News Intel GPUs - are the drivers getting better yet? (Maybe? Please?)

[DrMrLordX]

That's decidedly next-gen not current gen

More like "enterprise, not desktop". I'll be floored if anyone prioritizes CCIX on the desktop, but hey, you never know. Maybe Threadripper?

And besides, while possible, routing every single memory request through PCIe bus adds significant latency.

That is an issue, albeit not so much for an onboard iGPU as you indicated.

 

[Red Squirrel]

Yeah, to my admittedly amateur understanding there simply isn’t a market for it because the basic users don’t want to pay the extra for mid range performance and the upper end doesn’t want 1050 built in performance.
I suspect cooling is also a huge challenge because isn’t that area on the board real close to the cpu?

Right now there would be a market though given it's near impossible to buy GPUs. But yeah I guess the way it works now is the motherboard just has a port and no extra chips so the cost is low vs if they actually built the whole gpu in. But yeah I think it's just reality that now and the next couple years until they build more fabs is just not a good time to do any upgrades anyway.

 

[Fanatical Meat]

Right now there would be a market though given it's near impossible to buy GPUs. But yeah I guess the way it works now is the motherboard just has a port and no extra chips so the cost is low vs if they actually built the whole gpu in. But yeah I think it's just reality that now and the next couple years until they build more fabs is just not a good time to do any upgrades anyway.

certainly now is a missed opportunity for motherboard makers, just for how long and what happens next is the question.

 

[Gideon]

A bit offtopic but to reinforce my older point. AMDs own engineers also see similarites between the old and the new (e.g. the package being the new motherboard and I/O die the old Northbridge):

https://twitter.com/i/web/status/1395458960463912964

 

[mikk]

According to Moore's Law Is Dead DG2 is coming in Q1 2022, although he cannot rule out a limited Q4 launch (and possibly mobile only).

 

[uzzi38]

According to Moore's Law Is Dead DG2 is coming in Q1 2022, although he cannot rule out a limited Q4 launch (and possibly mobile only).

He also can't rule out the possibility that Intel would have decided to move DG2 to N7P for...reasons.

Disregard any sort of opinions from him. Frankly speaking I wouldn't even be considering anything he's written were it not for the images.

 

[mikk]

But why? Have a look to Ampere, it was unclear for a very long time and close to launch on which process it will be based on. I believe the first right leak came out in July when Ampere launched only 2.5 months later in September. Most if not all other specs were known at this point.

 

[guidryp]

But why? Have a look to Ampere, it was unclear for a very long time and close to launch on which process it will be based on. I believe the first right leak came out in July when Ampere launched only 2.5 months later in September. Most if not all other specs were known at this point.

MLID is not Kopite7Kimi.

Kimi drops the occasional accurate Tweet. MLID is a blowhard with an inaccurate clickbait YT channel... I thought I read he gave up on this, after people started calling him on his repeated inaccurate guesses...

 

[uzzi38]

But why? Have a look to Ampere, it was unclear for a very long time and close to launch on which process it will be based on. I believe the first right leak came out in July when Ampere launched only 2.5 months later in September. Most if not all other specs were known at this point.

Yes, except there's a big difference between Intel and Nvidia. Nvidia went from initial tape-out to product on shelves in record times. There was no A1 silicon. DG2 however... Well lets just say it's extremely late and actually should have already launched by Intel's initial timeline provided to OEMs.

 

[mikk]

MLID is not Kopite7Kimi.

Kimi drops the occasional accurate Tweet. MLID is a blowhard with an inaccurate clickbait YT channel... I thought I read he gave on this, after people started calling him on his repeated inaccurate guesses...

You missed the point. The point is it was unclear 3 months before launch on what process it's based on. It wasn't even clear if it's TSMC or Samsung. And here we have uzzi38 disregard everything from MLID just because he can't 100% confirm yet if it's really N6 (as rumored for a longer time) or N7P from TSMC.

Yes, except there's a big difference between Intel and Nvidia. Nvidia went from initial tape-out to product on shelves in record times. There was no A1 silicon. DG2 however... Well lets just say it's extremely late and actually should have already launched by Intel's initial timeline provided to OEMs.

Doesn't matter really, Xe is not bound to a specific process and obviously it's in ES state, manufacturing hasn't yet started. It makes no sense what you are saying.

 

[jpiniero]

Yes, except there's a big difference between Intel and Nvidia. Nvidia went from initial tape-out to product on shelves in record times. There was no A1 silicon. DG2 however... Well lets just say it's extremely late and actually should have already launched by Intel's initial timeline provided to OEMs.

I was under the impression that the N6 order from TSMC was for Q4 delivery.

 

[uzzi38]

You missed the point. The point is it was unclear 3 months before launch on what process it's based on. It wasn't even clear if it's TSMC or Samsung. And here we have uzzi38 disregard everything from MLID just because he can't 100% confirm yet if it's really N6 (as rumored for a longer time) or N7P from TSMC.




Doesn't matter really, Xe is not bound to a specific process and obviously it's in ES state, manufacturing hasn't yet started. It makes no sense what you are saying.

What doesn't make sense? That MLID is a known BS-er that is including clear BS on slides like this which absolutely ruins his credibility, when in reality the node for DG2 has been known for a while now?

When I say disregard what he says, I mean that he has ruined his reputation enough with stupid comments like this that frankly speaking you should not be taking anything he says at face value. Companies do not just decide to change which node they produce their product on 7-9 months before launching them

 

[uzzi38]

I was under the impression that the N6 order from TSMC was for Q4 delivery.

It is yes. And current timelines are now 2022 for DG2. They weren't originally.

 

[Gideon]

Here is what MLID said about Ampere in May 2020. At least a quarter or two closer to launch than DG2, good luck believing him

http://imgur.com/a/jwdZDtQ

 

[maddie]

Guys. Stop saying there's no Santa.

 

[mikk]

What doesn't make sense?

This:

He also can't rule out the possibility that Intel would have decided to move DG2 to N7P for...reasons.

Disregard any sort of opinions from him.

For the reasons I told you. And because it didn't make sense other people are helping you out. You should be aware that many of the specs can be read out from an ES sample and PCB, you can't simply do this for the underlying process.

 

[guidryp]

You missed the point. The point is it was unclear 3 months before launch on what process it's based on. It wasn't even clear if it's TSMC or Samsung. And here we have uzzi38 disregard everything from MLID just because he can't 100% confirm yet if it's really N6 (as rumored for a longer time) or N7P from TSMC.

No, you missed the point. MLID is BS. It shouldn't be used as a source of info.

My own personal guesses have been more accurate than MLID. The difference is I don't try to pass my guesses off as inside info to make a buck off it.

 

[IntelUser2000]

@Gideon

To be fair, MLID's leaks on Intel were more accurate than the rest. He was the first one to reveal the Cypress Cove codename for Rocketlake's core right? Same with Redwood Cove, though we'll see on that one.

There's nothing very off in MLIDs recent Xe leaks versus what's actually out there.

We can't outright dismiss everything he says because some of his sources might be correct, but some are feeding him misleading information. The problem is companies are absolutely refusing to reveal anything juicy but the desire of people to know what's happening in the future still has to be satiated.

If you want 100% accurate leaks you won't get that. You'll only get it if the company decides to tell you.

 

[guidryp]

@Gideon

To be fair, MLID's leaks on Intel were more accurate than the rest. He was the first one to reveal the Cypress Cove codename for Rocketlake's core right? Same with Redwood Cove, though we'll see on that one.

There's nothing very off in MLIDs recent Xe leaks versus what's actually out there.

We can't outright dismiss everything he says because some of his sources might be correct, but some are feeding him misleading information. The problem is companies are absolutely refusing to reveal anything juicy but the desire of people to know what's happening in the future still has to be satiated.

If you want 100% accurate leaks you won't get that. You'll only get it if the company decides to tell you.

There are much more accurate leakers.

Like Kopite7Kimi especially for NVidia products.
Like Gurman and Kuo especially Apple products.

The MLID guy comes across as insufferably arrogant, combined that with wild inaccuracy, making him a non starter.

Not being able to dismiss everything he says is meaningless. A stopped clock is correct twice/day, so in theory you can't always dismiss it, but it's still 100% useless for telling time.

 

[Tup3x]

Clearly he has some info about the cards - images seem to be legit. Everything else though... Who knows.

 

[IntelUser2000]

Not being able to dismiss everything he says is meaningless. A stopped clock is correct twice/day, so in theory you can't always dismiss it, but it's still 100% useless for telling time.

Yes but for Intel stuff he's been quite accurate. Granted not all, but trending in the 70%+.

Maybe for graphics it's different and he is accurate on CPUs, but we're on an Intel thread right?

 

[Tup3x]

Xe-HPG DG2:

https://twitter.com/i/web/status/1399966271182045184

 

[DrMrLordX]

Yeah I'd be scared too if I had to do driver optimizations for Intel. Hah. That's going to be the one thing everyone should be eyeing carefully once Intel manages to put DG2 on the market: how well will the drivers work? If the current gen iGPU drivers are any indicator, it could be kind of rough.

 

[guidryp]

From my perspective as a GoG gamer, I'd probably always stay far away from an Intel GPU.

For new games they seem to work well enough, but old games and Intel Drivers are a path to endless headaches.

 

[Dayman1225]

Xe-HPG DG2:

https://twitter.com/i/web/status/1399966271182045184

Estimates put it around 380-400 mm^2. Similar size to the RTX 3070 despite being on a smaller node (8nm SS vs 6nm TSMC). Rumours point to 3070-3070Ti perf for top model which isn’t too bad, if priced right it could be a great product for 1st gen.