Slashdot Mirror
Intel Abandons Discrete Graphics
Stoobalou writes with this excerpt from Thinq: "Paul Otellini may think there's still life in Intel's Larrabee discrete graphics project, but the other guys at Intel don't appear to share his optimism. Intel's director of product and technology media relations, Bill Kircos, has just written a blog about Intel's graphics strategy, revealing that any plans for a discrete graphics card have been shelved for at least the foreseeable future. 'We will not bring a discrete graphics product to market,' stated Kircos, 'at least in the short-term.' He added that Intel had 'missed some key product milestones' in the development of the discrete Larrabee product, and said that the company's graphics division is now 'focused on processor graphics.'"
They've never been able to bring the most innovative designs to market.. they bring 'good enough' wrapped in the x86 instruction set.
If x86 was available to all I think we'd see Intel regress to a foundry business model.
This is bad news for one reason. Competition. There are only 2 major players in discreet graphics right now and that is horrible for the consumer. Now the good. Intel SUCKS at making gpus. I mean seriously. So either way Intel has no hope of making a 120 core GPU based off of x86 being cheap or fast enough to compete. Go big or stay at home. Intel stay at home.
I kind of think Larrabee was a hedge.
If you think about it, around the time it was announced (very early on in development, which is not normal), you had a bunch of potentially scary things going on in the market.
Cell came out with a potentially disruptive design, Nvidia was gaining ground in the HPC market, OpenCL was being brought forth by Apple to request a standard in hybrid computing.
All of sudden it looked like maybe Intel was a little too far behind.
Solution: Announce a new design of their own to crush the competition! In Intel-land, sometimes the announcement is as big as the GA. Heck, the announcement of Itanium was enough to kill off a few architectures. They would announce Larrabee as a discrete graphics chip to get gamers to subsidize development and....profit!
Lucky for them, Cell never found a big enough market and Nvidia had a few missteps of their own. Also, Nehalem turned out to be successful. Add all that up, and it becomes kind of clear that Larrebee was no longer needed, negating the fact that it was a huge failure, performance-wise.
Intel is the only company that can afford such huge hedge bets. Looks like maybe another one is coming to attack the ARM threat. We'll see.
FUNK!
The Larrabee chips actually looked pretty good. There was a lot of hype, especially from Intel. They demoed things like Quake Wars running a custom real-time ray-tracing renderer at a pretty decent resolution. Being able to use even a partial x86 ISA for shaders would have been a massive improvement as well, both in capabilities and performance.
From what I've been able to piece together, the problem wasn't even the hardware, it was the drivers. Apparently, writing what amounts to a software renderer for OpenGL/DirectX that got good performance was beyond them.
Another part was an odd insistence on doing all the rendering in software, even stuff like texel lookup and blitting, but that's another story.
There was a company called Rapid Mind, which built library & tools for writing code to target various GPUs, multi-core CPUs, etc. Something similar OpenCL, I suppose, but easier to program (theoretically -- I never actually tried it). Intel bought it and killed it.
Another company, Havok, developed a successful physics & AI library. They were going to port it to to GPUs. Then Intel bought it and canceled the GPU port.
___
If you think big enough, you'll never have to do it.