News Intel GPUs - waiting for B770

[LightningZ71]

Eh? During the long period of core amount stagnation the iGPU size was actually what Intel "improved" between generations. Which of course on the desktop was completely superficial considering essentially everybody gaming used a dGPU anyway. The fact that Intel has been doing iGPUs for so long already doesn't seem to help it pushing out dGPUs.

"Improved" is a very generous assessment of their iGPU activities between Broadwell and Comet Lake. There are minor additions to the media engine and modest feature adds to the EUs, but, aside from a few "Iris" products that almost exclusively went to MACs, and an abortive MCM with AMD in Kady Lake G, it's barely a blip with respect to performance... That level of effort certainly wasn't taking anyone away from even the worst of dGPUs, with the 1030 roundly beating every one of those standard iGPU solutions by a country mile.

 

[LightningZ71]

Their chips already consume too much power and run too hot. Pushing that up even higher isn't really an option.

In the mobile space, that can be customized at the firmware and software level. For that additional performance and power draw, you get to eliminate from many designs low end dGPU provisions, allowing for the added cooling to make the higher total thermal dissipation more manageable. The power being concentrated in one chip, usually on a more refined process, would only serve to increase power efficiency of the whole system at similar performance levels.

 

[DrMrLordX]

The power being concentrated in one chip, usually on a more refined process, would only serve to increase power efficiency of the whole system at similar performance levels.

That also makes cooling potentially more expensive. OEMs are cheapskates.

 

[LightningZ71]

Oh, OEMs will certainly find a way to mess it up. However, it also could make it a lot more efficient with materials usage.

 

[igor_kavinski]

That also makes cooling potentially more expensive. OEMs are cheapskates.

They could create a keyboard with metallic keys with heatpipes underneath. That way, they would get a really large surface area to dissipate the heat. But it would give new meaning to the term "hotkey".

 

[Panino Manino]

I should had think harder to express me batter and write something like "a moment f weakness", but SHAME ON ALL OF YOU! For even joking that I'm taking Intel's side.

Back to topic, even with these GPUs aren't that great, even so don't Intel have an opportunity? If AMD really step up their APU game and focus on the high end, it seems that Intel will have an open niche to fill up. They don't need more than that, just some price segment where they can start their long run.
Still believe Intel have an important part to play fighting Nvidia along with AMD.

 

[igor_kavinski]

If AMD really step up their APU game and focus on the high end, it seems that Intel will have an open niche to fill up. They don't need more than that, just some price segment where they can start their long run.

Yeah if AMD's bright idea of future low end GPU's is similar to 6400/6500 XT, then sure. Intel gets to carve their own slightly better low end niche.

 

[moinmoin]

don't Intel have an opportunity?

Intel had and still has plenty opportunities, but still hasn't made use of a single one yet. Which is why this thread turned out the way it did. Not because Intel's dGPUs are bad (or not, we still don't know) but because the only thing Intel's dGPUs managed to do so far is missing great opportunities.

Note that this thread started in 2017 and is huge at 87 pages so far, and we are still waiting for the first real dGPU resulting from this whole effort.

 

[StinkyPinky]

When will the first Intel video cards be coming out???

A day late and a buck short.

 

[moonbogg]

A day late and a buck short.

A year late and a Bitcoin behind.

 

[NTMBK]

I should had think harder to express me batter and write something like "a moment f weakness", but SHAME ON ALL OF YOU! For even joking that I'm taking Intel's side.

Back to topic, even with these GPUs aren't that great, even so don't Intel have an opportunity? If AMD really step up their APU game and focus on the high end, it seems that Intel will have an open niche to fill up. They don't need more than that, just some price segment where they can start their long run.
Still believe Intel have an important part to play fighting Nvidia along with AMD.

In all seriousness, yes, Intel has a shot. It will require long term commitment and a willingness to eat big losses in the medium term, but hopefully their new leadership is willing to do that.

In my view, this generation was always meant more to lay the groundwork, and start building up the ecosystem, the driver teams, the distribution channels, etc. It's built on TSMC! Intel really wants to be building GPUs in their own fabs, getting the benefits of vertical integration, and putting more money into their own process R&D instead of funding their competitor. They need a wider product range in order to sustain the R&D costs of fabs in the Angstrom era. They want GPU market share for the same reason they want a foundry business.

It's in their long term strategic interests to be building huge numbers of Intel GPUs in Intel fabs. I think they are willing to ride out a bad generation, to play the pricing games they need to be competitive, in order to achieve that long term.

 

[moonbogg]

In all seriousness, yes, Intel has a shot. It will require long term commitment and a willingness to eat big losses in the medium term, but hopefully their new leadership is willing to do that.

In my view, this generation was always meant more to lay the groundwork, and start building up the ecosystem, the driver teams, the distribution channels, etc. It's built on TSMC! Intel really wants to be building GPUs in their own fabs, getting the benefits of vertical integration, and putting more money into their own process R&D instead of funding their competitor. They need a wider product range in order to sustain the R&D costs of fabs in the Angstrom era. They want GPU market share for the same reason they want a foundry business.

It's in their long term strategic interests to be building huge numbers of Intel GPUs in Intel fabs. I think they are willing to ride out a bad generation, to play the pricing games they need to be competitive, in order to achieve that long term.

Or they'll just fold like a cheap card table and throw the whole thing out.

 

[Stuka87]

In the mobile space, that can be customized at the firmware and software level. For that additional performance and power draw, you get to eliminate from many designs low end dGPU provisions, allowing for the added cooling to make the higher total thermal dissipation more manageable. The power being concentrated in one chip, usually on a more refined process, would only serve to increase power efficiency of the whole system at similar performance levels.

Not always. In a system with a discrete GPU, the discrete GPU is often powered down. And the much lower power integrated GPU is doing basic screen drawing and video rendering. If the CPU has a big GPU on it, it would require Intel to be able to really ramp the power down on the GPU when its doing basic work.

Also, its typically much easier to cool two chips putting out 100W each than a single chip putting out 200W (random numbers, not actual numbers). You need a lot more heatsink material to be able to draw away 200W from a single point. And laptops being height limited puts a limit on this.

Now, if Intel can find a way to make their chips consume way less power, then yes, having a big integrated GPU would work fine. This is how AMD is doing it. They have a pretty decent integrated GPU, but the CPU side is lower power.

 

[maddie]

Not always. In a system with a discrete GPU, the discrete GPU is often powered down. And the much lower power integrated GPU is doing basic screen drawing and video rendering. If the CPU has a big GPU on it, it would require Intel to be able to really ramp the power down on the GPU when its doing basic work.

Also, its typically much easier to cool two chips putting out 100W each than a single chip putting out 200W (random numbers, not actual numbers). You need a lot more heatsink material to be able to draw away 200W from a single point. And laptops being height limited puts a limit on this.

Now, if Intel can find a way to make their chips consume way less power, then yes, having a big integrated GPU would work fine. This is how AMD is doing it. They have a pretty decent integrated GPU, but the CPU side is lower power.

You're forgetting selective shutting down of unneeded logic. Everyone is going to very fine grained power gating. I see this extended to individual shader units and other component sub-blocks.

 

[Aapje]

In my view, this generation was always meant more to lay the groundwork, and start building up the ecosystem, the driver teams, the distribution channels, etc. It's built on TSMC! Intel really wants to be building GPUs in their own fabs, getting the benefits of vertical integration, and putting more money into their own process R&D instead of funding their competitor. They need a wider product range in order to sustain the R&D costs of fabs in the Angstrom era. They want GPU market share for the same reason they want a foundry business.

My prediction is that we are heading into a golden era for gaming in 2025+ when all three competitors are firing on all cilinders, boosted by huge war chests. Intel coasted for a decade on the CPU side and they will have to really compete now, while Nvidia also really needs to step up it's game. Intel's new strategies of entering the dGPU and foundry markets will shake the dust out of Intel, as they don't have past successes to coast on in those markets. And AMD finally seems to have figured out how to compete sustainably (at least as long as they have the console income).

DDR5 will most likely go to the moon in the next few years (while the price will rapidly drop once it is no longer optional when using newer CPUs). Add in the breakthroughs we've had and will have for chiplets. Exciting times ahead. We'll just have to get used to undervolting and underclocking...

We also need Intel and Samsung fabs to improve. Can't just let TMSC make all the top tier stuff.

 

[LightningZ71]

Intel can easily find footing in the market by using the vast regions of market pricing now being created by nvidia and AMD by their continuing push of MSRPs into the stratosphere with each generation. This has opened up many spaces for Intel to offer better performance per dollar while still being at worst revenue neutral until they have established market momentum. They just need something that they have routinely failed to provide to the market, stable, high performance drivers for 3d gaming. Of, they've made "playable" drivers for many games over the years, but, they did poorly with Xe. Their last performance "i"gpu product, KadyLake G, they literally walked away from supporting while products were still in warehouses to be sold, then pointed their fingers at the supplier that they refused to pay for it.

Prove it Intel.

 

[eek2121]

Or they'll just fold like a cheap card table and throw the whole thing out.

That will never happen. Intel needs the GPU segment to compete with NVIDIA and AMD in enterprise/datacenter/cloud markets.

People keep saying Intel is delayed, but I'm not seeing a delay. The roadmap I saw showed laptops getting a release, followed by desktops. I believe the official first desktop cards were due for release at the end of May, however I may be mistaken. I saw this roadmap in January or February. If I find it I post a link.

 

[moonbogg]

That will never happen. Intel needs the GPU segment to compete with NVIDIA and AMD in enterprise/datacenter/cloud markets.

People keep saying Intel is delayed, but I'm not seeing a delay. The roadmap I saw showed laptops getting a release, followed by desktops. I believe the official first desktop cards were due for release at the end of May, however I may be mistaken. I saw this roadmap in January or February. If I find it I post a link.

I can't imagine Intel ever taking the big performance jump needed to be relevant in the high-end or even mid-range gaming market. I think they will always be way too conservative in terms of raw performance and they will always come too late. I think their heads are stuck in mobile garbage GPU mode and they can't get unstuck. They will make some low-end garbage and pop champagne bottles and say "that'll do" then stick them in $800 garbage OEM builds sold at Walmart. They will champion their all Intel sticker PCs as being capable of full HD gaming with high-details and 60+ fps. These GPUs will be invisible to the enthusiast and high-end gaming crowd probably forever, and that's if they even exist.

 

[igor_kavinski]

Intel Acquires Siru Innovations to Bolster Push Into Graphics

The team at the Finland-based Siru Innovations formerly worked at AMD, Qualcomm, and BitBoys, a developer of desktop and mobile graphics.

www.pcmag.com

Intel got themselves the BitBoys, the infamous vaporware experts.

Weirdly, other sites are reporting a "partnership" while PCMag reports it as an acquisition.

 

[Aapje]

People keep saying Intel is delayed, but I'm not seeing a delay. The roadmap I saw showed laptops getting a release, followed by desktops. I believe the official first desktop cards were due for release at the end of May, however I may be mistaken. I saw this roadmap in January or February. If I find it I post a link.

They put Q1 in their slides and they clearly missed that. They then claimed to have released the first products in Q2, but only a handful of laptops of a single make have been released in Korea. I don't see how you call this anything other than a paper launch. No laptops were provided to any reviewers either, which is unlike a normal release and also highly indicative of this being a launch to only technically meet their promises, while not doing so in a way that most of us would see as a launch (a non-negligible quantity being available in all regions).

 

[igor_kavinski]

I don't see how you call this anything other than a paper launch.

Seems investors can't tell the difference and Intel falls back to this mockery of their intelligence as a time honored tradition.

 

[moinmoin]

Weirdly, other sites are reporting a "partnership" while PCMag reports it as an acquisition.

It's an acquisition.

I’m excited to share that Intel has acquired Siru Innovations, a Finnish… | Balaji Kanigicherla 19 | Kommentit

I’m excited to share that Intel has acquired Siru Innovations, a Finnish technology company specializing in the development of graphics IP and software services. We are thrilled to welcome CEO Mikko Alho and the entre Siru Innovations team to Intel and the Accelerated Computing Systems and...

fi.linkedin.com

 

[eek2121]

I can't imagine Intel ever taking the big performance jump needed to be relevant in the high-end or even mid-range gaming market. I think they will always be way too conservative in terms of raw performance and they will always come too late. I think their heads are stuck in mobile garbage GPU mode and they can't get unstuck. They will make some low-end garbage and pop champagne bottles and say "that'll do" then stick them in $800 garbage OEM builds sold at Walmart. They will champion their all Intel sticker PCs as being capable of full HD gaming with high-details and 60+ fps. These GPUs will be invisible to the enthusiast and high-end gaming crowd probably forever, and that's if they even exist.

It's not like these cards are weak. It will come down to price and driver quality. Intel technically does not NEED a high-end card. Far more 3060s + 3070s have sold than 3080s + 3090s. Even if next gen from AMD/NVIDIA is significantly faster in the midrange, Intel need only tweak the price a bit. Since these things are on N6, they are going to be cheap to make so although will have to sacrifice margins, they could potentially price the cards competitively with next gen.

Also, next gen cards from NVIDIA and AMD are almost certainly Q4. Intel will likely have launched desktop and mobile parts by then. Note that OEMs will 'prefer' Intel for Intel laptops and desktops since they'll likely get huge discounts if going all Intel. that means you are likely to see a LOT of laptops that use Intel + NVIDIA switch to Arc.

 

[moonbogg]

Laptops are mobile garbage and not real computers.

 

[Aapje]

It's not like these cards are weak. It will come down to price and driver quality. Intel technically does not NEED a high-end card. Far more 3060s + 3070s have sold than 3080s + 3090s. Even if next gen from AMD/NVIDIA is significantly faster in the midrange, Intel need only tweak the price a bit.

The driver quality seems to suck real hard though. I'd rather pay more for a card that does actually work than for one that theoretically has better performance for the price, but is horrible to use in practice due to crashes, weird bugs, uneven performance, etc.

Intel has a lot of fanboys, but selling them crappy cards seems like a good way to cure those fanboys.