Intel CEO confirms first dGPUs in 2020

[Dayman1225]

Intel makes it a three-way race with AMD and Nvidia on graphics chips

Intel's CEO Brian Krzanich disclosed during an analyst event last week that it will have its first discrete graphics chips available in 2020. This will mark the beginning of the chip giant’s journey toward a portfolio of high-performance graphics products for various markets including gaming, data center and artificial intelligence (AI).
[...]
Intel’s executive vice president of the data center group, Navin Shenoy, confirmed that the company’s strategy will include solutions for data center segments (think AI, machine learning) along with client (think gaming, professional development).

Interesting to say the least. I am excited for more competition in this space and I hope Intel are able to rival AMD and Nvidia, they certainly have the manpower and R&D $$ to do so. Now lets just wait and see how they execute.

 

[Dayman1225]

Confirmation from Intel on Twitter

Intel's first discrete GPU coming in 2020

 

[HutchinsonJC]

I'm trying to recall, Intel has licensed some things from Nvidia for some 20 years? Is that enough to compete? What are thoughts on that?

And if it wasn't for that agreement, this would be a seemingly impossible endeavor no?

 

[Headfoot]

Licensing and actually using technology are actually pretty far apart. They are probably using in house tech but have to pay out licenses due to patents, even if they wont use any materials provided by nVidia (or others). Realistically, it's probably a little of column A and little of column B, some parts they probably 'build' and others they 'buy'

 

[Qwertilot]

They could be good, but I'd be a bit sceptical a priori - it isn't like they're going to have any kind of process lead and there's a lot of technological catch up to do.

 

[f2bnp]

Let's wait and see what comes of this. Intel has tried it twice before and didn't really nail it, perhaps they will this time.

The i740 was actually a nice chip for its time, it was however hyped to the moon and when it came out the combination of price/performance wasn't ideal. Not a bad chip overall, had Intel worked on a 2nd iteration/followup I think we would have seen a really great product. Its legacy lived on the i810 integrated GPU and then later on the GMA chips.

Then there was Larabee which never made it to the market, but what a monster that was. Someone got hold of one somewhat recently and was trying to get it running.
https://www.vogons.org/viewtopic.php?f=63&t=57207

 

[cmdrdredd]

We don't know what segment they will target. They may not ever compete on the higher end, meaning nothing to compare to a 1080 performance level, but they may have very competitive lower end models. Who knows.

 

[SPBHM]

it will be really interesting, they are already making IGPs that are as big as many dGPUs, GT4e Skylake is like 160mm2 of GPU and that's excluding the edram

 

[tviceman]

Hopefully this gets Nvidia off of resting on the laurels in the consumer segment with their tech lead over AMD.

 

[maddie]

it will be really interesting, they are already making IGPs that are as big as many dGPUs, GT4e Skylake is like 160mm2 of GPU and that's excluding the edram

Saying it like that sounds terrible.

 

[LightningZ71]

I don't think that Intel will venture too far into the top end of the dGPU market. Up until this year, it was arguable that they might have been able to produce a dGPU chip that was "good enough" and use their process advantage to muscle it past the best that Nvidia and AMD could do through the independent foundries. Now, with the foundries getting to a "rough" parity with Intel's best process, they won't have that process advantage to lean on. They will have to compete on chip design as well as process tech.

With that being the case, if they just took what they have as an iGPU, re-purposed the general design into a dGPU with the required memory controller, etc, and clocked it to the moon with enough VRAM to be relevant, then I can see it being a solid mid-market competitor. Their iGPUs aren't horrible, just targeted to a slightly different market with the performance to match.

 

[JoeRambo]

While Intel GPU performance is not horrible, their drivers are. It will take Intel years to reach parity in driver performance and compatibility with "quirks" that are in the games ( both drivers working around game quirks, and games being designed with AMD/NV driver peculiarities in mind ). Sure they can target DX12/Vulcan and subset of latest DX11 games in 2020, but it is the legacy games that will have problems.

 

[SPBHM]

Saying it like that sounds terrible.

well, yes but, they haven't made a proper new IGP architecture for a few years, and they are always made for low power usage...

give it a high TDP, improved arch for high performance, dedicated memory, a big die... and maybe it will perform, even if not on par with Nvidia,

if they want to gain market share all they need is a well priced "gtx 1050-1060" style card, which is not easy to achieve sure, but it should be realistic, and their "marketing/discounts" machine can work its magic.

now if they are going for high end in 2020... I'm skeptic.

 

[maddie]

well, yes but, they haven't made a proper new IGP architecture for a few years, and they are always made for low power usage...

give it a high TDP, improved arch for high performance, dedicated memory, a big die... and maybe it will perform, even if not on par with Nvidia,

if they want to gain market share all they need is a well priced "gtx 1050-1060" style card, which is not easy to achieve sure, but it should be realistic, and their "marketing/discounts" machine can work its magic.

now if they are going for high end in 2020... I'm skeptic.

One thing we know is that Raja is in charge.

We have a lot if interviews with him laying out a vision for his graphics nirvana. Extremely high-res, high refresh rate, low power, HDR virtual reality indistinguishable from reality. He's still the same person and probably sees Intel as having the resources to achieve it vs AMD, who starved him.

A reasonable assumption is that he sold this same vision to Intel. In that case we will see them aiming high.

 

[Despoiler]

I'm still skeptical. My feeling is that Intel will be a third player, but far distant to AMD or Nvidia. I'm confident Raja can get Intel's software support where it needs to be. He took AMD from awful to an alternative in ~2.5 years. I don't know what Intel can do on the GPU front. Their current biggest GPU is ~12x slower than AMD and Nvidia. Their perf per watt when normalized is much worst than AMD's. Even if they solve those challenges, will Intel develop a competitive fab process for big GPUs? This is an uphill slog if I've ever seen one.

 

[tviceman]

I'm still skeptical. My feeling is that Intel will be a third player, but far distant to AMD or Nvidia. I'm confident Raja can get Intel's software support where it needs to be. He took AMD from awful to an alternative in ~2.5 years. I don't know what Intel can do on the GPU front. Their current biggest GPU is ~12x slower than AMD and Nvidia. Their perf per watt when normalized is much worst than AMD's. Even if they solve those challenges, will Intel develop a competitive fab process for big GPUs? This is an uphill slog if I've ever seen one.

I agree. TSMC is essentially caught up in node advancement, so Intel's fabled manufacturing advantage is out of the window. I think the only advantage Intel will have is a chip-cost advantage with not needing to outsource manufacturing. But even then, if you're looking at 20-25% less perf/w vs. AMD, which would be 75-100% less perf/w vs. Nvidia, it will take an awfully high total-package-power for a low end and even mid-range chip to compete in performance.

 

[HutchinsonJC]

so Intel's fabled manufacturing advantage is out of the window

Yeah, and if we consider that Intel is looking at 2020, I'm leaning toward a thought that they may even lose more ground by then.

if they want to gain market share all they need is a well priced "gtx 1050-1060" style card

In another thread about AMD's ryzen/epyc, and talking in terms of the big business world, where upper management kinda dictate to you, based on experience and reputations, the tried and true... you end up where a lot of IT folks have this "AMD is off limits" being pushed on them.

Reputation, future compatibility, etc being the big point there.

With that said, and with Intel's gpu coming in 2020, how many folks will jump off the Nvidia or AMD wagon to even try an Intel "1060" level card? How many people will try something new, unless it's priced very (emphasis) well? And we know how Intel can be with margins. Reviews need to be kinda glowing to get people to take interest, I think. Otherwise, I think there's a huge chunk of gpu buyers that will stick with what they know.

I'm kinda looking forward to the bits and pieces of "leaked" info or official info that will come out from now until then in regard to Intel's renewed push for graphics acceleration. Should be interesting, anyway.

 

[EXCellR8]

This is a ways off... and meanwhile AMD and NVIDIA will have better products released so who knows.

 

[Mopetar]

Hopefully this gets Nvidia off of resting on the laurels in the consumer segment with their tech lead over AMD.

That would imply Intel having better GPUs than AMD (or why else would NVidia change their behavior) and we know that comparing current Intel and AMD graphics technology that this is not the case. Intel is also using the person who was heading AMD's graphics efforts to this point, so I don't think there's a lot of good reason to believe that Raja is going to be able to accomplish with Intel what he couldn't with AMD. You could perhaps argue that AMD didn't have the money to fund GPU development to the necessary extent, but I don't think that the direction that Raja was trying to head towards worked out well in the grand scheme of things, so I'm not sure that throwing additional money at it makes it any more workable.

If anything, I think Intel will try to chase the compute side of the market. Even if they're not as good as AMD or NVidia, there's a lot of margin in those product segments that means even if Intel is the third-rate product, they can likely still make a decent amount of money.

 

[HutchinsonJC]

If anything, I think Intel will try to chase the compute side of the market.

I have no doubt that this is the case. Intel has been losing a lot of ground to Nvidia for various types of compute.

Edit: I do not, however, think they'll want to settle for 3rd or be happy with 3rd.

 

[Mopetar]

I do not, however, think they'll want to settle for 3rd or be happy with 3rd.

I'd have a hard time expecting other outcomes as being more likely. Intel might have more resources, but it's quite clear that they lack the same level of expertise in that area. Compare similarly with NVidia's efforts to develop their own CPU. How long was project Denver in the works for and what did NVidia ultimately have to show for it?

Intel needs to aim for a market where they can be profitable being in third unless they're willing to spend a lot of money subsidizing all of the time need to get to AMD/NVidia levels of ability while turning out inferior products that don't fund the future development effort.

 

[HutchinsonJC]

I have no contention with the likeliness that they will end up 3rd. I just don't think they'll be happy or settle with it.

there's a lot of margin in those product segments that means even if Intel is the third-rate product, they can likely still make a decent amount of money.

3rd rate isn't going to make the margins Intel is known for wanting.

 

[VirtualLarry]

While Intel GPU performance is not horrible, their drivers are. It will take Intel years to reach parity in driver performance and compatibility with "quirks" that are in the games ( both drivers working around game quirks, and games being designed with AMD/NV driver peculiarities in mind ). Sure they can target DX12/Vulcan and subset of latest DX11 games in 2020, but it is the legacy games that will have problems.

Given the various bugs and lack of certain features over the years, I feel confident in saying, that Intel just isn't as strong on the software front as AMD and NV. Even IF they catch up on the hardware front, and that's a really big IF, they still need to catch up on the software / game interfacing / game tweaking / driver front.

 

[VirtualLarry]

Two additional thoughts. What if Intel goes for lower-price / higher-volume? Imagine ebay five years from now, and all of those "Chinese faked NV GPUs"... are instead, Intel lower-end GPUs.

And what if Intel starts to bundle products? Like they're currently doing, for their i5+ and i7+ CPUs. (The "+" meaning that an Optane Memory Cache NVMe drive is INCLUDED in the package.)

I mean, AMD has been selling their "Game Crate" at Newegg, and they often get sold out. (Ryzen 5 1600 CPU, MSI RX 580 GPU, MSI AM4 mobo).

What if Intel starts selling i7 + dGPU combo kits, as a single higher-end "Gamer Package".

Remember, Combat was the highest-selling Atari 2600 game. Because it came with the Console.

Intel might just turn their iGPU strategy, of including one with every GPU, into their dGPU strategy, through forced bundling. (What if it wasn't even possible to buy an Intel CPU at retail WITHOUT one of their dGPUs in the retail package?)

 

[Mopetar]

3rd rate isn't going to make the margins Intel is known for wanting.

25% of a small pie is better than 0% of a big pie. No shareholder will complain about additional profit either. They only care about margins if your market isn't growing.