cancel
Showing results for 
Search instead for 
Did you mean: 

Longlife X299 Platform maybe we get Icelake X LGA 2066 soon

restsugavan
Level 13
86463

X299 Platform extending lifespan to end of 2021 with new hope
10nm+ SuperFin Icelake X LGA 2066 with 28 Core / 56 Thread / 28 x 1.25 MB L2 cache / 28 x 1.5 MB L3 cache
Sunny Cove Microarchitect PCIe 4.0 coming soon Q1 2021
https://www.hd-tecnologia.com/exclusiva-intel-lanzara-rocket-lake-s-y-z590-h570-b560-y-h510-a-finale...
W11CANARY 26085.1 Core i9 7980XE 02007006 MCE ME 11.12.95.2499 R6E OFFICIAL BIOS 3801 SAMSUNG OG9 FW 1019.0 SSD 970 EVO PLUS 1 TB x 3 NVIDIA RTX 4090 GAME READY 551.86 64GB GSKILL DDR4 3200MHz JBL 9.1 Sound Bar DTS-X
2,660 Views
9 REPLIES 9

Int8bldr
Level 11
Yeah I saw that too on: https://videocardz.com/newz/intel-rocket-lake-s-500-series-motherboards-roadmap-leaked

In a way I'm not unhappy, if X299 gets one more year.
Would be nice if intel release a Ice lake X gen CPU as a final X299 refresh.
That would give an upgrade path and extend the x299 platform life several more years beyond 2021.

-Nax-
Level 8
Oh !!! Here we go for a Rampage VI Extreme Encore Omega ?

What a very long time support for X299, in these times of 2 generations CPU by chipset it is just ... unbelievable from Intel if it is true.

G75rog
Level 10
Icelake will use LGA 4189 socket so it will be new motherboard time.

I think we all know the reason why intel is not coming out with a replacement for x299 until at the earliest h2 of 2021.
It's of course because they simply don't have a product to offer that would be remotely competitive to the TR series.

Int8bldr wrote:
I think we all know the reason why intel is not coming out with a replacement for x299 until at the earliest h2 of 2021.
It's of course because they simply don't have a product to offer that would be remotely competitive to the TR series.


I think they already had TR destroyer products on their hand now. Skylake X was design and taped-out on 2014, If you look at AMD core that can be
compete was release on 2018-2019. Sunny Cove was taped-out on 2017. Willow Cove was taped-out 2018. And lastest Golden Cove was taped-out
last year.

Few days ago, Intel was testing 72 Core Sapphire Rapid CPU with new AMX technology (Advance Matrix Extenstion) .86518

and also many next gen platform that could be easily destroy TR or Ryzen Core easily. We're only waiting.
W11CANARY 26085.1 Core i9 7980XE 02007006 MCE ME 11.12.95.2499 R6E OFFICIAL BIOS 3801 SAMSUNG OG9 FW 1019.0 SSD 970 EVO PLUS 1 TB x 3 NVIDIA RTX 4090 GAME READY 551.86 64GB GSKILL DDR4 3200MHz JBL 9.1 Sound Bar DTS-X

restsugavan wrote:
I think they already had TR destroyer products on their hand now. Skylake X was design and taped-out on 2014, If you look at AMD core that can be
compete was release on 2018-2019. Sunny Cove was taped-out on 2017. Willow Cove was taped-out 2018. And lastest Golden Cove was taped-out
last year.

Few days ago, Intel was testing 72 Core Sapphire Rapid CPU with new AMX technology (Advance Matrix Extenstion) .86518

and also many next gen platform that could be easily destroy TR or Ryzen Core easily. We're only waiting.


Interesting! AND I want it !

But my point is still valid, intel will not be able to get this out for another year, i.e H2 2021

AMX tech instruction set would be great IF and I repeat IF you can run AMX instructions at same speed as other instructions (no AMX negative offset like with AVX 512). The AVX 512 negative clock offsets are just too large - this has effectively made server admins disable AVX 512 all together (because if one process is running AVX 512 then all cores on the same CPU gets down clocked to AVX 512 levels, destroying through put for all other processes also foe processes not doing any AVX 512).

further AMX is only BF16 I think, that makes the TOPS non comparable. AVX 512 can do proper FP32 and FP64 vectors - amazing.
If you learn how to oc AVX 512 (low offsets) you can get some serious proper 64 bit and 32 bit FLPS work done with a core i9 10980 XE.

restsugavan wrote:
I think they already had TR destroyer products on their hand now. Skylake X was design and taped-out on 2014, If you look at AMD core that can be
compete was release on 2018-2019. Sunny Cove was taped-out on 2017. Willow Cove was taped-out 2018. And lastest Golden Cove was taped-out
last year.

Few days ago, Intel was testing 72 Core Sapphire Rapid CPU with new AMX technology (Advance Matrix Extenstion) .86518

and also many next gen platform that could be easily destroy TR or Ryzen Core easily. We're only waiting.


For us this won't destroy anything to be honest since DL really has no impact in enthusiast space which is why Intel is not releasing cooper lake for HEDT. Ice Lake also uses new refresh PCH same as cooper lake so I don't see it coming to X299. I guess it's possible I just don't see the point in downgrading a PCIE4 platform to PCIE3 to be honest.

https://www.tomshardware.com/news/intel-10nm-xeon-ice-lake-sp-sunny-cove-core-architecture
https://www.servethehome.com/3rd-generation-intel-xeon-scalable-cooper-lake/5/

restsugavan
Level 13
If you learn how to oc AVX 512 (low offsets) you can get some serious proper 64 bit and 32 bit FLPS work done with a core i9 10980 XE.


About AVX512 clock offset had been address since Sunny Cove Microarchitect on Icelake X according from Intel slide by link below.
AVX512 clock running better on Sunny Cove Microarchitect compare to existing CPU which you running now. :cool:

https://rog.asus.com/forum/showthread.php?120186-Compare-Skylake-X-VS-Icelake-X-Architect-by-slide

Also Willow Cove and Golden Cove many things had been optimized ( Latency , Throughput , Bandwidth , Cache mechanism )
At my point here. Even Willow Cove (Tigerlake SP) will wiped out all AMD CPU easily many area.
DDR5 PCIe 5 CXL and all AVX512 without lower clock penalty. Just wait for LGA 4189 mate, :cool:
W11CANARY 26085.1 Core i9 7980XE 02007006 MCE ME 11.12.95.2499 R6E OFFICIAL BIOS 3801 SAMSUNG OG9 FW 1019.0 SSD 970 EVO PLUS 1 TB x 3 NVIDIA RTX 4090 GAME READY 551.86 64GB GSKILL DDR4 3200MHz JBL 9.1 Sound Bar DTS-X

restsugavan wrote:
About AVX512 clock offset had been address since Sunny Cove Microarchitect on Icelake X according from Intel slide by link below.
AVX512 clock running better on Sunny Cove Microarchitect compare to existing CPU which you running now. :cool:

https://rog.asus.com/forum/showthread.php?120186-Compare-Skylake-X-VS-Icelake-X-Architect-by-slide

Also Willow Cove and Golden Cove many things had been optimized ( Latency , Throughput , Bandwidth , Cache mechanism )
At my point here. Even Willow Cove (Tigerlake SP) will wiped out all AMD CPU easily many area.
DDR5 PCIe 5 CXL and all AVX512 without lower clock penalty. Just wait for LGA 4189 mate, :cool:


Thank you! I've seen those slides on Icelake X and they look to be excellent.

It took some time for me working on the BIOS settings to get my 10980XE to be able to OC with AVX 512 on at relatively low offsets. but I got there in the end.

For me that's the holy grail!

If you want to do Neural Net training you need FP32 or even FP64 (inference can be done with BF16 but training not so much) and AVX 512 can do 16 Fused FP32 multiplications and adds per clock cycle per core IF you program and use AVX 512 correctly (i.e. utilize both the AVX-512 ports that every core has every clock! ).
That is insanely good. 16 by 16 matrix tile multiplications can be done extremely fast this way with some clever AVX 512 assembly (need to re-order data vectorized in RAM, mange cache carefully and make excessive use of the 32 512 bit zmm registers!).
To day's complier's are not good at optimize for AVX-512 (it's hard to vectorize code even for humans) - I think this is the other reason why AVX-512 has a bad reputation (e.g. Linus statement on AVX 512 "painful death" thing), the first reason being the high default negative AVX 512 offsets.

Regardless, I hope you are right with the next gen and it comes soon. However, I'm happy with my 10980XE now that I managed to get AVX 512 to run sustainable with out burning up. It was just a lot harder than it should have been and took a lot longer time than it should have taken (lots of ucode tests and lots of BIOS work + hand coding AVX-512 matrix FMA's in assembly).