]

And rumors are than AMDs Genoa will also have a cTDP of up to 400W, so it seems like everyone is increasing their power envelope. Given that PCIe cards and OAM around that TDP already exist with air cooling I'd bet it isn't really an issue.

I thought that was only for the 128 or 196 core CPU's ??

Edit: 96 core, and the Turin is 256 cores for 600 watt:

AMD EPYC "Turin" with Zen5 cores rumored to feature maximum TDP of 600W - VideoCardz.com

AMD EPYC 7005 Turin more than doubles maximum TDP of Milan 2022 and beyond, a decade of high-wattage PC components. It is no longer NVIDIA Lovelace, Hopper, and AMD RDNA3. AMD EPYC Turin is now joining the list of devices requiring lots of power. Presumably, a special version of the EPYC 7005...

videocardz.com

 

[igor_kavinski]

And it gets even worse. 700W for short periods!

AMD EPYC "Turin" with Zen5 cores rumored to feature maximum TDP of 600W - VideoCardz.com

 

[Abwx]

View attachment 55421

And it gets even worse. 700W for short periods!

AMD EPYC "Turin" with Zen5 cores rumored to feature maximum TDP of 600W - VideoCardz.com

For one ms...

 

[igor_kavinski]

For one ms...

That can't be good for the CPU, can it? Imagine the thermal stress that much heat will place on the CPU die. That's a lot of current coursing through transistors.

 

[Abwx]

That can't be good for the CPU, can it? Imagine the thermal stress that much heat will place on the CPU die. That's a lot of current coursing through transistors.

700W during 1ms amount to 0.7W during 1 second, that s nothing compared to the average power and is negligible given the chip thermal inertia.

DT CPUs have also very high peaks for 1ms duration, actually much more than this SKU relatively to the average power.

 

[igor_kavinski]

DT CPUs have also very high peaks for 1ms duration, actually much more than this SKU relatively to the average power.

Do you have figures for 5950X and 12900K, for comparison?

 

[Abwx]

Do you have figures for 5950X and 12900K, for comparison?

I have no numbers under the hand other than Haswell 4790K whose 1ms peak power was rated at 160W or so, for the 12900K Intel has published the short duration peak power requirement for PSUs to correctly feed this CPU, they can be found with some googling.

 

[igor_kavinski]

I have no numbers under the hand other than Haswell 4790K whose 1ms peak power was rated at 160W or so, for the 12900K Intel has published the short duration peak power requirement for PSUs to correctly feed this CPU, they can be found with some googling.

Intel Alder Lake ATX12V peak current recommendations are higher than Rocket Lake's - VideoCardz.com

Is the 12900K a 165W CPU?

 

[Abwx]

Intel Alder Lake ATX12V peak current recommendations are higher than Rocket Lake's - VideoCardz.com

Is the 12900K a 165W CPU?

Dunno why using flawed datas from this site, offical specs are better found here although only continuous power are mentionned :

Processeur Intel® Core™ i9-12900K (30 Mo de cache, jusqu'à 5,20 GHz) Caractéristiques techniques

Processeur Intel® Core™ i9-12900K (30 Mo de cache, jusqu'à 5,20 GHz) fiche de synthèse comprenant les spécifications, les caractéristiques, les prix, la compatibilité, les documents de conception, les références de commande, les codes de spécification, etc.

ark.intel.com

Anyway there were enough reviews as to one has to wonder why asking such a useless question...

 

[igor_kavinski]

Anyway there were enough reviews as to one has to wonder why asking such a useless question...

Apologies for irking you. Just confused if 165W is for an upcoming HEDT CPU...

 

[Abwx]

Apologies for irking you. Just confused if 165W is for an upcoming HEDT CPU...

No irk here but guess that i ll have to make a clarification about what are exactly those 1ms peaks, to summarize if the CPU is on a low power state at say 0.8V then getting to the max frequency at all core at say 1.3V will be seen as a short circuit by the supplying devices.

Said otherwise the CPU equivalent capacitance need to be charged instantly from say 0.8V to 1.3V while frequency (and hence power comsumption) is also increasing at the same time, that will create a very short peak that exceed largely the rated TDP.