Discussion Intel current and future Lakes & Rapids thread

[poke01]

Xeon Sapphire Rapids-SP 2S Geekbench

Quanta Cloud Technology Inc. QuantaGrid D54Q-2U - Geekbench

Benchmark results for a Quanta Cloud Technology Inc. QuantaGrid D54Q-2U with an Intel Xeon Platinum 8480+ processor.

browser.geekbench.com

78,000 points for 112 cores

Thats bad... Intel is behind

this an 96 core EYPC zen 4.

 

[inf64]

Thats bad... Intel is behind

this an 96 core EYPC zen 4.

That Genoa part has an issue while running the multithreaded AES-XTS benchmark, the score is lower than the single core score.

 

[nicalandia]

That Genoa part has an issue while running the multithreaded AES-XTS benchmark, the score is lower than the single core score.

As to be expected from unreleased hardware and BIOS.

The same goes for the Sapphire Rapids-SP parts, those are only preliminary benchmarks, that paint a trend in performance that we are aware of.

We know that Zen4 at ISO Core and ISO Speed are either a match for Golden Cove or slightly ahead(Which will be problematic for Intel because AMD has numerical core advantage and process node advantage that will make EPYC 1S and 2S systems more efficient). The unknowns are what impact on performance will the 4 tile design have on Sapphire Rapids-SP, does the Mesh of half Rings being extended that far(across four tiles) and only 1.875 MiB per core has any impact on performance?

On general purpose apps we can almost be certain that AMD will be ahead.. On specific task that take advantage of HBM2e on package memory and AMX I believe Intel will produce impressive performance not seen before.

 

[nicalandia]

Here is the same Intel Xeon Platinum 8480+ scoring 37,794 Points due to being a single CPU 56C/112T instead of a 2S system.

Unknown - Geekbench

Benchmark results for an Unknown with an Intel Xeon Platinum 8480+ processor.

browser.geekbench.com

And a single 32C/64T

PEGATRON Lightning - Geekbench

Benchmark results for a PEGATRON Lightning with an AMD Eng Sample: 100-000000866-01 processor.

browser.geekbench.com

 

[mikk]

MLID says Granite Rapid gets 5-10% more IPC on Redwood+ architecture and Intel 3 node. No Lion Cove?

 

[nicalandia]

MLID says Granite Rapid gets 5-10% more IPC on Redwood+ architecture and Intel 3 node. No Lion Cove?

Thats correct, Redwood Cove+ on a better node. Its going to need more than 10% on IPC to still be relevant against Turin Zen5. Sapphire Rapids and Emerald Rapids will have to endure such a difficult task against Zen4, Zen4X3D

 

[naad]

Genoa single core gets similar score to a 4.4/4.5GHz Cezanne APU, quite impressive.
We've been fed 2.7-3.0GHz single core mainstream (non-F) server chips for years, finally improvements are in order.

 

[eek2121]

Thats bad... Intel is behind

this an 96 core EYPC zen 4.

View attachment 69106

Intel plans on making up for it with more sockets. Compute density with Intel will actually be higher than with AMD. A 448 core Intel system will likely be ahead by a somewhat decent amount. Don’t ask about power consumption, however.

 

[Exist50]

MLID says Granite Rapid gets 5-10% more IPC on Redwood+ architecture and Intel 3 node. No Lion Cove?

Ignoring for a sec his IPC predictions (which I'll note, he's already more than halved vs his prior claims), if GNR is RWC-based ("+" or otherwise), then it will not realistically compete head to head with Turin. Should close much of the SPR/EMR-Genoa gap from the increased core count and node shrink, but they're not going to be competitive with that core and similar process nodes.

 

[Exist50]

Intel plans on making up for it with more sockets. Compute density with Intel will actually be higher than with AMD. A 448 core Intel system will likely be ahead by a somewhat decent amount. Don’t ask about power consumption, however.

If they need two blades for 8S, then they're not likely to have better or comparable compute density. And >2S is a small and dying market anyway. Long term, I imagine 1/2S systems will be the norm, interconnected with CXL for coherency.

 

[Geddagod]

MLID says Granite Rapid gets 5-10% more IPC on Redwood+ architecture and Intel 3 node. No Lion Cove?

I'm confused, I did not hear anything about redwood cove+ being 5-10% more IPC than either redwood cove or raptor cove, in the video.

 

[nicalandia]

I'm confused, I did not hear anything about redwood cove+ being 5-10% more IPC than either redwood cove or raptor cove, in the video.

Intel's Own CEO Pat said that it would be an enhanced core on a smaller node.

 

[Geddagod]

Intel's Own CEO Pat said that it would be an enhanced core on a smaller node.

Ik but the IPC figures? Tbh I don't expect a 5-10% IPC uplift from redwood cove+, maybe like >5 or ~5? because I doubt they could change the architecture too much without having to move the timeline again.
Plus I think @Exist50 thinks that you meant over 5-10% ipc over raptor cove, Tom didn't give any figures, he just said a bigger uplift from alder lake to raptor cove. So Tom expects slightly more than the 10-20% IPC uplift he cited before for redwood cove.

 

[Exist50]

Ik but the IPC figures? Tbh I don't expect a 5-10% IPC uplift from redwood cove+, maybe like >5 or ~5? because I doubt they could change the architecture too much without having to move the timeline again.
Plus I think @Exist50 thinks that you meant over 5-10% ipc over raptor cove, Tom didn't give any figures, he just said a bigger uplift from alder lake to raptor cove. So Tom expects slightly more than the 10-20% IPC uplift he cited before for redwood cove.

Yeah, I didn't actually watch the video. But if he's still claiming 10-20% or whatever, much less anything higher, well then get your popcorn ready, because this'll be a trainwreck you don't want to miss. Well, more for him than anyone else, but we all know how that goes.

 

[IntelUser2000]

Pat said 10-12% not 5-10%. Another time he said "10% plus". Again MLID has some sources but they are tainted by rampant speculations of his own.

 

[Geddagod]

Pat said 10-12% not 5-10%. Another time he said "10% plus". Again MLID has some sources but they are tainted by rampant speculations of his own.

I recall the 10% plus but not the 10-12%. Not that I doubt it won't be in that range.
I believe this was the exact quote-
"We did change the timing of Granite Rapids, and we had a big internal debate show on should we even keep the Granite Rapids name because it was the same platform, but it was a new core on a new process. So to some degree, it was a very different product. But some said, hey, you delayed Granite Rapids. Hey, I say I enhance Granite Rapids, with a much higher performance product, a much — 18% process, a major new core, that’s 10-plus percent in the core. So a much better product and aligned to the customers’ timing. And they said, hey, Sapphire Emerald Granite was too compressed. "

 

[Exist50]

I recall the 10% plus but not the 10-12%. Not that I doubt it won't be in that range.
I believe this was the exact quote-
"We did change the timing of Granite Rapids, and we had a big internal debate show on should we even keep the Granite Rapids name because it was the same platform, but it was a new core on a new process. So to some degree, it was a very different product. But some said, hey, you delayed Granite Rapids. Hey, I say I enhance Granite Rapids, with a much higher performance product, a much — 18% process, a major new core, that’s 10-plus percent in the core. So a much better product and aligned to the customers’ timing. And they said, hey, Sapphire Emerald Granite was too compressed. "

"A major new core" almost makes me wonder if it was intended to be upgraded to Lion Cove, but they were forced to backtrack sometime after Gelsinger's statement. Surely something they don't even deign to give a new name would not be worth calling that? Or MLID is simply wrong. Certainly won't discount that possibility. But if it is just a Redwood Cove derivative, that would be extremely disappointing.

 

[IntelUser2000]

"A major new core" almost makes me wonder if it was intended to be upgraded to Lion Cove, but they were forced to backtrack sometime after Gelsinger's statement. Surely something they don't even deign to give a new name would not be worth calling that? Or MLID is simply wrong. Certainly won't discount that possibility. But if it is just a Redwood Cove derivative, that would be extremely disappointing.

We can't focus on the naming too much. Generally a new name suggests a big change, but it's not an absolute thing. They can do whatever they want, it's their project after all. Goldmont Plus, anyone?

Technically, it is a plus and an expansion of the predecessor, but the performance increase was pretty fantastic.

Also they sandbagged Raptorlake. "Up to double digit performance boost". So we now know they were talking about single thread and planned to "beat" that.

Not saying they are sandbagging GNR. It's that we don't know crap.

As for MLID:

2024: "Oh I was never sure about that. The source doesn't have a 100% track record."

 

[Exist50]

We can't focus on the naming too much. Generally a new name suggests a big change, but it's not an absolute thing. They can do whatever they want, it's their project after all. Goldmont Plus, anyone?

Yeah, this is just reading the tea leaves, but if they were willing to give "Raptor Cove" its own name for such meager gains, then I wouldn't expect anything more from "RWC+". I certainly doubt such an upgrade would count as a "major new core". Barely even "new"! But I've been quite open about my pessimism regard Redwood Cove.

 

[IntelUser2000]

Yeah, this is just reading the tea leaves, but if they were willing to give "Raptor Cove" its own name for such meager gains, then I wouldn't expect anything more from "RWC+". I certainly doubt such an upgrade would count as a "major new core". Barely even "new"! But I've been quite open about my pessimism regard Redwood Cove.

Well 10-12% is at least of significance. The P core in Raptorlake is just the L2 cache and the prefetcher, which the E cores got too!

 

[Exist50]

Well 10-12% is at least of significance. The P core in Raptorlake is just the L2 cache and the prefetcher, which the E cores got too!

I'm not convinced we'll see 10-12% IPC over Raptor Cove, much less Redwood Cove. But I guess we'll have to wait and see.

 

[nicalandia]

Very good Info on Sapphire Rapids and Emerald Rapids..



So either Emerald Rapids is Ramping Up Sooner than expected or...Sapphire Rapids is Ramping Up Later as in late 2023?

https://twitter.com/x/status/1584941088443625474

 

[jpiniero]

If Cost of Power is a first order of factor, shouldn't customers be buying the other guys?

 

[DrMrLordX]

Very good Info on Sapphire Rapids and Emerald Rapids..

That may be good news for Intel that Sapphire Rapids' delays have essentially compressed the launch window for the product without pushing Emerald Rapids back. But yes it does make you wonder if that means that Sapphire Rapids will have even more delays.

 

[lightisgood]

So either Emerald Rapids is Ramping Up Sooner than expected or...Sapphire Rapids is Ramping Up Later as in late 2023?

Rumor has it that EMRs might be scheduled in early 2023.
Now, I'm putting a simple interpretation on SPRs/EMRs.

SPRs : low-end ( 600~700sqmm die area?).
EMRs : mid-range ( ~800sqmm die area).
SPRs with EMIB: high-end ( ~1600sqmm die area ).
EMRs with EMIB: extream ( >=1600sqmm die area).