[News] Intel Hires Raja Koduri, to Develop Discrete GPUs, This Time for Real

[StyM]

https://www.techpowerup.com/238620/intel-hires-raja-koduri-to-develop-discrete-gpus-this-time-for-real


Intel hired Raja Koduri, who resigned as head of AMD's Radeon Technologies Group (RTG), earlier this week. Koduri has been made Senior Vice President and Chief Architect of Intel's future discrete GPUs. That's right, Intel has renewed its dreams to power high-end graphics cards that compete with AMD and NVIDIA. Intel's last attempt at a discrete GPU was "Larrabee," which evolved into a super-scalar multi-core processor for HPC applications under the Xeon Phi line.

This development heralds two major theories. One, that Intel's collaboration with AMD RTG on graphics IP could only go further from here, and what is a multi-chip module of Intel and AMD IP now, could in the future become a true heterogeneous die of Intel's and AMD's IP. Two, that the consolidation of AMD's graphics assets and IP into a monolithic entity as RTG, could make it easier to sell it lock, stock, and barrel, possibly to Intel.

:eek:

[iddqd]

he has literally never been good at any job that he's held.

[Metroid]

It is going to be fun if the next intel discrete igpu is a cooker like vega hehe

[drmrlordx]

Wait.

Discrete iGPU?

[Particle]

he has literally never been good at any job that he's held.
What makes you say that?


It is going to be fun if the next intel discrete igpu is a cooker like vega hehe
He'll have better process technology to play with as well as a different supporting design team, so he might do even better than he did at AMD. Intel has real money to spend on research and development.

[iddqd]

What makes you say that?

his decade-long portfolio of failed projects

[kromosto]

This means intel will announce a card that can compete with 1080ti release it 2 years late with a performance between 2080ti and 3070 which will make Koduri remember his family and resign from intel to be with them...

[Particle]

his decade-long portfolio of failed projects

Nothing interesting comes to mind? I don't know his work history, so I was hoping for something substantive to look at.

[Buckeye]

I would think a company like Intel will not be making a gaming GPU, but rather a Compute card, perhaps ASCI cards or a mix of both. Besides the Xeon Phi cards which are mostly outdated now and have a very specific use Intel has nothing to throw in the Compute area and it needs to compete with Nvidia and provide its own solution.

[Metroid]

And Vega was a good example on compute area although it failed in the gaming area, I def would not want a cooker inside my desktop.

[iddqd]

Vega is good at nothing. It's not competitive at either compute or graphics.


Nothing interesting comes to mind?

The last 'successful' card he worked on at (still) ATi was the R700 (Radeon HD 4XXX series). But back then his involvement was much smaller - he wasn't in charge of the whole design, only the software part of it. Note that this is the time when ATi's graphics drivers were particularly garbage. Then, he left to become chief graphics architect at Apple, and worked on the GPUs in iPhones 1, 3G, and 4. These weren't built from scratch, but rather, modified PowerVR designs. Note that at the time, iPhones were notorious for terrible, underperforming GPUs that were a royal pain in the ass to program for. He's also partially responsible (through his connections at Intel) for Apple's PowerPC -> x86 switch, although you can't really attribute the success to him; Intel did all the hard engineering work.

Anyway, he left before the iPhone 5 came out, with a much-improved GPU. Since his absence, iPhone GPUs became better and better (and are actually somewhat painless to program for now), and actually outperform Qualcomm's offerings. Of course, he went right back to AMD, where he (unsuccessfully) tried to spin RTG off, threatening to quit and go to Intel. He kept throwing tantrums until Lisa Su gave up and just gave him free reign to do whatever he wanted at RTG. He immediately put his newfound freedom to work by producing several hopelessly uncompetitive designs: starting with Hawaii (r9 29X), then following it up with the pointless Tahiti & Tonga (r9 280, 285X), and then the even less competitive Fiji (Fury). The 290 was actually better positioned in the market than Fury. This trend continues with his latest failure, Vega, which is in an even more awkward position than the Fury was. The card literally has no reason to exist.

[drmrlordx]

Vega is actually a pretty good compute card. It's quite good for mining Monero.

[iddqd]

If it were anywhere close to competitive with what nvidia offers (the Tesla cards), they would be charging $6000 for it. It's not though, so they can't.

[vario]

If it were anywhere close to competitive with what nvidia offers (the Tesla cards), they would be charging $6000 for it. It's not though, so they can't.

Well i hope you are right and intel bought themselves a problem and not a solution. Because if intel would do a good gpu, amd would be in a lot of trouble and we need the competition.As for vegas, im not so sure its as bad as you portrait it, i think that most of its problems come from glofo, not some inherent design flaw. Although we should have gotten Big polaris chip instead of waiting for vega, and vega maybe next year at 12nm with a decent supply of hbm2 at good clocks.
And i have one, bought at launch price, and its been pretty good, well, some quirks with the drivers for it (and i pretty much never had any driver problems with ati!) but thats it.Im pretty sure it was better buy longterm than 1070, recentlu it became quite obvious that last gen cards from both nvidia and ati dont get shader optimizations for newest games...
And well, in some compute cases it must be better than what nv is offering, its the fastest at ethereum and monero for sure.

[iddqd]

Fat Polaris would have clowned all over Vega for sure, giving it even less of a reason to exist.

[Metroid]

Regarding Vega being a gaming card, the idea about HBM was to cut power consumption by half and AMD/Raja with Vega did the opposite but it was not just half, it increased 100%, from around 200w to 400w, that was unbelievable. If they can't see where was the mistake then they better look for a different type of job.