Intel Teases Discrete GPU - "Arctic Sound"

[ZcanKal]

28 minutes ago, Drak3 said:

If you're talking about Intel, they own x86. They license out x86 to AMD, in exchange for AMD licensing out AMD64/x64 to Intel. Without each other, both are basically worthless to the modern world.

would a 32 bit co-processor be that big of a bottleneck though?  Weren't the 900 series' co-processors 32 bit ARM?

 

24 minutes ago, Okjoek said:

Intel is huge in the mobile market. wouldn't surprise me if they go for some 75W or less solutions that compete with things like the RX 550, 560 ,GT1030, GTX 1050

Starting small would end up well.

[Ryujin2003]

Shouldn't there be a summary to go along with the news post?

[Starelementpoke]

Now I just want Intel to end up with VEGA 2: The VEGAING

[Shreyas1]

2 hours ago, Godlygamer23 said:

People rip on Intel for creating sub-par IGPs, but they definitely have the ability to create something potent. 

I mean, they have the money and most of the GPU market share so why not

[TheRandomness]

People seem to have forgotten that Larrabee existed. Too bad it never got out the factory doors.

[ignaloidas]

2 hours ago, i_got_laid_by_a_dragoness said:

And apologies if this was already posted, went through the first 5 pages of News and didn't see anything about it.

You could've used search...

but I guess that works too.

[ZcanKal]

5 minutes ago, TheRandomness said:

People seem to have forgotten that Larrabee existed. Too bad it never got out the factory doors.

It has been discussed on this thread.  We're talking about a dGPU that actually releases, since they got Raja, it seems that they're actually going through with it this time.

[Nowak]

2 hours ago, i_got_laid_by_a_dragoness said:

AMD names them whatever the fuck they want

AMD names their graphics architectures after stars. Used to be islands, but Polaris, Vega and Navi are stars.

[Sniperfox47]

7 minutes ago, Daring said:

AMD names their graphics architectures after stars. Used to be islands, but Polaris, Vega and Navi are stars.

Yeah. Used to be islands, now stars. I.e. whatever the heck they want since those two things are totally unrelated.

[Razor01]

Funny thing is Intel's Larrabee project also had a Polaris code name chip lol.

[Humbug]

2 hours ago, M.Yurizaki said:

Part of me doesn't really expect them to launch with a high-end beast to compete with the 70-80-Titans of NVIDIA. But if they can break out a solid $300-$400 competitor, then things will get really interesting.

Safe bet that they will not be able to compete with Nvidia and AMD in their first gen. But this is Intel after all. So if they stay at it for a few generations I think they will compete.

 

LOL it's so funny, 8 years ago everybody in the industry was like 'discrete GPUs will be dinosaurs', 'the future is fusion', 'integrated graphics are getting better so fast' they will take over compute and gaming etc

 

Fast forward to 2018. The gap between integrated graphics and discrete graphics performance is as big as ever. AMD and Nvidia cannot even produce enough discrete GPUs to satisfy the demand, and Intel wants to get in on the action. LOL, industry experts did not predict this.

 

[ZcanKal]

5 minutes ago, Humbug said:

Safe bet that they will not be able to compete with Nvidia and AMD in their first gen. But this is Intel after all. So if they stay at it for a few generations I think they will compete.

 

LOL it's so funny, 8 years ago everybody in the industry was like 'discrete GPUs will be dinosaurs', 'the future is fusion', 'integrated graphics are getting better so fast' they will take over compute and gaming etc

 

Fast forward to 2018. The gap between integrated graphics and discrete graphics performance is as big as ever. AMD and Nvidia cannot even produce enough discrete GPUs to satisfy the demand, and Intel wants to get in on the action. LOL, industry experts did not predict this.

 

People also believed for quite some time that the earth couldn't sustain more than 1 billion people, but see how that worked out.

[Mira Yurizaki]

7 minutes ago, Humbug said:

Safe bet that they will not be able to compete with Nvidia and AMD in their first gen. But this is Intel after all. So if they stay at it for a few generations I think they will compete.

 

LOL it's so funny, 8 years ago everybody in the industry was like 'discrete GPUs will be dinosaurs', 'the future is fusion', 'integrated graphics are getting better so fast' they will take over compute and gaming etc

 

Fast forward to 2018. The gap between integrated graphics and discrete graphics performance is as big as ever. AMD and Nvidia cannot even produce enough discrete GPUs to satisfy the demand, and Intel wants to get in on the action. LOL, industry experts did not predict this.

 

It's probably because people didn't realize that sticking a 150W TDP GPU with a 95W TDP GPU isn't exactly a... Wait never mind, AMD thought releasing a 220W TDP processor was a good idea.

 

EDIT: It's the FX-9590 if anyone is curious.

[Taf the Ghost]

3 minutes ago, Humbug said:

Safe bet that they will not be able to compete with Nvidia and AMD in their first gen. But this is Intel after all. So if they stay at it for a few generations I think they will compete.

 

LOL it's so funny, 8 years ago everybody in the industry was like 'discrete GPUs will be dinosaurs', 'the future is fusion', 'integrated graphics are getting better so fast' they will take over compute and gaming etc

 

Fast forward to 2018. The gap between integrated graphics and discrete graphics performance is as big as ever. AMD and Nvidia cannot even produce enough discrete GPUs to satisfy the demand, and Intel wants to get in on the action. LOL, industry experts did not predict this.

 

dGPU sales have actually been on a rather steep decline from the late 2000s. 

 

https://www.anandtech.com/show/10613/discrete-desktop-gpu-market-trends-q2-2016-amd-grabs-market-share-but-nvidia-remains-on-top

 

What changed is the OEM space, not the Consumer space. There's also this issue that GPUs haven't been increasing in performance at nearly the rate since that period. They met a sufficiency level (ability to play 1080p video content) and sales have been down since, along with people switching to mobile. Which is partially why people forget that Intel is still sells more GPUs than both AMD & Nvidia combined, and I believe it's been that ways since around 2000.  First the motherboard-based parts, then the iGPUs.

 

Issue for dGPUs is drivers. It's all about the drivers.

[Taf the Ghost]

12 minutes ago, ZcanKal said:

People also believed for quite some time that the earth couldn't sustain more than 1 billion people, but see how that worked out.

They weren't wrong, though oddly the person most associated with those statements made them well after the problem was solved with technology.

[Humbug]

5 minutes ago, Taf the Ghost said:

Which is partially why people forget that Intel is still sells more GPUs than both AMD & Nvidia combined, and I believe it's been that ways since around 2000.  First the motherboard-based parts, then the iGPUs.

 

Issue for dGPUs is drivers. It's all about the drivers.

Oh ya Intel sells the most GPUs.

But they always sell GPUs to people who just wanna put a display on screen. Not for people who want to utilize the power of parallel processing for compute and graphics applications.

[Dabombinable]

As long as the thing doesn't rely on system memory like a piece of shit again, I'll be happy (that's why the i740 had so little onboard vRAM-it relied heavily on the AGP BUS+system RAM).

[Mihle]

"Enter the market with a bang"

 

Does that mean it will have a self destruct feature?

 

[ScratchCat]

1 hour ago, Mihle said:

"Enter the market with a bang"

 

Does that mean it will have a self destruct feature?

 

In other news - Intel cross licenses patents with Samsung, specifically those about the Note 7.

3 hours ago, M.Yurizaki said:

It's probably because people didn't realize that sticking a 150W TDP GPU with a 95W TDP GPU isn't exactly a... Wait never mind, AMD thought releasing a 220W TDP processor was a good idea.

 

EDIT: It's the FX-9590 if anyone is curious.

IBM can be worse however they do design CPUs for datacenters where noise and mass of the cooling solution are less of a priority.

Quote

A 12 core POWER8 is "limited" to 3.1 GHz if you want to stay below the 190W TDP mark. Clockspeeds higher than 4 GHz are only possible with 8-cores and a 250W TDP. This makes us really curious what kind of power dissipation we may expect from the 4.2 GHz 10-core POWER8 inside the expensive E870 Enterprise systems (300W?).  

https://www.anandtech.com/show/9193/the-xeon-e78800-v3-review/6

 

4 hours ago, Daring said:

AMD names their graphics architectures after stars. Used to be islands, but Polaris, Vega and Navi are stars.

Don't forget the Evergreen series - Cypress, Cedar, Juniper.

In fact they named some after mythical Gods i.e. Zeus, Thor, Loki.

Patterns are not their strong point:

Edit: The last two point do not seem reliable.

[leadeater]

6 hours ago, Humbug said:

Fast forward to 2018. The gap between integrated graphics and discrete graphics performance is as big as ever. AMD and Nvidia cannot even produce enough discrete GPUs to satisfy the demand, and Intel wants to get in on the action. LOL, industry experts did not predict this.

Well I wouldn't call tech journalists industry experts personally lol. They have a track record of being total idiots when it comes to predicting future trends. Single package HSA really could only ever have been a thing for mobile and small form factor devices,.

[leadeater]

2 hours ago, ScratchCat said:

IBM can be worse however they do design CPUs for datacenters where noise and mass of the cooling solution are less of a priority.

Intel Xeons are no less power hungry at full tilt, TDP ratings on those are a total joke. 250W-400W per CPU depending on model and generation is normal and that's before the big power jump for AVX-512 which will push 400W-500W per CPU.

[Trixanity]

1 minute ago, leadeater said:

Intel Xeons are no less power hungry at full tilt, TDP ratings on those are a total joke. 250W-400W per CPU depending on model and generation is normal and that's before the big power jump for AVX-512 which will push 400W-500W per CPU.

I was under the impression that generally Intel's TDP are for base clocks (or at best some form of single core boost) meaning it's utterly useless. I'm not saying there's a right way to do it but that's just ridiculous. They could at least be realistic about it.

[leadeater]

8 minutes ago, Trixanity said:

I was under the impression that generally Intel's TDP are for base clocks (or at best some form of single core boost) meaning it's utterly useless. I'm not saying there's a right way to do it but that's just ridiculous. They could at least be realistic about it.

Personally I don't know when the last time I looked at TDP for Intel Xeons, it's just so pointless. Desktop CPUs aren't any better but I think X299 took the crown away from Xeons recently for the most "You are kidding right? That's not even close to reality".

[AlexGoesHigh]

Not to say these won't be great but no matter how great they will be it's going to be for shit if the drivers and support for it aren't great amd and Nvidia still support 5+ year old hardware where as Intel is probably not even updating Skylake drivers (I'm saying this as how Intel has  supported stuff in the past idk if Skylake still has support or not) when are just 3 years old and even when they do support stuff they take forever to fix stuff and that's outside of games that is something of its own topic.

 

I'm not saying they should go all the way to they point of releasing a new game ready drivers on every triple a release in order to succeed but a once a month or bimonthly + fix stuff and release on a timely manner would be great because the way they support their drivers now for the integrated GPU is embarrassing for a company the size of Intel.

[porina]

35 minutes ago, leadeater said:

Intel Xeons are no less power hungry at full tilt, TDP ratings on those are a total joke. 250W-400W per CPU depending on model and generation is normal and that's before the big power jump for AVX-512 which will push 400W-500W per CPU.

If you have code that can make use of AVX-512 (at least for FP64 ops), you're getting roughly 2x the peak IPC compared to Intel FMA3, or 4x that of Zen. Don't know if anyone has worked out the performance per watt for it though.

25 minutes ago, leadeater said:

Personally I don't know when the last time I looked at TDP for Intel Xeons, it's just so pointless. Desktop CPUs aren't any better but I think X299 took the crown away from Xeons recently for the most "You are kidding right? That's not even close to reality".

Is this that point in the thread where someone goes that TDP is not equal to expected power consumption? Doh, I just did it...

 

For benching the 7800X is certainly a beast running AVX-512, I think I got it to about 4.5 GHz compared to a borderline throttling 4.9 non-AVX workloads. I was on the limits with ambient custom loop cooling (not seriously tried sub-ambient yet). I hate to imagine the heat the top end Skylake-X CPU would put out at 3x the cores!

 

To swing it vaguely back to topic, I hope whatever GPU Intel eventually comes up with, it if could do FP64 AVX-like code natively, it could be very interesting for compute uses that nvidia and AMD haven't really tapped into. Personally not interested in the trend towards low precision for machine learning and similar applications.