Well, this is the kind of news you don’t hear every day: Intel and AMD are teaming up to develop a processor that combines an Intel CPU with an AMD GPU. From Intel’s press release:
The new product, which will be part of our 8th Gen Intel Core family, brings together our high-performing Intel Core H-series processor, second generation High Bandwidth Memory (HBM2) and a custom-to-Intel third-party discrete graphics chip from AMD’s Radeon Technologies Group* – all in a single processor package.
It’s a prime example of hardware and software innovations intersecting to create something amazing that fills a unique market gap. Helping to deliver on our vision for this new class of product, we worked with the team at AMD’s Radeon Technologies Group. In close collaboration, we designed a new semi-custom graphics chip, which means this is also a great example of how we can compete and work together, ultimately delivering innovation that is good for consumers.
This is the first partnership between these two sworn rivals in several decades, and that alone makes it quite notable. I didn’t really know whether to put this in the Intel or AMD category, but I chose Intel because it appears above AMD in our list (which isn’t alphabetical because reasons).
Will Intel choose proprietary AMD drivers or FLOSS drivers?
(Talking about Linux here)
FLOSS drivers general have better performance for gaming, Vulkan, and video.
If that’s what Intel will choose, some additional development resources will be dedicated for improving AMD drivers?
Intel drivers cover a bit more OpenGL (though both are 4.5 compliant), a bit less Vulkan, way more OpenCL.
It will be interesting to see how would Intel deal with those differences inside single generation.
For Linux, FOSS drivers ARE the official drivers for the kernel (amdgpu.ko, thus also the Wayland DRM driver) and X11 (xf86-video-amdgpu). It’s the OpenGL, OpenCL, and Vulkan libraries that are separate at the moment.
Yes/No
AMD have opensource Linux kernel driver and both proprietary and floss userspace graphic drivers.
However proprietary driver currently works with downstream version of kernel driver, while floss one can happily work with vanilia kernel driver and general is developed in tandem with upstream kernel efforts.
So Intel could possibly choose AMD’s pro driver, and would work on downstream kernel driver, and that would not benefit floss driver in short term.
I thought Intel had decent integrated graphics nowadays. Have they not improved the Iris Pros at all since their launch?
They are still lacking and uses a lot of power. And the Iris models and mostly fast due to edram which they could give an AMD iGPU too.
These chips are going to have HBM on the package, so same setup as the Iris models with much more bandwidth.
I don’t understand the point of this, especially from Intel’s position. It basically acknowledges a weakness in their own line of graphics chips, which is not normal corporate behavior, especially for a would-be monopolist. (It’s not as surprising from AMD’s perspective, but it still seems odd.)
OTOH, for the customer, this may end up being a good thing, assuming driver support is good (present and future).
I was thinking exactly this, but from AMD’s side. They (AMD) finally have a decent CPU, and with their GPU paired with it, can be a compelling alternative to Intel. Why enter into a deal like this with a competitor when it so clearly undercuts their confidence in their own CPU, and give a clear strategic advantage away?
From Intel’s side, no doubt they’d prefer to use something else like nVidia – but they don’t play nicely with this kind of IP licensing.
Nvidia has pissed off Intel, and Intel has a pretty good relationship with AMD. They’re competitors, but they’re pretty much joined at the hip because of x86 cross-licensing agreements.
So basically, they are cross licensing their GPU in the same way they cross license x64 – I suppose that’s reasonable. I’m not sure it demonstrates strong strategic thinking though. It’s very tactical.
They’re not croslicensing, AMD still makes the GPU. They’re tecnically integrating.
You’re not wrong; this is integration.
However, I don’t think this chip exists without the graphics licensing agreement AMD and Intel signed earlier this year. While it’s mainly a shield against patent infringement, it could be more. We’ll have to wait and see.
https://wccftech.com/intel-strikes-deal-amd-graphics-license/
How did Nvidia piss off Intel? I’d imagine Nv being pissed at Intel for not letting them make x86 chips… (how did we have so many of x86 manufacturers in the 90s, anyway / why it wasn’t a problem for them to license x86 / how did we end up with a duopoly?)
Intel and NVidia have a weird relationship. I’d figure NV would be more pissed when intel basically turned the socket for their processors into a monopoly and stopped giving licenses for 3rd party chipset when the i-series was introduced. That almost sent nvidia under.
Also remember a lot of x86 vendors back in the 80s/90s were 2nd sources. I.e. they were simply fabbing intel designs, and paying them a royalty, since some of the early contracts with large PC vendors, IBM and Compaq (and a few others) kind of specified intel had to allow 2nd sourcing, since they lack the capacity to meet the possible demand at that time.
I think only a few 3rd parties have been able to license the x86 and make their own designs. AMD, obviously, and Via. I think IBM had a temporary license to do their own 386/486 ISA implementations.
I wish the industry would have settled around an open ISA instead. Copyrighted instruction sets suck.
x86 instructions set sucks, period.
Honestly, that’s such a tired trope ;-). Since the 386, i.e. the mid 80s, on the x86 has been a fairly standard ISA.
Standard ISA doesn’t mean a “clean and orthogonal ISA”, in which the x86 specifically sucks, period. Look at 68k, mips, sh, to get an idea of what a better ISA looks like.
Of course x86 improved, especially the x96-64 gap introduced by AMD, but I digress, the legacy compatibility still spoils the fun.
Yawn. This is even more tired tripe. This keeps being repeated, usually someone read it on usenet and regurgigated it, rinse and repeat.
E.g. M68K has its own sets of inelegancies as an ISA. MIPS is even more ironic; like most RISC designs it was not meant to be exposed to the programmer that much, etc, etc.
As I said, past 386… x86 is not really that flawed/warty of an ISA. And with x64, it’s just an ISA that support lots of programming models, so it ended being pretty flexible after all.
My main issue with x86 is not the subjective qualitative/stylistic masturbatory exercises some people engage, but rather that I’d prefer if the industry had standarized around an open ISA. So the barrier of entry would have been much lower to outfits trying to produce their own processors/toolchains/etc.
Hm, I don’t remember any notable (or any at all?) Nv chipsets for Intel, apart from the chipset of Xbox1 …which Nvidia turned into a succesfull line of AMD chipsets (so I guess they would be more pissed at AMD when the latter bought ATI). IIRC Via had more presence on Intel platform & mostly they were cut off by Intel…
Yes, a lot of x86 vendors in the 80s were 2nd sources, but not so much in the 90s, I think – they had a few independent designs: I remember NexGen (which was bought by AMD and formed the basis for K6), Cyrix (bought by Via), IDT Winchip …somehow, it seems as if it was easier to license x86 from Intel back then, and the license wasn’t lost in the case of buyout of the company (as it would happen now if somebody bought AMD)
nvidia had a nice business going with the nforce chipset for intel during the Core days. They were basically the best chipsets for SLI (obviously), although they had some reliability issues.
Yeah, I think Intel got very protectionistic of X86 by the middle of the 90s. I was a kid back then, but I remember how they basically introduced the name pentium, so they could copyright the product line after the 486. And at some point they also ended up kicking out any 3rd parties from their sockets.
I don’t know if ISAs were commonly copyrighted back in the 70s/80s. So probably that was the reason for the number of x86 cloners back then?
Hm, for some reason I forgot about Nforce for Core / SLI. Perhaps because ultimatelly it was a fairly niche product (so it probably didn’t hurt Nvidia that much when Intel blocked it) – essentially, a halo product.
And I thought we already established that there were plenty of x86 cloners also, or perhaps especially (with indpendent designs; and even more advanced than Intel, for example IIRC Nexgen was the first that translated x86 instructions into internal RISC-like), in the 90s…
Actually losing the chipset business hurt nvidia plenty. Part of the settlement that nvidia got from Intel were used to cover the damages.
The microinstructions is really more of an out-of-order thing. Intel also did it starting with the Pentium pro in the mid 90s as well.
I see. Hm, perhaps I forgot those Nv SLI Intel chipsets because they weren’t that popular here, being a bit premium after all (as is SLI), while my place is more about value solutions… Which Nvidia also had – the Geforce 6100 chipsets with integrated GPU for Athlon 64 were phenomenally good value, I built a computer with it for my buddy for ~200^a'not, and it performed quite good in contemporary games.
That was probably the first shot fired. Nvidia retaliated and has held a grudge with both sides going back and forth.
Nvidia also doesn’t play well with others. They want to do things their way.
This doesn’t undercut their own CPU products in any way. Ryzen doesn’t generally ship with integrated graphics, relying on a dedicated AMD or nVidia GPU for graphics processing. Sure, some of the mobile chips will start shipping with integrated Radeon chips at some point soon.
No, if anything, this increases AMD’s market penetration. OEM’s have always mainly shipped Intel processors, and in laptops (the most sold PC form factor today) these machines tend to rely on Intel’s rather lackluster integrated graphics. By shipping AMD GPU’s as the integrated graphics, AMD has a whole new market open to them. Consumers get fast and powerful integrated graphics, OEM’s get to ship “reliable” and “proven” Intel chips. Everybody wins.
Ryzen doesn’t, but APU’s do, and AMD is releasing Zen based APU’s, not just for mobile, but desktop ones too. Even aside from that though, they still sell lots of one or two generation old APU’s (in fact, those are probably some of the best selling x86 CPU’s right now, you can play GTA 5 on a one generation old 80 USD APU and get triple digit frame rates from the integrated GPU).
In my opinion, a better assessment would be that it’s not going to undercut their own CPU products simply because they’re half the cost of equivalent Intel offerings. AMD covers the low cost end-user market far better than Intel, and Intel covers the sickeningly overpriced corporate market much better than AMD.
Intel’s market position is stagnating. They haven’t had any substantial performance updates to their flagship chips in years, and the PC market is flat. They tried getting in on tablet/cellphone market, but that fell flat. (Yeah, they were worse then the notoriously awful ARM manufacturers.)
They need to do something to keep advancing, and they need to keep their customers. In this case, they probably needed to appease Apple, and Apple’s lust for high DPI displays. This wouldn’t be the first time Intel spun some custom silicon for them and offered it to the general public.
you do realize Intel makes a hell of a lot more than desktop processors, right?
You do realise Microsoft makes a hell of a lot more than Office and Windows, right?
You do realize Ikea makes a hell of a lot more than just chairs and tables right?
Edited 2017-11-08 04:04 UTC
Yes, but that is the straw that stirs the drink.
I would be interested to know how it’ll be marketed.. will the Intel or AMD brand be prominent?
I assume Intel…
It could be argued that some external force could be at play here.
Like Apple threatening to quit Intel CPUs unless reasonably powerful iGPU is bundled with them
Who are they going to get processors from in the future then? AMD?
I’d figure some of the new ryzen APUs should be pretty attractive to apple.
AMD, and or internal.
Even threat of introducing their own chipsets in 5y future would be enough for Intel to think seriously about what Apple wants, now that Apple do have their own mobile offering.
They’ve been doing this for years in mobile under the fashion of Intel making the mobile CPU and AMD providing the GPU.
That went quite well until AMD started selling their own mobile CPU / APU and all of the sudden no longer offered support for any platform that had this combination available.
So many problems came from this “small” thing for people upgrading from Windows 7 to 8 / 8.1 until AMD made it clear it will support at least in legacy fashion this configuration.
I’m not sure if this new direction is targeting the “PC” Market. It’s more like they keep trying to make Apple linked to their offerings for their mac line that they needed to do this one way or the other..
Presuming this venture bears fruit then Apple should be able to design a mac mini that has good CPU/GPU. A modular hardware setup would be good; at least various CPU/GPU combinations.
A pity they have avoided first class support for native OpenGL/Vulkan (instead only embracing “Metal 2”) since this possible future mac-mini would have been a nice cross-platform graphics-app development box.
Would make much sense to have such a chip for potential new Mac minis / iMacs and (if mobile version will happen) MacBook Pro.