It's always Sunny in Xeon - Intel 10nm 24 core processor found

[williamcll]

While this is not the first Intel 24 core 10nm CPU, it is the first server chip to appear however. Performance wise it is slightly slower than the AMD equivalent EPYC 7402P which does have the higher clocks.

Quote

Following Cannon Lake's, Ice Lake is Intel's second family of processors to feature the 10nm process node. The 10nm+ chips are based on the Sunny Cove microarchitecture. While Ice Lake has debuted on the mobile plaform first and is found in laptops, it would appear that Intel is finally ready to shift its focus to the desktop, but more specifically, the server market.

 

Ice Lake is expected find its home on Intel's Whitley platform. The new processors are also rumored to usher in native support for PCIe 4.0 and DDR4-3200 RAM modules on an Intel desktop platform.

The Ice Lake Xeon processor in question surfaced with the Intel $0000 codename to disguise itself from unwanted eyes. The chip seemingly features 24 CPU cores and 48 threads. Geekbench reported a 2.19 GHz base clock and 2.89 GHz boost clock. The 24-core Ice Lake processor's cache design comes down to 1.25MB of L2 cache per core and a total of 36MB of L3 cache on the chip itself.

 

The Epyc 7402P raked in single-and multi-core scores of 4,498 points and 42,155 points in the same Geekbench 4 benchmark. AMD's offering outperformed the Ice Lake part by up to 9.7% in single-threaded performance. Multi-threaded performance between the Epyc 7402P and 24-core Ice Lake was in the same ballpark.  Rumor has it that Intel's Ice Lake server processors will arrive in the third quarter of this year.

Source: https://browser.geekbench.com/v4/cpu/15530472

https://www.tomshardware.com/news/intel-ice-lake-24-core-cpu-teases-10nm-on-server-platform
Thoughts: There would be very little reasoning for firms to buy this chip if their price isn't competitive. But at least it's good to know that PCIe 4 is finally getting official support. I haven't heard any news about the consumer desktop offering however, hopefully that will change in the near future. 

[GDRRiley]

Wow then looks like 10nm has worst yields than 14nm did when xeons launches on it if we end up with a 24 core max. That’s broadwell level. Intel better hope they can get a 60+ core unit make or this gen will be DOA

[GDRRiley]

Just now, VegetableStu said:

videocardz says they've delayed all their remaining 2020 launches, so.....

so then it will be if it launches after Milan not surprised they pushed it back

[CarlBar]

Didn't intel allready say backend of last year that they where skipping 10nm for the bulk of their products. I'd assume this is some sort of development sample where intel is basically using 10nm to play around with idea's for their eventual 7nm products.

[Dabombinable]

1 minute ago, CarlBar said:

Didn't intel allready say backend of last year that they where skipping 10nm for the bulk of their products. I'd assume this is some sort of development sample where intel is basically using 10nm to play around with idea's for their eventual 7nm products.

That's highly likely. From what I last heard, Intel was finding their 10nm node underwhelming, and actually more expensive than 22nm (low yields?).

[CarlBar]

1 hour ago, Dabombinable said:

That's highly likely. From what I last heard, Intel was finding their 10nm node underwhelming, and actually more expensive than 22nm (low yields?).

 

 

Bear in mind as with anything like this a big part of the cost comes down to production volume vs the development costs, given how little Intel has produced on the node and how much time and effort has gone into developing it it's likely they're running into the inverse of economies of scale driving their costs up.That said it's not a bad node to play around with idea's on because intel put a lot of effort into getting it running including an entire fabbing facility which otherwise probably wouldn't be doing very much. 14nm and 22nm are basically maxed out for intel ATM, 10nm they likely have more capacity than they know what to do with.

[porina]

4 hours ago, williamcll said:

Thoughts: There would be very little reasoning for firms to buy this chip if their price isn't competitive.

It is a lot more complicated than price per core-GHz. I'd hope they will have two unit AVX-512 implementations, probably with the VNNI update we saw in Cascade Lake.

 

4 hours ago, GDRRiley said:

Wow then looks like 10nm has worst yields than 14nm did when xeons launches on it if we end up with a 24 core max. That’s broadwell level. Intel better hope they can get a 60+ core unit make or this gen will be DOA

One unreleased CPU test result is hardly anything to go by, other than we possibly know they can offer 24 cores. We don't know what the maximum core count will be. We don't know what final clocks will be.

 

For some of my uses at least, it is easier to extract multi-thread scaling with a monolithic design than the fragmented CCX approach AMD have taken. So it will depend on the workload too.

 

1 hour ago, CarlBar said:

Didn't intel allready say backend of last year that they where skipping 10nm for the bulk of their products. I'd assume this is some sort of development sample where intel is basically using 10nm to play around with idea's for their eventual 7nm products.

Ice Lake SP (these server CPUs) I think was originally 1H this year, then got pushed back towards the end. I'd assume it'll be largely similar to the existing mobile offerings but with an uncore change like we was comparing Skylake-X to consumer Skylake-S. The question is, since we had some time between Ice Lake mobile and these eventual server parts, have they managed to tweak 10nm to do much better with that time?