Monday, November 1st 2021

Intel Drops Xe-HP Server GPU Plans, to Stick with HPC and Client Graphics

Intel has dropped plans to build Xe-HP server GPUs commercially. This line of products would have powered cloud-based graphics rendering instances, for cloud-gaming or cloud-rendering applications. An announcement to this effect came from Raja Koduri, overseeing the development and monetization of Xe. Koduri stated that Xe-HP based instances were originally set up to power Intel's oneAPI devcloud as a software development vehicle for oneAPI and the upcoming Aurora supercomputer of the Argonne National Laboratory.

The company will now focus on Xe as a compute accelerator in the form of Xe-HPC "Ponte Vecchio," and discrete graphics in the client segment, leveraging the Xe-HPG graphics architecture. The smallest derivatives, the Xe-LP, powers integrated graphics solutions found in the company's Core processors (11th Gen and later). Back in the August 2021 Architecture Day presentation, Intel's technical brief for Xe HPC revealed that the silicon itself features certain on-die hardware relevant to graphics rendering (more here). This would have gone on to power the Xe-HP server GPU solutions.
Sources: Raja Koduri (Twitter), AnandTech
Add your own comment

20 Comments on Intel Drops Xe-HP Server GPU Plans, to Stick with HPC and Client Graphics

#1
GURU7OF9
Dont beleieve a word of it! Servers are biggest cash cow! There must be some other forces at play here !
Posted on Reply
#2
Flanker
Still. Waiting. For. Intel. Graphics. Card. To. Hit. The. Market.
Posted on Reply
#3
Ferrum Master
FlankerStill. Waiting. For. Intel. Graphics. Card. To. Hit. The. Market.
Imagine the popping sound when some fat fly hits a windshield.
Posted on Reply
#4
Vya Domus
I'm not surprised, judging from what they were planning these would have been monstrously expensive and it just wouldn't have made sense for a lot of companies to buy.
Posted on Reply
#5
Chrispy_
Chicken-and-egg problem here; Intel and Raja made a mistake by thinking that they could just roll out server GPU compute cards without a solid background history of driver, API, and software support.
As a corporate GPU buyer I don't buy server GPUs for the GPU performance, I buy them because those are the specific make and model recommended by the software solution we need to run on them.

Intel NEED TO GET XE/ARC PRODUCTS OUT ONTO THE MARKET YESTERDAY
  • If the consumer cards are successful enough to show a reliable track-record to big-dollar, risk-averse corporate buyers
  • If they are cost-effective in a cut-throat market
  • If the drivers turn out to be robust
  • If the developers adopt Intel's standards instead of Nvidia proprietary APIs
  • ....then, and only then, will Intel have a chance to get into the general GPGPU server market.
As I've said dozens of times already, Intel Xe/Arc news is just irrelevant noise until they finally put their product on the market. Every single article I post this on marks another week/fortnight/month of Intel failing to release a product that they told us was imminent at the end of 2017. I'll probably still be posting this paragraph well into 2022.
Posted on Reply
#6
sepheronx
so when do they expect to push pre-orders so i can get one?
Posted on Reply
#7
Vayra86
Chrispy_Chicken-and-egg problem here; Intel and Raja made a mistake by thinking that they could just roll out server GPU compute cards without a solid background history of driver, API, and software support.
As a corporate GPU buyer I don't buy server GPUs for the GPU performance, I buy them because those are the specific make and model recommended by the software solution we need to run on them.

Intel NEED TO GET XE/ARC PRODUCTS OUT ONTO THE MARKET YESTERDAY
  • If the consumer cards are successful enough to show a reliable track-record to big-dollar, risk-averse corporate buyers
  • If they are cost-effective in a cut-throat market
  • If the drivers turn out to be robust
  • If the developers adopt Intel's standards instead of Nvidia proprietary APIs
  • ....then, and only then, will Intel have a chance to get into the general GPGPU server market.
As I've said dozens of times already, Intel Xe/Arc news is just irrelevant noise until they finally put their product on the market. Every single article I post this on marks another week/fortnight/month of Intel failing to release a product that they told us was imminent at the end of 2017. I'll probably still be posting this paragraph well into 2022.
That's what I'm xeing too...

Also, Raja and miscalculating the market/marketing... what's new?
Posted on Reply
#8
xkm1948
Wonder how much of it is difficulty of developing their “oneAPI” that can take CUDA code. Good hardware needs good software support, especially for computation acceleration. They really need to polish oneAPI if they want to counter CUDA
Posted on Reply
#9
TheinsanegamerN
Chrispy_Chicken-and-egg problem here; Intel and Raja made a mistake by thinking that they could just roll out server GPU compute cards without a solid background history of driver, API, and software support.
As a corporate GPU buyer I don't buy server GPUs for the GPU performance, I buy them because those are the specific make and model recommended by the software solution we need to run on them.

Intel NEED TO GET XE/ARC PRODUCTS OUT ONTO THE MARKET YESTERDAY
  • If the consumer cards are successful enough to show a reliable track-record to big-dollar, risk-averse corporate buyers
  • If they are cost-effective in a cut-throat market
  • If the drivers turn out to be robust
  • If the developers adopt Intel's standards instead of Nvidia proprietary APIs
  • ....then, and only then, will Intel have a chance to get into the general GPGPU server market.
As I've said dozens of times already, Intel Xe/Arc news is just irrelevant noise until they finally put their product on the market. Every single article I post this on marks another week/fortnight/month of Intel failing to release a product that they told us was imminent at the end of 2017. I'll probably still be posting this paragraph well into 2022.
Reminds me of the late/hot arrival of vega when all we wanted int he consumer space was a bigger polaris and teh commercial guys wanted fixes to drivers and better API support.....
Posted on Reply
#10
Steevo
So Intel can’t deliver a product they have been talking about for how long now? It’s almost like words have meaning, unless their Intel words, which are clearly superior to the other words, but have no meaning without the special Intel word decoder ring, and really that is just Raj sending smoke signals asking for help or to be let go cause he said the thing and can’t produce it.

I feel bad for the guy at this point, so much hype the train is at terminal speed with the track ending at Argonne, an unproven tech sold to the public (let’s go Brandon) before they roll back other segments in the coming months or “adjust performance expectations”’on products, IE muzzle their jackass.
Posted on Reply
#11
Totally
GURU7OF9Dont beleieve a word of it! Servers are biggest cash cow! There must be some other forces at play here !
They're probably too power hungry to be competitive. History seems to be repeating itself since this seems to be playing out exactly like Vega/Polaris.
Posted on Reply
#12
iO
Bad yields and low efficiency because of their still underperforming 10nm node are the main reasons says the rumor.

Great omen for the Xe family if they got to kill 1/3 of it before it even launches..
Posted on Reply
#13
Totally
iOBad yields and low efficiency because of their still underperforming 10nm node are the main reasons says the rumor.

Great omen for the Xe family if they got to kill 1/3 of it before it even launches..
I could only bring myself to believe half of that or that is half the story considering Raj is involved and these were the same issues Vega and Polaris. If it was someone else I'd be inclined to give benefit of doubt but would still raise an eyebrow since the node isn't new, isn't the only product they have on the node, and is was something they knew about long ago.
Posted on Reply
#14
Darmok N Jalad
With the way things are going, Intel’s consumer GPUs will probably hit the market right when there’s a big mining bust and the market floods with new and used GPUs from the last decade.
Posted on Reply
#15
trsttte
GURU7OF9Dont beleieve a word of it! Servers are biggest cash cow! There must be some other forces at play here !
You're missing something, they're droping Xe-HP but they still have Xe-HPC for the server space, if anything Xe-HP was redundant and removing it may be a sign that they won't try to segment their market as much as nvidia (selling the same thing under Geforce and Quadro) which they also can't do anyway because they have zero track record.
iOBad yields and low efficiency because of their still underperforming 10nm node are the main reasons says the rumor.

Great omen for the Xe family if they got to kill 1/3 of it before it even launches..
Xe-HPG will be made (and always was) at TSMC, Xe-HP (now supposedly killed) and Xe-HPC were going to use their own fabs (and still are afaik)
Darmok N JaladWith the way things are going, Intel’s consumer GPUs will probably hit the market right when there’s a big mining bust and the market floods with new and used GPUs from the last decade.
Ethereum keeps delaying PoS, if Intel launches early in 2022 they are likely to still avoid the big flood of used cards, if they start delaying they'll inevitably hit a saturated market when the first miners start dumping stock
Posted on Reply
#16
iO
trsttteXe-HPG will be made (and always was) at TSMC, Xe-HP (now supposedly killed) and Xe-HPC were going to use their own fabs (and still are afaik)
The Xe-HPC’s compute tiles are also from TSMC. Initially planned to be fabbed both internally and at TSMC but they dropped the idea and went TSMC only.
Only the cache, EMIB and interposer tiles are made by Intel internally.
Posted on Reply
#17
yotano211
Ferrum MasterImagine the popping sound when some fat fly hits a windshield.
Imagine getting a front flat tire towing a 45k lb load.

Happened to me last night.
Posted on Reply
#18
Ferrum Master
yotano211Imagine getting a front flat tire towing a 45k lb load.

Happened to me last night.
You can name that tyre Raja.

It would align with the similarities

Imho Intel cards will pop like dead flies in the market. They will be expensive, slow and buggy, and already outdated.
Posted on Reply
#19
AsRock
TPU addict
I wish they get it over with already, not as it be the 1st time they failed, or maybe it be a success.

It be about time when was the last time 1980's ?.
Posted on Reply
#20
yotano211
Ferrum MasterYou can name that tyre Raja.

It would align with the similarities

Imho Intel cards will pop like dead flies in the market. They will be expensive, slow and buggy, and already outdated.
That's how my tire went, pop pop pop.....
Posted on Reply
Add your own comment