Raja Koduri, Executive Vice President & Chief Architect, Leaves Intel

[kondamin]

The article and this whole thread is calling into question whether or not ‘Intel is working on it.’ It is unclear to many given Intel recent moves such as planned layoffs, product cancellations and refreshes.

As I posted earlier, I believe Arc will be cancelled after the Alchemist refresh.

They have scrapped more important technology, like xpoint.

 

[Jermelescu]

Actually yes. His record is quite good for bringing up new architectures that make deep impacts. Everyone needs to step back and remember that all of HIS work ended the moment the card ships. Just like Jim Keller left AMD over a year before Zen 1 launched, yet his direct input extended all the way to laying foundations on Zen 3. Once the stamp of approval hits the design and it begins packaging it's time to jump to the next project.

Similarly this could mean that Arc is on track and Raja's involvement is not pivotal to the existing road map, so he's leaving before the next major architecture begins to take form. He's likely had direct input on what we will only know about in 4 years time.

Mr Keller designs stuff that changes an industry. Raja doesn't hold a candle to that. Polaris/RDNA and Arc are seriously underwhelming, whereas Apple's A chips ans Zen pretty much redefined efficiency. In two different market segments, no less.

Unlike most of you know, Jim Keller did just a bit of design on Zen, as his primary job was to manage and lead the teams that worked on it.

 

[iO]

And in 4-5 months or so Gelsinger will kill off ARC officially.

 

[ThrashZone]

I mean that's not at all what anyone claimed, but yeah.

Hi,
Well that is what he said last time he left/ fired from asus I believe was the company and then he went to intel shortly after so yeah guess he tiered fast of family
Not sure it passes the small test this time.

 

[bug]

Unlike most of you know, Jim Keller did just a bit of design on Zen, as his primary job was to manage and lead the teams that worked on it.

It matters little what their day job is made of. I was just talking about results. Mr Keller puts you on top, Mr Koduri gets you second place in a two runners race.

 

[R-T-B]

Hi,
Well that is what he said last time he left/ fired from asus I believe was the company and then he went to intel shortly after so yeah guess he tiered fast of family
Not sure it passes the small test this time.

I mean family can be tiring. That's the one part I'll grant him, lol.

 

[Haile Selassie]

They have scrapped more important technology, like xpoint.

3DXpoint was never profitable and they did not secure enough design awards.
GPU is a different story. Having GPU of this calibre puts them into contention for next gen console SoC.

 

[L'Eliminateur]

I've said it before(you can check my posts!) and i'll say it again as this pretty much confirms it: Raja is a smoke and snake-oil salesman, he ALWAYS overhype, overpromises and underdelivers HARD.
He hypes and hypes and lies through his teeth then fibs and vanishes from [whatever company he's in]

Most if not ALL the products he oversought where late underwhelming failures.

Look at the track record as other have said, when Raja leaves, companies start delivering good product, and spare me the "oh but development is long those are probably made by him" bullshit because they're not.

I've always been surprised how he keeps getting hired..... (he kind of reminds me of r night shamalamadingdong that was given carte blanche on whatever film project he wanted until the industry learned the hard way he made only one good movie -his first- then all trash and flops)

And it comes to no surprise that he'll apparently will go to an AI startup: the PERFECT place to sell snake oil nowadays(like "the cloud" before) "if->then"(aka: AI) with his usual bullshit.

 

[ThrashZone]

Hi,
Well if anything the release showed he never used the item they produced and he can't write or manage vbios or drivers production either

 

[TheinsanegamerN]

Everyone associates Raja with Polaris, when he actually set the foundation for Navi while at AMD. Nonetheless, even Polaris was amazing value at the time (I miss those days compared to the shit show we have in the gpu market today). Absolutely nothing viable at a decent price, and all our wishes for AMD to become competitive so we can see better pricing did not come to fruition as you can see today (the fact Intel is our last hope is a very sad reality indeed).
With ARC, I dismissed it at first until I recently started paying close attention to it in reviews, and its actually quite impressive for what it is. The ray tracing implementation is solid. It beats out a 6700xt for much less. This is only a solid foundation for Intel to continue. If they stop now it would be very shortsighted of them. Anyways, thats what Raja (and other other architects) do in general. They join, lead a project or program up to a run and maintain/growth phase, and then move on to kick start something new. Jim Keller did that with Zen, and Raja did the same with Navi at AMD.

View attachment 288729

That is a very......selective history.

Raja did some amazing work at ATi back in the day, serving as CTO from 2001 to 2009. That era gave us some great chips, the evergreen design, ece. It should be noted, however, that the foundations for the legend that was the r300 were actually made by artX, by Dr. Wei Yen, a former silicon graphics engineer who also led the n64 GPU project.

In regards to polaris, yes raja did head up that design. Polaris was good, primarily because it was cheap. What it did not do was significantly put itself above the 290/x/390/x hawaii chips that AMD had already made, and sold, often for $200-$300 by the end of their lifespan. Many, including myself, asked where the higher end polaris chips were, when were we going to get an improvement over hawaii, and Raja's answer?

"well, buy 2 and crossfire"

That was a TERRIBLE response. Crossfire was already a sinking ship, as was SLI, the writing had been on the wall for years that big monolithic giant chips were the future of high end gaming. What it signaled everybody was that RTG, under raja, had spent the last 4 years as "rebrandeon"; reskinning their old GCN cards, and now had simply given up making a high end counterpoint to nvidia. Many, including myself on this very forum, called out a major issue, that was for all his bluster on "$200 GPUs making up a majority of buyers" there was no getting around nvidia being handled the entirety of the mid range and up on a silver platter.

The result? Nvidia ran with it. The $350 1070, the $550 1080, and the $700 1080ti had effectively 0 competition for, in some cases, nearly 18 months. The 1070 alone had higher margins per card then anything AMD sold, and outsold the entirety of polaris. As did the 1060. And to a point, so did the 1080. Nvidia made massive bank which would help fund turing, ampere, ece down the line.

Raja's genius plan was to pull a 3dfx, and split the resources of the meager RTG into two separate projects. The result of this was the embarrassment known as vega; a hot, power hungry, expensive card that was slower then the cheaper geforce cards and over a year late to market. Vega was an abysmal failure in the high end GPU space, even for the claims it was decent at some server work, RTG's server market would collapse during this era.

I could go on. I didnt even get into the embarrassment that was the "overclockers dream" fury/X cards. Nor did I get into drivers, because as you may have noticed, outside of a 4 year span Raja was in charge of ATI/AMD/RTG graphics (the span which gave us the great first gen GCN cards), and during that time period AMD gained its abysmal reputation for driver issues and instability. rDNA2, and AMD's modern, far better drivers, occurred after his departure. Lets not even get into the abandoned Radeon VII.

Now lets fast forward to intel. Raja joined in 2017. Now making a GPU from scratch is hard. Even so, with billions behind him, decades of experience, and some work already done in the form of the iris cores, Raja went to work.

The result? Alchemist was delayed. Repeatedly. It launched nearly 2 years later then planned. It's drivers were woefully incompetent, unexplained since intel's iGPU drivers, while not the best, at least worked properly. It's performance was not what was promised, power consumption was quite high, and it had numerous quirks like being useless without ReBAR. The launch was a wet fart, with little fanfare and few models available. Shortly after launch the GPU division is split in two, and now that raja is not the head, suddenly intel is responsive on getting better drivers out and promoting their products. Supposedly today, intel is shifting numbers similar to AMD, and raja is no longer there.

Raja, he may have done some great things in the 2000s, but his 2010s and early 2020's performance has been, to say the least, lackluster. Polaris was a weak shining star in a sea of half baked silicon and terrible software support; one that AMD had in its hand and just did nothing with despite its popularity. Intel's GPU division will likely do far better with him gone. Many want to compare him to jim keller, but keller always left after successful, often revolutionary launches. Athlon 64, the early generations of apple's A silicon, zen and zen+. Raja on the other hand either leaves right after a sputter of a launch or right before SHTF, let us not forget that 2009 was the last good year of evergreen, in 2010 fermi 2.0 embarrassed AMD all the way to the bank.

 

[wNotyarD]

That is a very......selective history.

Raja did some amazing work at ATi back in the day, serving as CTO from 2001 to 2009. That era gave us some great chips, the evergreen design, ece. It should be noted, however, that the foundations for the legend that was the r300 were actually made by artX, by Dr. Wei Yen, a former silicon graphics engineer who also led the n64 GPU project.

In regards to polaris, yes raja did head up that design. Polaris was good, primarily because it was cheap. What it did not do was significantly put itself above the 290/x/390/x hawaii chips that AMD had already made, and sold, often for $200-$300 by the end of their lifespan. Many, including myself, asked where the higher end polaris chips were, when were we going to get an improvement over hawaii, and Raja's answer?

"well, buy 2 and crossfire"

That was a TERRIBLE response. Crossfire was already a sinking ship, as was SLI, the writing had been on the wall for years that big monolithic giant chips were the future of high end gaming. What it signaled everybody was that RTG, under raja, had spent the last 4 years as "rebrandeon"; reskinning their old GCN cards, and now had simply given up making a high end counterpoint to nvidia. Many, including myself on this very forum, called out a major issue, that was for all his bluster on "$200 GPUs making up a majority of buyers" there was no getting around nvidia being handled the entirety of the mid range and up on a silver platter.

The result? Nvidia ran with it. The $350 1070, the $550 1080, and the $700 1080ti had effectively 0 competition for, in some cases, nearly 18 months. The 1070 alone had higher margins per card then anything AMD sold, and outsold the entirety of polaris. As did the 1060. And to a point, so did the 1080. Nvidia made massive bank which would help fund turing, ampere, ece down the line.

Raja's genius plan was to pull a 3dfx, and split the resources of the meager RTG into two separate projects. The result of this was the embarrassment known as vega; a hot, power hungry, expensive card that was slower then the cheaper geforce cards and over a year late to market. Vega was an abysmal failure in the high end GPU space, even for the claims it was decent at some server work, RTG's server market would collapse during this era.

I could go on. I didnt even get into the embarrassment that was the "overclockers dream" fury/X cards. Nor did I get into drivers, because as you may have noticed, outside of a 4 year span Raja was in charge of ATI/AMD/RTG graphics (the span which gave us the great first gen GCN cards), and during that time period AMD gained its abysmal reputation for driver issues and instability. rDNA2, and AMD's modern, far better drivers, occurred after his departure. Lets not even get into the abandoned Radeon VII.

Now lets fast forward to intel. Raja joined in 2017. Now making a GPU from scratch is hard. Even so, with billions behind him, decades of experience, and some work already done in the form of the iris cores, Raja went to work.

The result? Alchemist was delayed. Repeatedly. It launched nearly 2 years later then planned. It's drivers were woefully incompetent, unexplained since intel's iGPU drivers, while not the best, at least worked properly. It's performance was not what was promised, power consumption was quite high, and it had numerous quirks like being useless without ReBAR. The launch was a wet fart, with little fanfare and few models available. Shortly after launch the GPU division is split in two, and now that raja is not the head, suddenly intel is responsive on getting better drivers out and promoting their products. Supposedly today, intel is shifting numbers similar to AMD, and raja is no longer there.

Raja, he may have done some great things in the 2000s, but his 2010s and early 2020's performance has been, to say the least, lackluster. Polaris was a weak shining star in a sea of half baked silicon and terrible software support; one that AMD had in its hand and just did nothing with despite its popularity. Intel's GPU division will likely do far better with him gone. Many want to compare him to jim keller, but keller always left after successful, often revolutionary launches. Athlon 64, the early generations of apple's A silicon, zen and zen+. Raja on the other hand either leaves right after a sputter of a launch or right before SHTF, let us not forget that 2009 was the last good year of evergreen, in 2010 fermi 2.0 embarrassed AMD all the way to the bank.

The secret in hiring Raja is firing him later. You start badly, but then let time fix everything

 

[Vayra86]

He has only been there 5 minutes, they released one generation and hes gone? wow.

Fun facts!

Jim Keller (engineer) - Wikipedia

en.wikipedia.org

In August 2012, Jim Keller returned to AMD, where his primary task was to lead development of new generation of x86-64 and ARM microarchitectures called Zen and K12.[15][14] After years of being unable to compete with Intel in the high-end CPU market, the new generation of Zen processors has restored AMD's ability to do just that.[3][13] On September 18, 2015, Keller departed from AMD to pursue other opportunities, ending his three-year employment at AMD.[20]

In January 2016, Keller joined Tesla, Inc. as Vice President of Autopilot Hardware Engineering.[21]

Raja took two years longer Probably because of Xe initial delays. As an engineer, how much more can you do once designs are complete? I think as high profile workforce these guys drop their knowledge at a company, make something happen and move on. Everything they do after that is going to feel less interesting to them if they stay. A lot of a product's success hangs not just on its design but also the world around it, the company pushing it, etc.

Everyone associates Raja with Polaris, when he actually set the foundation for Navi while at AMD. Nonetheless, even Polaris was amazing value at the time (I miss those days compared to the shit show we have in the gpu market today). Absolutely nothing viable at a decent price, and all our wishes for AMD to become competitive so we can see better pricing did not come to fruition as you can see today (the fact Intel is our last hope is a very sad reality indeed).
With ARC, I dismissed it at first until I recently started paying close attention to it in reviews, and its actually quite impressive for what it is. The ray tracing implementation is solid. It beats out a 6700xt for much less. This is only a solid foundation for Intel to continue. If they stop now it would be very shortsighted of them. Anyways, thats what Raja (and other other architects) do in general. They join, lead a project or program up to a run and maintain/growth phase, and then move on to kick start something new. Jim Keller did that with Zen, and Raja did the same with Navi at AMD.

View attachment 288729

Navi was in fact delayed several times because of Raja and the blundering spree of releases that came out of his mind. Fury, Polaris, Vega and Radeon VII were all complete failures. Fury and Vega were the same failure on repeat (counting on cheaper HBM), Polaris failed because of its delayed time to market; it was surpassed in every way by Nvidia's midrange, also on price. Raja spent a lot of time rebranding and extending yesterday's-news GCN as well. We had Tonga. We had yet another 7970 as the 280X. All of that was fuel for AMD to get 1,5 ~ 2 generations behind on Nvidia. Imho the more accurate description of what Raja was doing there: he was throwing shit at the wall hoping something sticks. Nothing did, because there was no clear strategy and there was no real (marketing) story. Even the release of Vega, with frontier editions and all, was an exercise in trying to do way too much with a pretty poor product, forced into it because the cost of making one had to be made up somehow. Polaris and the aforementioned 'just crossfire them'... ouch. Just painful.

By the time "Navi" appeared in RDNA1, the overall conclusion was... its actually quite shit, and mostly just respinned GCN, and again, way too late. It wasn't until RDNA2 that AMD finally caught up. And then RDNA3 happened... and its a big pile of meh again, albeit with interesting new hardware.

It matters little what their day job is made of. I was just talking about results. Mr Keller puts you on top, Mr Koduri gets you second place in a two runners race.

Its not quite so easy to make a competitive GPU as it was for Keller to make strides on CPU, given the fact AMD was lagging behind massively on Intel.

You could say the releases are remarkably similar, first and second gen Ryzen weren't exactly beating every workload either. Especially on the gaming front. Constant Agesa updates required. Lots of early TLC required, bad ram support... Single thread king was still Intel. What Zen mostly did early on was nudge the market into 'moar cores' and it did offer more cores for the money.

And is ARC in such a bad place today? I mean, sure, its not topping charts, but its moving for sure. Is it competitive... eh... not quite, but Ryzen 1 wasn't killing sales either. It could certainly have been better, but perhaps a few generations from now, we might conclude Raja actually did put Intel on track much like Keller did AMD.

 

[BoboOOZ]

And is ARC in such a bad place today? I mean, sure, its not topping charts, but its moving for sure. Is it competitive... eh... not quite, but Ryzen 1 wasn't killing sales either. It could certainly have been better, but perhaps a few generations from now, we might conclude Raja actually did put Intel on track much like Keller did AMD.

I think technically ARC could be fine. The worst thing about it was that it came late, had it arrived 9 months earlier it would have been a commercial success, most probably.

It could still be fine if Intel holds on and does just midrange like AMD did with the 5700, while they keep working on drivers and software.

The biggest threats to ARC are due to coming out just after the crypto bubble, in a very bad market, and due to Intel potentially not having enough cash to hang in there for so long.

Oh, and about ARC moving, not so much. There was a lot of ARC shipped, not that much sold, because it's still not in a great position competing with the 6600s from AMD.

 

[lemonadesoda]

When person A can make a success of something, but only once, it means that it was not person A who was responsible for that success, but the environment and the team under A, that made the success despite A. But A stole the credit and lived off its coattails.

 

[eidairaman1]

Strike 2, Rajas Overpromise yet again.

 

[eidairaman1]

Oh dear, it's a shame as GPU market could use a third player.

Yeah Imagination Technologies Power VR