Intel Xe Architecture for dGPUs

What are Passive Die Stiffeners? They're some kind of interconnect or PHY?

 

Could they be exactly what they claim? Passive pieces of silicon (+maybe routing for the emib, since they say emib is under them & hbm) to make the package hold together better?

 

ahh ok, that makes sense. I didn't realize it's emib so it's package filler.

 

Well that looks, expensive.

 

Probably won't run Crysis too well either.

 

At this point it might be able to run it in software emulation

 

Can't wait for the Larrabee vs Xe video from DF

 

For a consumer yes.
If you work in datacenters, hardware is the cheap part while software is the more costly part.

 

What's the point of "releasing" something like that ?

 

To shift it into a different Accounting Bucket in order to start taking depreciation or some other accounting aspect?

 

A first "test" of the DG1.
A can't see, what driver version was used. But, it sounds as if several driver updates have been received.
I'm supprised about this premature state of software...
https://www.igorslab.de/en/set-with...h-massive-driver-problems-at-benchmarks-leak/

Edit: The AIDA64 screenshot shows driver: 27.20.100.8890
does anybody know, if this is a recent one?

 

Are you saying that because of the crashes /errors ? If so, the card is a test sample, right ? So for all we know, the hardware is causing this...

 

Those are from early november last year and meant for Xe MAX (which while same chip is using different configuration which could possibly cause issues)
Current driver build is 27.20.100.9168

 

Strange, they didn't use an up-to-date driver...

Why would one send out broken hardware? To let everybody test if the mechanics fit?

 

Maybe Xe MAX isn't supported in the regular releases yet? Dunno

 

Intel DG1 GPU Teardown, Failed Benchmarks, and Why It Won't Work on Most Systems

https://www.tomshardware.com/news/intel-dg1-gpu-teardown-and-first-failed-benchmarks

------------------------------------------------------------------------

What a POS

 

This seems to be only a recap of igor's article, why post this?

 

Tomshardware speculates performance is around GTX 1030.

 

Again, test sample. It's not uncommon for test sample to not work very well without special software or a specific setup... Legit review will come pretty soon I guess anyway.

I'm pretty interested about their 10nm process