News Intel Xe HPG Gaming GPU: Taking on AMD and Nvidia

[Admin]

Dedicated Intel Xe HPG graphics cards will launch in 2021, with hardware ray tracing support, and they won't be fabricated at Intel.

Intel Xe HPG Gaming GPU: Taking on AMD and Nvidia : Read more

 

[warezme]

Intels Achilles heel has always been drivers. If they haven't sorted that out, the hardware won't matter.

 

[jimmysmitty]

Intels Achilles heel has always been drivers. If they haven't sorted that out, the hardware won't matter.

In GPUs, yes. But they also have never really done more than just had a half decent iGPU. In others theirs are typically pretty good.

I hope they can break in and kill the duopoly that AMD and nVidia have. AMD and nVidia have price fixed before and having a third party in to push competition and possibly get us better pricing is never a bad thing.

 

[JarredWaltonGPU]

Intels Achilles heel has always been drivers. If they haven't sorted that out, the hardware won't matter.

I discussed this in the Xe LP article, but the latest drivers have been pretty good for me. There were one or two issues I think, but with UHD 630 and Iris Plus, every game I tested at least worked. I also discussed the drivers in the recent integrated graphics testing.

https://www.tomshardware.com/uk/news/intel-xe-lp-graphics-specs
https://www.tomshardware.com/features/intel-gen11-core_i7_1065g7-tested
https://www.tomshardware.com/features/amd-vs-intel-integrated-graphics

It's by no means a 'solved' problem, but compatibility is not nearly as bad as in the Haswell and earlier days.

 

[TerryLaze]

I hope they can break in and kill the duopoly that AMD and nVidia have. AMD and nVidia have price fixed before and having a third party in to push competition and possibly get us better pricing is never a bad thing.

Yeah,why do I doubt that intel will somehow have aggressive pricing on...anything.
I mean I wish but it's very unlikely,it's not like intel has to make sales to survive,especially desktop/gaming GPUs.

 

[Chung Leong]

Seems like a mistake to not use HBM. A first generation product is not going to achieve high volume in any event, so why launch something mediocre? Given Nvidia's market position and advantage in game optimization, an Intel solution has to have 50% more raw horsepower to be competitive. I'm not seeing that here.

 

[jimmysmitty]

Yeah,why do I doubt that intel will somehow have aggressive pricing on...anything.
I mean I wish but it's very unlikely,it's not like intel has to make sales to survive,especially desktop/gaming GPUs.

It depends on the case. Core 2 was very aggressive price wise. The Core 2 Quad Q6600 was much cheaper and outperformed the Quad FX setup, AMDs then equivalent quad core system pre-Phenom.

In this case I would like to think since Intel is new to the market so to speak they would be more aggressive to start.

 

[AnimeMania]

I think Intel's Achilles heel has been killing off ideas/products before they have a chance to catch on. Nobody wants to buy a Graphics Card that won't be supported a year from now.

 

[TCA_ChinChin]

AMD and nVidia have price fixed before and having a third party in to push competition and possibly get us better pricing is never a bad thing.

When did they price fix? Having Intel compete is nice though.

 

[spongiemaster]

When did they price fix? Having Intel compete is nice though.

I don't think they ever have. It's more of a case of non-competing for the past few years. Nvidia has chosen whatever price can maintain their margins because AMD can't compete. Then a year or two later, AMD releases cards that just slightly undercut Nvidia on price, but not by enough to warrant any real response from Nvidia, keeping ASP's inflated.

 

[JarredWaltonGPU]

Seems like a mistake to not use HBM. A first generation product is not going to achieve high volume in any event, so why launch something mediocre? Given Nvidia's market position and advantage in game optimization, an Intel solution has to have 50% more raw horsepower to be competitive. I'm not seeing that here.

HBM is terribly expensive. Even AMD learned its lessons and has backed off of pushing HBM2 in the consumer segment. It has its use cases, like enterprise / HPC stuff, or when you want to use as little area as possible (eg, Navi 12 for Apple MacBook), but the price/performance just isn't necessary or beneficial for consumer products.

As far as a raw horsepower advantage, I think you're stuck in the past. Iris Plus Gen11 graphics has about 1.1 TFLOPS of computational power. It performs basically like a 1.1 TFLOPS AMD or Nvidia GPU. By which I mean that it's woefully underpowered, only half the performance of a lowly GTX 1050. Intel really needs more ALUs / EUs for its graphics ambitions, and it looks like Xe HPG may do the trick. I don't expect a 10 TFLOPS Intel part to outperform a 10 TFLOPS Nvidia part, but a 15 TFLOPS Xe HPG could probably put up a good fight against RTX 2080 Ti.

We definitely need to see Xe HPG in action before coming to any conclusions, though. Even Xe LP won't really tell us what to expect from Xe HPG, since HPG is going to reworks the architecture for sure. Fundamentally, it's just a matter of getting optimized ALU cores that can run gaming workloads fast, and Intel has the R&D resources to try and make that happen. Whether it can succeed (and overcome internal politics) is a major question.

 

[TerryLaze]

It depends on the case. Core 2 was very aggressive price wise. The Core 2 Quad Q6600 was much cheaper and outperformed the Quad FX setup, AMDs then equivalent quad core system pre-Phenom.

In this case I would like to think since Intel is new to the market so to speak they would be more aggressive to start.

Well back in 2006 intel had quite a few consecutive quarters of decline and AMD was still doing very well financially because they hadn't bought ATI yet so intel had to be aggressive.
Now there is nothing that forces intel's hand,they are going to sell any quantity they want to laptop/portable and server/compute is what these were designed for in the first place so it's going to sell well there as well.

With the current situation it would only hurt intel if they sell out of gaming GPUs and then can't make any more because their fab space is filled out already,it's better for them to make them out to be premium things with high price just to show what they can do and whenever they can make them in bulk they can lower prices.

 

[Chung Leong]

HBM is terribly expensive. Even AMD learned its lessons and has backed off of pushing HBM2 in the consumer segment. It has its use cases, like enterprise / HPC stuff, or when you want to use as little area as possible (eg, Navi 12 for Apple MacBook), but the price/performance just isn't necessary or beneficial for consumer products.

Unloading a pile of graphic cards at steep discounts is terribly expensive too. Intel is trying to break into a new market. It can't take consumer interest for granted. Unless Xe HPG is a lot better than the Nvidia and AMD counterparts, it's going to stay on the shelf.

 

[JarredWaltonGPU]

Unloading a pile of graphic cards at steep discounts is terribly expensive too. Intel is trying to break into a new market. It can't take consumer interest for granted. Unless Xe HPG is a lot better than the Nvidia and AMD counterparts, it's going to stay on the shelf.

I suspect there's going to be a lot of partner deals with the OEMs by Intel. Because as much ground as AMD has made up, it's still far, far behind in terms of volumes in the OEM space. Of course, how to convince OEMs to use a dedicated GPU is still an issue -- if you have a "good enough" GPU for general use, you don't need a dedicated GPU. I can't see most businesses wanting to add a dedicated GPU at all, whether AMD, Intel, or Nvidia. I worked in IT for years, and it was always integrated graphics and lowest common denominator stuff.

Still, unless HBM use would massively boost the GPU's performance (and it wouldn't, since GDDR6 is hitting similar bandwidths, or at least close enough), adding it would increase costs and make the card even less desirable.

 

[Chung Leong]

I suspect there's going to be a lot of partner deals with the OEMs by Intel. Because as much ground as AMD has made up, it's still far, far behind in terms of volumes in the OEM space. Of course, how to convince OEMs to use a dedicated GPU is still an issue -- if you have a "good enough" GPU for general use, you don't need a dedicated GPU. I can't see most businesses wanting to add a dedicated GPU at all, whether AMD, Intel, or Nvidia. I worked in IT for years, and it was always integrated graphics and lowest common denominator stuff.

Gaming laptop is small but growing segment. That's a much easier nut for Intel to crack than the dGPU market. I wonder how many video card makers will make Xe HPG products. The answer could conceivably be zero. There just isn't much of an upside. Selling a Xe means not selling an RTX or Radeon. Intel will have to spread a lot of co-marketing dollars to get manufacturers to sign up.

 

[TerryLaze]

I wonder how many video card makers will make Xe HPG products. The answer could conceivably be zero. There just isn't much of an upside.

Why do you think that intel would want anybody else to make Xe products?!
Or even that they would be able to make so many chips that they could sell them to 3rd party.
Intel has fabs and all the means to make the GPUs by themselves as fast as they can which would probably be faster than waiting on some company that has to first make a bunch of AMD and Nvidia cards,they will start with what would be reference models if the other companies would be making them.

 

[Chung Leong]

Why do you think that intel would want anybody else to make Xe products?!
Or even that they would be able to make so many chips that they could sell them to 3rd party.

If Intel doesn't intend to sell many chips then it wouldn't tap TSMC. Unrealistic optimism seems to be the driving force in the company these days.

 

[JarredWaltonGPU]

Why do you think that intel would want anybody else to make Xe products?!
Or even that they would be able to make so many chips that they could sell them to 3rd party.
Intel has fabs and all the means to make the GPUs by themselves as fast as they can which would probably be faster than waiting on some company that has to first make a bunch of AMD and Nvidia cards,they will start with what would be reference models if the other companies would be making them.

Intel announced plans to work with third party AIB partners for Xe HPG, and I'm sure Asus, Gigabyte, MSI, and others will at least give it a shot -- might be like a weak sauce Radeon VII release where everyone only sells a few hundred cards that are straight up copies of the reference model, though. It will depend on what sort of price and performance the Xe HPG delivers.

 

[TerryLaze]

If Intel doesn't intend to sell many chips then it wouldn't tap TSMC. Unrealistic optimism seems to be the driving force in the company these days.

I can't know how many wafers intel can produce to know if the amount they got from tsmc is many or just a handful but my guess is it's not that many.
Also looking at intel's sale numbers their optimism isn't unrealistic intel made the exact same amount of money the last two years which means either a ginormous coincidence or intel producing the maximum amount of product they are able to and selling it all.

 

[Chung Leong]

Intel announced plans to work with third party AIB partners for Xe HPG, and I'm sure Asus, Gigabyte, MSI, and others will at least give it a shot -- might be like a weak sauce Radeon VII release where everyone only sells a few hundred cards that are straight up copies of the reference model, though. It will depend on what sort of price and performance the Xe HPG delivers.

Supporting another AMD chip requires far less effort than supporting a brand new platform. Plus AMD has a loyal fan base. We're looking at a situation where even if the Xe HPG is priced at zero, there'll be no taker. I think Intel will have to toss a lot of dirt cheap Xe LP chips into the mix to make the effort worthwhile. Maybe that's why Intel contracted out to TSMC. They don't want their precious fabs tied up making zero-margin parts.

 

[TerryLaze]

Supporting another AMD chip requires far less effort than supporting a brand new platform. Plus AMD has a loyal fan base. We're looking at a situation where even if the Xe HPG is priced at zero, there'll be no taker. I think Intel will have to toss a lot of dirt cheap Xe LP chips into the mix to make the effort worthwhile. Maybe that's why Intel contracted out to TSMC. They don't want their precious fabs tied up making zero-margin parts.

Oh come on now, you don't even believe that yourself.
AMD has a loyal fan base and what?! intel that does 72B of sales a year doesn't?
Any 3rd party GPU maker will sell any intel card for the name alone even at a high price,no matter how much you personally think that it sucks but people see intel as sign of quality, just look at CPU sales and any other sales intel does.
Yes intel doesn't want to lose fab space for low margin cards which is the only reason why intel might not make a lot of gaming chips and why we might not see more than a few token 3rd party cards.

 

[Chung Leong]

Any 3rd party GPU maker will sell any intel card for the name alone even at a high price,no matter how much you personally think that it sucks but people see intel as sign of quality, just look at CPU sales and any other sales intel does.

In the sub-$100 segment, I'm sure cards with the Intel name on it would sell like hotcakes. That's why I said the availability of Xe LP at a low price might be a condition for AIB participation. Once we get into the budget-gaming segment though, consumers are far more discerning. They want assurance that games will actually work with the card. The enthusiast segment is tougher still. Most consumers here have previously owned and are using an Nvidia or AMD product. An Xe HPG halo product is bound to move very slowly. Without volume, of course, price is not much of a lever.

The situation would be different if Intel releases something that crushes the RTX 3090. Bragging right is worth a lot to people in this day and age. That seems extremely unlikely at this point.

 

[JayNor]

Intel is busy converting their image processing apps to use their new GPUs. They claim the film/animation industry uses these apps already.

https://software.intel.com/content/www/us/en/develop/tools/oneapi/rendering-toolkit.html

 

[shrapnel_indie]

They haven't really released a single GPU of this new lineup... yet. I'll believe performance numbers once the hardware is actually on the market.

 

[jimmysmitty]

When did they price fix? Having Intel compete is nice though.

https://www.tomshardware.com/news/n...e fixing of GPU processors and graphics cards.

I would have to look at where this went but it was enough to get a case started.

Ironically, in the emails they also stated Intel would never release a dGPU.

Well back in 2006 intel had quite a few consecutive quarters of decline and AMD was still doing very well financially because they hadn't bought ATI yet so intel had to be aggressive.
Now there is nothing that forces intel's hand,they are going to sell any quantity they want to laptop/portable and server/compute is what these were designed for in the first place so it's going to sell well there as well.

With the current situation it would only hurt intel if they sell out of gaming GPUs and then can't make any more because their fab space is filled out already,it's better for them to make them out to be premium things with high price just to show what they can do and whenever they can make them in bulk they can lower prices.

AMD was doing well but also had capacity issues with OEM and vendor channels. Some smaller OEMs and vendors also got looked over for larger ones when K8 was doing well. But a lot of this came from HPC and server markets as AMD was vastly more competitive during that period than K10 or Bulldozer, hell they dropped out of that market until Zen leaving Intel with the whole market to cash in on.

I honestly think intel will do whatever they can. I think they plan to hit the entire market and will still market it as their all in one solution. I am interested to see if they have figured out a way to use a iGPU with their dGPU. That would be quite an advantage if the extra cores on a desktop CPU can be used with a dGPU to offload some work. Mainly against nVidia as AMD would probably come up with a similar solution.

They haven't really released a single GPU of this new lineup... yet. I'll believe performance numbers once the hardware is actually on the market.

No but any performance numbers from Tiger Lake should give us an idea. Not an exact idea but enough to extrapolate and increase based on moving to GDDR6X. And that should be early September for that.

However with the rumors of the RTX 3090 I think Intel wont quite be in the realm of taking names.