Workstation Xeon W (Ice Lake variant).

KarelG

Member
Jan 29, 2020
43
12
8
Looks like Intel is preparing also workstation Xeon W-33xx variant of Ice Lake Xeons. Puget Systems does have W-3335 in their benchmark database from yesterday:


This is indeed interesting move since performance comparison with Xeon W-32xx and W-22xx may be a little bit problematic for Ice Lake based implementations due to low turbo clock especially when considering lower core count variants...
 

kingmouf

Member
Jun 15, 2016
30
4
8
40
Looks like Intel is preparing also workstation Xeon W-33xx variant of Ice Lake Xeons. Puget Systems does have W-3335 in their benchmark database from yesterday:


This is indeed interesting move since performance comparison with Xeon W-32xx and W-22xx may be a little bit problematic for Ice Lake based implementations due to low turbo clock especially when considering lower core count variants...
I cannot find, so maybe they pulled it?
 

KarelG

Member
Jan 29, 2020
43
12
8
twitter use @hxl uploaded workstation cpus presentation:
Second slide claims that maximum frequency will be 4.0 GHz which is +300MHz higher than top-class server chips now. This may make sense to upgrade W-32xx line of chips which ends on 4.4 GHz, but probably is out of luck with W-22xx which ends on 4.8 GHz happily -- hence even envisioned +15% IPC gain of IceLake core is not able to provide more performant chip. That's probably also the reason why -W line of chips in LGA 2066 socket is so long and continues down to 2022.
 
Last edited:

111alan

Member
Mar 11, 2019
82
29
18
Haerbing Institution of Technology
Just got the 4189 variant, it seems that this gen is up on the clock speed and has no hard-limited difference between single core turbo and multi core turbo despite they never said so? 32x2 cores at 3.7GHz all core is the deadliest thing I've ever seen :X

Sorry for the NDA I signed, can't post the picture.
 

KarelG

Member
Jan 29, 2020
43
12
8
3.7GHz on all 32 cores? Wow! But, well, if turbo is 4.0GHz only, then there is only 300MHz difference between max single-thread perf and max multi-thread perf and that in the same power envelope. I'm curious what is IceLake holding back to not be able to turbo much higher on 1 core when 32 cores are so efficient that they can run on 3.7GHz in specified TDP.
 
  • Like
Reactions: nnunn

111alan

Member
Mar 11, 2019
82
29
18
Haerbing Institution of Technology
3.7GHz on all 32 cores? Wow! But, well, if turbo is 4.0GHz only, then there is only 300MHz difference between max single-thread perf and max multi-thread perf and that in the same power envelope. I'm curious what is IceLake holding back to not be able to turbo much higher on 1 core when 32 cores are so efficient that they can run on 3.7GHz in specified TDP.
Mine is an OEM model, which only has the turbo max of 3.7GHz. that's why I said that there may be no difference between all-core and single-core freq. There's another retail sku which displayed the same behavior. The listed TDP is 300W but in cinebench the package power only reached about 280w. They might've finally improved the lousy F-IVR in Skylake.

Actually since each core has its own voltage regulator, and there's an overall TDP limit, I don't know the reason why they were still using the concept of single/dual core turbo after Skylake in the first place, let alone icelake.
 

KarelG

Member
Jan 29, 2020
43
12
8
Sounds good! This looks like Intel may be competitive up to AMD TR PRO 3975WX which would be great. As always the question will be price...