Intel introduces new high-performance graphics brand: Arc

I'm actually looking more forward to what they do APU wise with this after. Intels big little stuff... sandwiched with a real APU could be much more interesting. I think over the next 2-4 years things like the Deck are going to be what drive gaming.
It might sound crazy... but I am more looking forward to a Intel Nuc type device with Say 4 big cores 2 little and a decent Arc (I guess) GPU running some fast system ram and Steam OS. :) To would be a super interesting product to me... more then a third player in a GPU space where everyone is using TMSC anyway so stock issues and price gouging to go with don't change anyway.

I think this is where it's going too, and I really like the idea of a nuc-like PC running steamos with an advanced APU in it. Something 2-3x the power of the Steam Deck's processor could be a real game changer for small form factor gaming systems, or "living room" systems running steamOS in big picture mode as a viable "PC gaming console".

I like that idea a lot.
 
I think this is where it's going too, and I really like the idea of a nuc-like PC running steamos with an advanced APU in it. Something 2-3x the power of the Steam Deck's processor could be a real game changer for small form factor gaming systems, or "living room" systems running steamOS in big picture mode as a viable "PC gaming console".

I like that idea a lot.
Can Intel build a better console, that’s basically what that would come down too.
A big part of that being a possibility is going to really depend on what their Linux GPU drivers look like.
 
Will any of these offerings actually be decent? I know intel is sitting on cash, lots of cash/cash-equivalents. They really look bad as a company. I hope that wont translate into these products.
 
Can Intel build a better console, that’s basically what that would come down too.
A big part of that being a possibility is going to really depend on what their Linux GPU drivers look like.

I don’t know if it’s got to be better, tbh, although given the price premium they usually put on NUCs it would certainly help!

Thankfully their Linux drivers are excellent (and open source) already. I can’t imagine that would change with Arc.

as a longtime Linux gamer I’m really interested to see what they can do with this.
 
It is, pretty close but that of course included, console, mobile, and PC gamers and there is undoubtedly some double entries in there.

ex. You use an outlook.com email for your Xbox, but use gmail for steam, and a secret Apple.com for your mobile. Unless you’ve got them all linked somewhere public that one user counts as 3 for their studies.

Exactly, if you have a cell phone, a console, and a PC, you are counting as 3 gamers in their numbers.
 
Let´s hope they come out with good stuff, more players in the market is a good thing. At least we know that Intel is big enough to be able to compete, I can´t see any other hardware company having the structure to go against a giant like NVidia.
 
I think they are shooting themselves in the foot by waiting until q1 2022 and targeting specifically the 3070 for performance. The likelihood of AMD or Nvidia paper launching a new product during that quarter are high and it wont exactly make the run of the mill consumer excited when performance is compared to "old" products.

Unless they do something insane like price them at 199 msrp that is.
 
Unless they do something insane like price them at 199 msrp that is.

It's Intel's MOS to lose money to gain market share. AMD and Nvidia know this, too, which is why they're fine with their shared thick margins right now. They aren't in competition with each other right now, not in the conventional sense. They're testing each other, sharpening their knives if you will. The possibility of a real throwdown is just around the corner.
 
Intel's drivers are pretty solid, I kinda hope they lay off any creature features and just put out stable kick-ass drivers, preferably for Linux too.
These cards will probably be a good option on Linux as long as they're decent. Intel is quite active in the Linux kernel community and there is no doubt the cards will be supported.
 
These cards will probably be a good option on Linux as long as they're decent. Intel is quite active in the Linux kernel community and there is no doubt the cards will be supported.
The cards they put out for Aurora were supposedly better than what AMD and NVidia were offering up at the time so that’s at least remotely promising. The GPU’s they’ve tossed into the 11th gen Mobile parts aren’t too bad either even at 10nm I would imagine that on their 7nm and TSMC 5 things only get better. I’m hopeful, I love that AMD is pushing them, I just hope they can continue to do so for the foreseeable future.
 
I think they are shooting themselves in the foot by waiting until q1 2022 and targeting specifically the 3070 for performance. The likelihood of AMD or Nvidia paper launching a new product during that quarter are high and it wont exactly make the run of the mill consumer excited when performance is compared to "old" products.

Unless they do something insane like price them at 199 msrp that is.
Or something insane like… have them available at launch….
 
I don't expect these to be overly expensive given Intel's bank and introduction into the high performance GPU market and having lots to improve probably in the driver department and functionality etc. They would shoot themselves in the foot to ride too much on the current GPU pricing inflation. If it performs RTX 3070 lvl I'd expect $399 MSRP => 600~$700 street pricing if RTX 3070 goes for around $1000 at the time of launch. They probably won't be able to ship enough volumes for Nvidia or AMD to care eitherway at launch.
 
Last edited:
Or something insane like… have them available at launch….
Intel is supposedly upgrading the following fab's over to 7nm:
D1B, RB1, D1C, D1D, D1X, Fab 34, Fab 42
Once complete they will have more 7nm capacity (supposedly equivalent to TSMC 5nm) than TSMC has capacity across their entire stack.
So assuming that Intel has worked out the 7nm kinks, and there isn't anything really saying they haven't at this stage then they should be good to go for Q3 2022. This also lines up with their i3's that they have announced that are being produced on the TSMC 3nm process, which Intel doesn't have anything to stand up against it and won't in their own timelines until 2024.
 
Some of the suggestions here sound downright ridiculous considering the circumstances. A company with no track record of producing good GPUs or supporting said GPUs is going to have a lot to prove before demanding Nvidia or AMD pricing. Until we see actual reviews there is absolutely no reason for excitement here IMO.
 
Some of the suggestions here sound downright ridiculous considering the circumstances. A company with no track record of producing good GPUs or supporting said GPUs is going to have a lot to prove before demanding Nvidia or AMD pricing. Until we see actual reviews there is absolutely no reason for excitement here IMO.
It's also a semiconductor manufacturer with nearly 3x the revenue of NVIDIA and AMD combined. And one that has been making (integrated) GPUs for quite a while. Of course be hesitant but more competition is something to look forward to.
 
I expect Intel to repeat similar methods from the past. OEMs, good deals for OEMs that buy a package deal, CPU, Motherboard (that is if Intel makes them again), GPU. Intel system will have Intel GPUs at a good price -> sell like hotcakes. You get multiple OEMs doing marketing for your own product that smashes anything Nvidia and AMD combined could do. As for retail, I just don't see much effort there.
 
I'm a bit confused, is this consumer Arc GPU (Xe-HPG DG2 codenamed Alchemist) being made on TSMC 7nm, 5nm, or 3nm? Or do we not know? I found conflicting reports but some were a year old.
 
I'm a bit confused, is this consumer Arc GPU (Xe-HPG DG2 codenamed Alchemist) being made on TSMC 7nm, 5nm, or 3nm? Or do we not know? I found conflicting reports but some were a year old.

All I saw in the links was marketing fluff and some generic gameplay with unknown performance numbers. And then the assumptions started flying around in all the posts....nothing substantial other than code names.
 
I'm a bit confused, is this consumer Arc GPU (Xe-HPG DG2 codenamed Alchemist) being made on TSMC 7nm, 5nm, or 3nm? Or do we not know? I found conflicting reports but some were a year old.
Unless Intel scrapped a generation these will be 7nm(+) TMSC.
https://www.reuters.com/article/technologyNews/idUSKBN29H0EZ
The delay likely has nothing to do with the Fab run. Intels first (ok first in a long long time) GPU has other points of slowness. With luck these will be good enough for Intel not to run from the segment and future chips won't see 2 year delays between engineering samples and actual sellable product.
 
I'm a bit confused, is this consumer Arc GPU (Xe-HPG DG2 codenamed Alchemist) being made on TSMC 7nm, 5nm, or 3nm? Or do we not know? I found conflicting reports but some were a year old.
The current GPU’s for the Aurora Supercomputer is made jointly on TSMC 5 and Intel 7nm. Given this is the consumer version of those cards I would expect them to keep the same processing node.
 
The current GPU’s for the Aurora Supercomputer is made jointly on TSMC 5 and Intel 7nm. Given this is the consumer version of those cards I would expect them to keep the same processing node.
I was under the impression that Aurora is actually using Ponte Vecchio GPUs mfged at 7nm at Intel.
https://www.alcf.anl.gov/aurora
Unless something has changed the Gov lists;
GPU Architecture
Xe arch-based “Ponte Vecchio” GPU; Tile-based chiplets, HBM stack, Foveros 3D integration, 7nm

I was under the impression that Intel was still planning to fab their own XE thing for Aurora the type of thing that makes zero sense to try and mass produce but that their fabs may be able to handle in smaller numbers for Aurora... with TMSC getting the DG2 contract. (which is a more sane design without the chiplet foveros stuff) Intel for sure has a 7nm+ contract with TMSC. Who knows though Intel is such a mess they probably have 4 different GPU projects going at once. lol
 
intel fabs their own stuff right? that is a huge 1up on everyone else.

but if it mines, MSRP goes out the window unless intel and push them out constantly and in high numbers while AMD and Nvidia wait on 3rd parties.
 
I was under the impression that Aurora is actually using Ponte Vecchio GPUs mfged at 7nm at Intel.
https://www.alcf.anl.gov/aurora
Unless something has changed the Gov lists;
GPU Architecture
Xe arch-based “Ponte Vecchio” GPU; Tile-based chiplets, HBM stack, Foveros 3D integration, 7nm

I was under the impression that Intel was still planning to fab their own XE thing for Aurora the type of thing that makes zero sense to try and mass produce but that their fabs may be able to handle in smaller numbers for Aurora... with TMSC getting the DG2 contract. (which is a more sane design without the chiplet foveros stuff) Intel for sure has a 7nm+ contract with TMSC. Who knows though Intel is such a mess they probably have 4 different GPU projects going at once. lol
Intel ran into troubles with yields and fell behind, the DoE threatened to sue Intel for breach of contract for failing to meet their deliverables. Intel worked out a deal with TSMC for space on their 5nm to pick up the slack. So now the chips are made with both processes, they use different SKU’s but perform nearly identically.
Intel has contracts for 7nm, 6nm (7+), 5nm, and 3nm; Intel’s monthly wafer allotment from TSMC is almost as large as AMD’s and has been for years, they use TSMC to produce a lot of their micro controllers and parts for many of their other projects.
 
  • Like
Reactions: ChadD
like this
It will be absolutely destroy the i740.

Beyond that, well, wait for reviews, don't get hopes too high.
I’m hoping it competes with the 1660 super and is available at truly insane quantity. I want to see like weekly restocks in the 100s at the local microcenter. That would basically cover the reasonable lower middle tier of gaming card and at large enough quantity bring the price to something reasonable. Right now the low end at microcenter is basically $189 for a 1050ti.
 
Thank you ChadD and Lakados. What is the total fab capacity of Intel 7nm vs TSMC 5nm? How do they compare today? Is intel at, say, 5% of TSMC? I'm a bit out of the loop.
 
Thank you ChadD and Lakados. What is the total fab capacity of Intel 7nm vs TSMC 5nm? How do they compare today? Is intel at, say, 5% of TSMC? I'm a bit out of the loop.
Intel’s roadmap has 7 facilities running 7nm for 2022, TSMC currently has 1 fab running 5nm, with a second being built in Arizona.

Output wise the two aren’t comparable, Intel operates 4x more facilities than TSMC does. That’s not a bad thing though, TSMC gets to be relatively agile in these respects, they get to do and try things Intel just can’t outside of a lab simply because their size allows for it and they have a far FAR better management structure.
 
Last edited:
Intel’s roadmap has 7 facilities running 7nm for 2022, TSMC currently has 1 fab running 5nm, with a second being built in Arizona.

Output wise the two aren’t comparable, Intel operates 4x more facilities than TSMC does. That’s not a bad thing though, TSMC gets to be relatively agile in these respects, they get to do and try things Intel just can’t outside of a lab simply because their size allows for it and they have a far FAR better management structure.

Does that take into account yields? Are intel's yields comparable?
 
Does that take into account yields? Are intel's yields comparable?
God only knows, neither Intel nor TSMC tend to talk about yields overly much. We do know their yields weren't great with their initial 7nm processes when they started producing the Ponte Vecchio chips but was that due to chip design or process or both who knows. Intel's Q4 earnings disclosure did state the 7nm process was back on track, for full delivery in 2023, not sure if that means fiscal 2023 which would be Q3 of 2022, or actual 2023 but Pat Gelsinger spoke on the call saying "he has personally reviewed progress on the company's 7nm process over the last week and that he is pleased with the "health and recovery of the 7nm program."
Bob Swan also chipped in this little tidbit as well "the company's yield issues with the 7nm process stemmed from difficulties with a sequence of steps in the company's production process, which introduced defects. By rearchitecting these steps, we have been able to resolve the defects"

So I am guessing they have gotten yields to a point where they are comfortable being able to roll forward, and as such have started the upgrade process on many of their 22/14 fab plants.

Interesting enough there was a rumor that Intel hired TSMC to help them fix their process in exchange they are paying TSMC a bonus on their existing and upcoming wafer contracts, but that is just something floating around some Reddit sites and probably 100% false, but fun to think of all the same.

https://seekingalpha.com/article/44...n-on-q4-2020-results-earnings-call-transcript

But in terms of raw output regardless of process density, Intel puts out more silicon than Samsung, TSMC, Micron, SK Hynix, and Kioxia combined. and those 5 take up positions 2-6 in terms of wafer output.
 
Last edited:
It will be interesting to see how the driver development for these cards is, if it's like 11th gen series with no graphics drivers for like 2-3 weeks or their vast amount of things they have just abandoned support for like VROC then it won't go anywhere.
 
https://www.anandtech.com/show/16895/a-sneak-peek-at-intels-xe-hpg-gpu-architecture

As part of today’s architecture day announcements, Intel is finally answering the burning question and disclosing the fab and node being used. As many have expected, Intel is indeed turning to TSMC to fab their gaming GPU, and they will be using TSMC’s N6 process to do it.

I unfortunately saw no mention of concrete plans to move the design to Intel fabs for Xe-HPG, but surely it's in the works if we know Xe-HPC is already taped out or in early production on Intel 7nm. Like many of you I just want enthusiast GPU prices to go down and I hope intel can move some production away from TSMC for this. Though even just at TSMC, with Intel's new GPU the total GPU production at TSMC ought to increase, I assume.
 
Last edited:
https://www.anandtech.com/show/16895/a-sneak-peek-at-intels-xe-hpg-gpu-architecture



I unfortunately saw no mention of concrete plans to move the design to Intel fabs for Xe-HPG, but surely it's in the works if we know Xe-HPC is already taped out or in early production on Intel 7nm. Like many of you I just want enthusiast GPU prices to go down and I hope intel can move some production away from TSMC for this. Though even just at TSMC, with Intel's new GPU the total GPU production at TSMC ought to increase, I assume.
The VCZ link I posted earlier said 6nm. I don't know why people were assuming 7nm or 5nm when TSMC's production has been tapped out on those fabs by other companies forever and a day now.
 
Back
Top