News Intel GPUs - Falcon Shores cancelled

[IntelUser2000]

That die size has to be wrong, right?

That's package size not die size. Die is roughly quarter of that or 400mm2. If someone does a perspective correction for measurements they should be able to get the figure quite accurate.

 

[Hitman928]

That's package size not die size. Die is roughly quarter of that or 400mm2. If someone does a perspective correction for measurements they should be able to get the figure quite accurate.

That makes a lot more sense. So roughly the same size as GA104.

 

[IntelUser2000]

That makes a lot more sense. So roughly the same size as GA104.

Yes, package sizes are usually quoted a mm x b mm, while die sizes are shown as c mm2.

Also you can just tell. Package sizes are generally way out of range so just by that you can figure out whether it's talking about the die or package. Of course you'd compensate for other variables like whether it's a desktop CPU or a one that'll go in smartphones. Lakefield for example is only 12mm x 12mm for the package size, then you need to do additional search.

Many tech sites make a mistake of not distinguishing between the two(probably because they don't understand it that well).

I think while I believe Anandtech articles have deteriorated in quality over the years I appreciate them not putting up speculation/rumor as part of their news feed.

 

[jpiniero]

Good. If it was 1:2 it would have been absolutely pathetic in the one area it was supposed to be good at - HPC.

They don't seem to care much about DP anymore since AI doesn't use it.

 

[uzzi38]

They don't seem to care much about DP anymore since AI doesn't use it.

No way.

AI workloads in HPC are growing, but FP64 is still the most important by a very long shot.

 

[Glo.]

The leaker also being surprised by the size, seems close to 400mm2, that's what I could tell......

Not surprising considering 384 EU SKU is derivwed from the full 512 EU's, and 512 EU DG2 has close to 400mm2 die size...

 

[jpiniero]

No way.

AI workloads in HPC are growing, but FP64 is still the most important by a very long shot.

That's the impression I'm getting from marketing. Plus the small increase in FP64 performance in Ampere. 45 TF of FP64 might be better than Hopper.

 

[7beauties]

Friends, school me if I'm wrong, but my understanding is that Intel is 'deflating their footballs'. Raja Kadouri may be brilliant and he knows computer graphics very, very well, but Intel's upcoming GPUs will only work optimally with Intel CPUs and chipsets but not nearly as well on Ryzen platforms. If this is true, it's a massive gyp. Imagine buying a Toyota Corolla only to realize you must use their special blend of gas or their specific tires to run well. Wouldn't that piss you off?

 

[jpiniero]

Friends, school me if I'm wrong, but my understanding is that Intel is 'deflating their footballs'. Raja Kadouri may be brilliant and he knows computer graphics very, very well, but Intel's upcoming GPUs will only work optimally with Intel CPUs and chipsets but not nearly as well on Ryzen platforms. If this is true, it's a massive gyp. Imagine buying a Toyota Corolla only to realize you must use their special blend of gas or their specific tires to run well. Wouldn't that piss you off?

There really hasn't been much on rumors of actual gaming performance in general. Far to early to figure out if there is anything you suggest.

 

[gdansk]

Friends, school me if I'm wrong, but my understanding is that Intel is 'deflating their footballs'. Raja Kadouri may be brilliant and he knows computer graphics very, very well, but Intel's upcoming GPUs will only work optimally with Intel CPUs and chipsets but not nearly as well on Ryzen platforms. If this is true, it's a massive gyp. Imagine buying a Toyota Corolla only to realize you must use their special blend of gas or their specific tires to run well. Wouldn't that piss you off?

That's all unlikely. Intel will not do that. What gave you that idea? DG1 laptop sales requirements?

 

[Glo.]

Friends, school me if I'm wrong, but my understanding is that Intel is 'deflating their footballs'. Raja Kadouri may be brilliant and he knows computer graphics very, very well, but Intel's upcoming GPUs will only work optimally with Intel CPUs and chipsets but not nearly as well on Ryzen platforms. If this is true, it's a massive gyp. Imagine buying a Toyota Corolla only to realize you must use their special blend of gas or their specific tires to run well. Wouldn't that piss you off?

Why would it not work with AMD CPUs?

Does it have special SIMDs to gimp performance of actual Intel GPUs? Drivers would gimp performance on AMD platform?

That would only harm Intel, not AMD. Can you imagine a backlash that people would grill Intel, with?

If Intel is serious about dGPUs(which, they are) they will go all out, regardless of platform.

P.S. I'm gonna tell you guys right now. The actual combo of AMD CPU and Intel dGPU is what I am gonna build next year.

 

[DeathReborn]

There really hasn't been much on rumors of actual gaming performance in general. Far to early to figure out if there is anything you suggest.

I heard that MLID says the drivers are "horrible" (worse than VEGA was 6 months before launch) and performance is around 3060.

 

[VirtualLarry]

Why would it not work with AMD CPUs?

Why didn't Intel's then-premier AX200 Wifi 6 m.2 card, work with anything besides intel 10th (and now, 11th)-gen CPUs and their associated chipsets?

It wouldn't surprise me in the least if Intel's new dGPUs only function in an all-Intel system.

 

[moonbogg]

I look forward to pictures of pallets of Intel GPUs in a single miner's garage.

 

[PingSpike]

Why didn't Intel's then-premier AX200 Wifi 6 m.2 card, work with anything besides intel 10th (and now, 11th)-gen CPUs and their associated chipsets?

It wouldn't surprise me in the least if Intel's new dGPUs only function in an all-Intel system.

Yeah, people are keep saying "Intel wouldn't do that". First of all, they already did with DG1 and second of all Intel totally loves doing exactly that with everything they make. Its a bad idea, but it always been a bad idea and they keep doing it anyway. So you really can't dismiss it out of hand.

 

[DrMrLordX]

Why didn't Intel's then-premier AX200 Wifi 6 m.2 card, work with anything besides intel 10th (and now, 11th)-gen CPUs and their associated chipsets?

Eh wot

My x570 motherboard has an AX200 built into it. It works, though the AX200 itself is buggy. It's also buggy in Intel machines.

 

[IntelUser2000]

Why didn't Intel's then-premier AX200 Wifi 6 m.2 card, work with anything besides intel 10th (and now, 11th)-gen CPUs and their associated chipsets?

It wouldn't surprise me in the least if Intel's new dGPUs only function in an all-Intel system.

That's because those are actually AX201s which are CNVi and only exist on Intel chipsets. You could say the AX201s give Intel and advantage because having some of its functions on a chipset saves real estate and cost but that's a different story.

The AX200 with everything, including AMD. I use the AX200 on a Broadwell laptop. Works great.

First of all, they already did with DG1 and second of all Intel totally loves doing exactly that with everything they make. Its a bad idea, but it always been a bad idea and they keep doing it anyway. So you really can't dismiss it out of hand.

DG1 doesn't count. DG1 isn't artificial segmentation, it just lacks a video BIOS chip that let's it boot without a specific chipset requirement.

I assume for DG1 they wanted to get a "discrete GPU" out fast so they hacked it quick and dirty.

It doesn't mean they can't artificially segment it to work with on their GPUs but hopefully they are not that stupid. I know that they know as a newcomer you need it to be open source and work with everyone.

They are great with Linux open source for example.

 

[Tup3x]

DG1 was a test run and only used in certain OEM builds. Obviously Arc cards do not have those restrictions (special OEM stuff aside).

 

[Glo.]

I look forward to pictures of pallets of Intel GPUs in a single miner's garage.

Jokes aside, how's current state of crypto mining algorithms on Intel GPUs?

Still doesn't work, at all?

 

[moonbogg]

Jokes aside, how's current state of crypto mining algorithms on Intel GPUs?

Still doesn't work, at all?

If it's broke, they will fix it. I fully expect these to be just another mining product. Gamers can't compete on price with people who see GPUs as money trees.

 

[DrMrLordX]

Jokes aside, how's current state of crypto mining algorithms on Intel GPUs?

Still doesn't work, at all?

From what I have read, there were a few ways to get some OpenCL-based mining software to work on Intel iGPUs if the software specifically targeted Intel iGPUs. Standard OpenCL software targeting an inspecific OpenCL-compliant iGPU/dGPU didn't run. Intel did something with their drivers in 2020 that broke all of that software. So if you want to mine on an Intel iGPU today, you can, but only if there are older drivers available for it.

 

[IntelUser2000]

Jokes aside, how's current state of crypto mining algorithms on Intel GPUs?

Still doesn't work, at all?

Don't think it mattered at this point. What would you mine on an iGPU? Without high bandwidth dedicated VRAM crypto mining is pointless.

You can mine on AMD iGPU because its a derivative of the dGPU part. Since it uses the same drivers and software it's supported simple as that.

Not that it matters. Back when the DAG file was small enough you could get 3-4MH/s @ 50W? iGPUs are efficient for gaming but on crypto mining it falls completely flat on it's face.

 

[Mopetar]

Yeah those measurements have to be off. It would make the die close to 1,600 mm^2 which would be larger than anything ever released by either AMD or Nvidia by about double.

You also wouldn't be able to fit very many of them on a wafer. Probably fewer than 30 from napkin math.

 

[jpiniero]

Not that it matters. Back when the DAG file was small enough you could get 3-4MH/s @ 50W? iGPUs are efficient for gaming but on crypto mining it falls completely flat on it's face.

I think the 5700G does 4 MH @ 25 W which is profitable at current ETH prices but barely.

 

[aigomorla]

I think while I believe Anandtech articles have deteriorated in quality over the years I appreciate them not putting up speculation/rumor as part of their news feed.

Ian Cutress making his own youtube channel and not cross posting on AT, speaks for itself on how far anandtech has gone down minus the forums.

https://www.youtube.com/c/TechTechPotato