News Raja Koduri Leaves Intel to Found Software Start-Up

[DaveLTX]

I have build pcs with ATI rage 9xxx 1xxx 2xxx 3xxx 4xxx 5xxx 6xxx 7xxx. Moved to nvidia come back with the 5xx series. Can't buy a Vega because the price and power. But even today can't find a cheap used Vega, before raja and after ATI (AMD) still suck.

I don't know what your fuss with AMD are. I've not been with AMD since my R9 290X and I've returned to AMD now with my 6900XT that I got for the price of a 3070.

I've not had driver issues or performance issues. Had a 3080 before this too so I'm very much in tune with current gen

 

[Ogotai]

But even today can't find a cheap used Vega, before raja and after ATI (AMD) still suck.

just like i said here, your issues with amd based products, the problem could just be your own fault, you push them too far, or you do something that doesnt work, you break it, then you blame amd for it.

 

[bit_user]

even today can't find a cheap used Vega,

Probably because they either break, or are in continual demand by people using them for things like Davinci Resolve.

I sold a Radeon VII on ebay, so I've looked at the listings enough to see several listed as "non-working / parts only". That's evidence that failures aren't too uncommon.

before raja and after ATI (AMD) still suck.

Why do you say they suck? The RX 6000 series seemed to do quite well against RTX 3000. Right now, they're a bargain.

 

[bit_user]

OMG. Has it been 5 years, already? Seems not that long ago that he landed at Intel.
: O

so intel are preparing to fully pull away from discrete GPU market.

Or, maybe it's because they're still serious about dGPUs that they're doing a re-org.

 

[bit_user]

Regardless, his portfolio will now have two GPU types in it, one of which was a failure and both of whom have ridiculous nomenclature. I mean, seriously, who thought that "Vega" was a good name for a video card? I thought that it was hokey as hell and who is going to take a card called "Battlemage" seriously? Arc wasn't a great name but it was sure as hell better than Battlemage.

I'm pretty sure I remember reading about how Raja wanted to name GPUs after stars. Hence, we got: Polaris, Vega, Navi, Arcturus, and Aldebaran (others?). The last two are CDNA chips (MI100 and MI200), so they didn't get a lot of press.

...but I seriously doubt he's responsible for Intel's branding strategy. Intel is/was a much bigger company and I'm sure their marketing department had quite some clout. I think there's very little chance they would leave their dGPU branding strategy up to engineers.

As for "taking it seriously", these names are for gaming GPUs, so it's little surprise they leaned into that culture. If you look at the datacenter equivalents, they have names like Artic Sound and Lancaster Sound. Intel knows their markets.

IMO, the names are the least problematic things about any of these products. Total sideshow.

Vega came from AMD's "stars" series of GPU codenames

With some rather unfortunate connotations around heat & power, I dare say.
: O

Prior to that, AMD used the "Arctic Islands," "Volcanic Islands," "Sea Islands," "Southern Islands," and "Northern Islands" as codenames.

I like when AMD did that, because each GPU die would have its own name within that theme.

Now, we just get variations on the same thing (i.e. Vega20, Navi31, etc.). According to an AMD engineer, the first digit is the generation and the second one is/was the order in which it was developed.

 

[bit_user]

Generative AI gaming? So, the kind of AI designed to replace character artists, environment artists, level designers, quest designers, and so on.

Not necessarily. Maybe the focus is more on actual rendering technology, which is indeed a very interesting new development.

Sure, let's make the lives of people, who are already suffering under terrible management and abusive hours, even more stressful.

There will always be a need for them, even if the amount of them becomes less. If the skillset required becomes more specialized, they'll have more clout (like developers). Ultimately, these are labor problems and the solutions don't really have to do with the specific technologies in use. Don't want to get sidetracked on that, but these struggles are hardly new.

he's going into a space that threatens a lot of my friends who don't have something to fall back on.

My best wishes to them.

 

[bit_user]

I've always been a little amazed at how this man can be divisive (and with divisive I mean most people, especially AMD enthusiasts, hating his guts).

We just remember how hot, late, and under-performing his products were. Polaris was mildly disappointing, non-competitive on perf/W, and we got 2 generations of it! Vega was like a year late, unprofitable for AMD, and underperformed where it was targeted. Nobody used it for AI, which was supposed to be a big selling point. The only reason that wasn't catastrophic for AMD was thanks to the crypto boom.

If we're being fair, Raja likely deserves some credit for RDNA. As with the prior products, it came almost too late. Maybe he did get stuck with more than his share of blame, though. But, when you're at the top, you get a lot of undeserved criticism along with undeserved praise (and it goes down much easier with the big compensation package).

Based on a few comments I've heard from a couple AMD folks, it sounds like he was full of hot air. They definitely weren't speaking respectfully of him.

 

[dalek1234]

"Insider information on Raja being pushed out of Intel"
Skip to time index 1:43, although the rest of the video is interesting too:

View: https://www.youtube.com/watch?v=HzV1RS5Oc6I

 

[dalek1234]

Let's say that Intel is planning to cancel Arc, getting rid of Raja before that announcement would make sense. Maybe that's the plan.

 

[bit_user]

Let's say that Intel is planning to cancel Arc, getting rid of Raja before that announcement would make sense.

Huh?

No, I think big companies like to release all their bad news at once. If they were about to kill Arc, then they'd just announce that. Whatever happens to Raja might be a footnote, in such an announcement. This almost certainly means they're not decided on killing it, but instead probably looking for someone to do a turnaround to try and save it.

 

[dalek1234]

Huh?

No, I think big companies like to release all their bad news at once.

You're probably right, announcing both at the same time does sound more sensible.

What I found curious from the video I posted above, is that Raja-leaving-Intel announcement communicated by Intel to employees, was buried as a footnote in another unrelated announcement. People from Intel who leaked the info say that senior people, when departing, receive a separate anoucemet to employees, not as a foot note in another announcement. Why is Intel trying to publicise this news, to emplyees, as little as possible, especially so knowing that this is atypical for Intel to do?

 

[bit_user]

People from Intel who leaked the info say that senior people, when departing, receive a separate anoucemet to employees, not as a foot note in another announcement. Why is Intel trying to publicise this news, to emplyees, as little as possible, especially so knowing that this is atypical for Intel to do?

I can't say, but he was re-org'd not that long ago:

https://www.tomshardware.com/news/intel-re-orgs-axg-graphics-group-raja-koduri-moves-back-to-chief-architect-role

Might've been something they did to prepare for his departure.

 

[Howardohyea]

so intel are preparing to fully pull away from discrete GPU market.

I'm very confident Intel will at the very least follow through their roadmap to Druid instead of cutting it now, but there's no saying what would be after that.

It's about damage limitation, imagine opening a shop and closing it the next day after serving only 5 customers, at the very least you want to recoup a quarter, if not a half, of the total money you invested, and who knows if it'll become profitable in the long run?

 

[Avro Arrow]

Vega came from AMD's "stars" series of GPU codenames, the other being Polaris. Prior to that, AMD used the "Arctic Islands," "Volcanic Islands," "Sea Islands," "Southern Islands," and "Northern Islands" as codenames. Things had gotten a bit screwy near the end (the "Arctic Islands" which later became Polaris) and so AMD was looking elsewhere, and with Vega they changed the codenames to "stars." But apparently the fans liked Vega enough that it became the actual product name.

Battlemage, like Alchemist, is just a codename. We don't call the Arc A770 an Alchemist, unless we're talking about the architecture, and we won't call the Arc B770 a "Battlemage" unless we're talking about the architecture.

I know what the nomenclature was, it was just handled REALLY weird.

The RX 580 wasn't called "POLARIS 580" and I don't remember seeing the word "HEMLOCK" draped across the cooling shroud of the HD 5970.

In that oddball era, Radeons were named like they existed in a different reality that didn't really say much about them. Instead of the more conventional Radeon nomenclature that was comprised of a two-digit prefix followed by a number (and possibly a suffix), names like "Vega 56" and "Radeon VII" were used. Those names were totally lacking in context compared to every other card to come out of Markham.

Maybe AMD thought that naming Radeons after stars would have brought them the same success that they achieved with "Deneb" and "Thuban".

 

[waltc3]

It's not like they had nothing, but admittedly AMD is still using Vega in their current generation laptop APUs (courtesy of Barcelo R, which is technically the refresh of a refresh actually). I've always been a little amazed at how this man can be divisive (and with divisive I mean most people, especially AMD enthusiasts, hating his guts).

Raja is one those people who garnered far more attention than he actually deserved, courtesy of AMD, which elevated his public profile, which I think was probably a mistake as Koduri's head swelled accordingly. If it was up to Raja, AMD might never have gotten past Vega in some form or other. As it is, AMD's GPU products and ambitions have skyrocketed since he left, which says a lot. BTW, AMD's newest mobile gaming chips will use RDNA3, which lets Vega out. You have to recall that AMD is only now getting serious about the laptop segments of the markets, and Vega is already dead in the desktop discrete markets. I think that singling out "rock stars" of the hardware biz is a big mistake, just as when Jim Keller left AMD and went to work for Intel for a brief period, I saw Intel fans boasting as to how Keller's presence would see Intel catching up with AMD soon--but that was before Keller left Intel after a brief employment. IIRC, Keller said he couldn't get Intel to listen to him, so he left... But the main thing to take away is that successful CPU and GPU hardware designs are 100% collaborative efforts involving hundreds of people, or more, and sometimes a narrow vision near the top can restrict or setback everything. While Keller rarely if ever flaunted himself that I saw, Raja seemed to bask in the attention. I don't recall AMD giving Raja a hard time when he left...with Polaris and Vega, and with the Mantle API, Raja had pretty much given AMD everything he had. Hardware rock stars rarely meet their billings, which is reason enough to avoid creating them.

 

[DaveLTX]

Raja is one those people who garnered far more attention than he actually deserved, courtesy of AMD, which elevated his public profile, which I think was probably a mistake as Koduri's head swelled accordingly. If it was up to Raja, AMD might never have gotten past Vega in some form or other. As it is, AMD's GPU products and ambitions have skyrocketed since he left, which says a lot. BTW, AMD's newest mobile gaming chips will use RDNA3, which lets Vega out. You have to recall that AMD is only now getting serious about the laptop segments of the markets, and Vega is already dead in the desktop discrete markets. I think that singling out "rock stars" of the hardware biz is a big mistake, just as when Jim Keller left AMD and went to work for Intel for a brief period, I saw Intel fans boasting as to how Keller's presence would see Intel catching up with AMD soon--but that was before Keller left Intel after a brief employment. IIRC, Keller said he couldn't get Intel to listen to him, so he left... But the main thing to take away is that successful CPU and GPU hardware designs are 100% collaborative efforts involving hundreds of people, or more, and sometimes a narrow vision near the top can restrict or setback everything. While Keller rarely if ever flaunted himself that I saw, Raja seemed to bask in the attention. I don't recall AMD giving Raja a hard time when he left...with Polaris and Vega, and with the Mantle API, Raja had pretty much given AMD everything he had. Hardware rock stars rarely meet their billings, which is reason enough to avoid creating them.

Vega was dead ever since AMD finally put RDNA2 in their laptop CPUs, as to Barcelo being a respin of Cezanne, it is what it is. GCN allegedly takes up less die area but also, it appears the reason is that RDNA2 iGPU blocks seem to require DDR5 or come with them.
On the same node there's no question that RDNA2 is hell of a lot more efficient so the only reason for them to continue making them for cost sensitive markets that require more performance than mendocino? even mendocino requires LPDDR5

Yes indeed, Keller mentioned in passing that Intel was practically like talking to a stubborn child

 

[-Fran-]

And here I still have and use my Vega64 on my 1440p 144Hz Samsung G8 with no problems whatsoever. From what I'm reading here, it should be dead or giving me issues, but alas...

Anyway, the writing was on the wall for Raja when AXG was chopped. MLiD gets kudos on this one, at least partially so far, on at least making people very aware things were not rosy at all inside of Intel at the time and all of that is transpiring even with Raja's and Pat's denials for stockholder value (which is obvious they'd deny in a legally non-bounding way).

I still think Intel's GPU venture is salvageable, but it won't be cheap and they're already bleeding money and patching the leaks as they try to move forward... Looking bleak. I smell the same destiny as Optane and the modem division: they'll be gotten rid of out of nowhere with no notice.

Regards.

 

[Chung Leong]

The first nail in the farewell coffin for Intel ARC graphics cards for gamers. Raja Koduri jump the ship. Or kicked out by Gelsinger (the best or easiest way without much fuzz/fuss or speculations). Yep, Pat Gelsinger will finally have the chance to get rid of this loss project.

I suspect the truth is probably not too far from the official version: guy just wanted to move on. Intel's GPU effort has moved into the stage where the company is trying to recoup its investment. It's a lot less fun than the initial stage, when the team gets to spend money like there's no tomorrow.

 

[RedBear87]

As it is, AMD's GPU products and ambitions have skyrocketed since he left, which says a lot.

Did they? AMD was the underdog back then and it's still the underdog now; AMD had a niche for efficiency back in the previous generation, but even that was lost when Nvidia switched back to TSMC for A. Lovelace, now they need to push more power for comparable rastering performance and inferior ray tracing... Putting it all on Raja's head looks unfair to me, honestly.

We just remember how hot, late, and under-performing his products were. Polaris was mildly disappointing, non-competitive on perf/W, and we got 2 generations of it! Vega was like a year late, unprofitable for AMD, and underperformed where it was targeted. Nobody used it for AI, which was supposed to be a big selling point. The only reason that wasn't catastrophic for AMD was thanks to the crypto boom.

Curiously Polaris remained fairly popular on a price-to-performance ratio for a lot of time, while AMD being bad at AI didn't really stop being an issue after he left, even now ROCm is basically nearly unknown, usable only on Linux if you're really stubborn.

 

[bit_user]

I'm very confident Intel will at the very least follow through their roadmap to Druid instead of cutting it now, but there's no saying what would be after that.

It's about damage limitation, imagine opening a shop and closing it the next day after serving only 5 customers, at the very least you want to recoup a quarter, if not a half, of the total money you invested, and who knows if it'll become profitable in the long run?

Intel isn't afraid of sunk costs, and I'm sure they've barely put anything into Druid, right now. According to what Koduri said late last year, probably the bulk of the team is still on Battlemage.

 

[bit_user]

The RX 580 wasn't called "POLARIS 580" and I don't remember seeing the word "HEMLOCK" draped across the cooling shroud of the HD 5970.

Don't think too hard about "marketing logic". If a name sounds good, they're likely to use it.

FWIW, HD 5970 wasn't called Hemlock - the GPU's codename was Tahiti. That's a fine code name, but I don't think it works as well as Polaris, with respect to marketing and selling GPUs.

 

[bit_user]

AMD's newest mobile gaming chips will use RDNA3, which lets Vega out.

AMD already moved past Vega iGPUs more than a year ago, in the Ryzen 6000 series. They used Zen 3+ and RDNA2 on TSMC N6.

I think that singling out "rock stars" of the hardware biz is a big mistake, just as when Jim Keller left AMD and went to work for Intel for a brief period, I saw Intel fans boasting as to how Keller's presence would see Intel catching up with AMD soon--but that was before Keller left Intel after a brief employment. IIRC, Keller said he couldn't get Intel to listen to him, so he left... But the main thing to take away is that successful CPU and GPU hardware designs are 100% collaborative efforts involving hundreds of people, or more,

I used to agree with this, but I've since come to think that Jim's praise is largely well-deserved. He's also had a longer and more illustrious career than Raja. As for the team aspect, that seems like a major part of Jim's contribution, in fact. Upon his return to AMD, it sounds like the first thing he had to do was breakdown their old way of thinking and operating. I'll quote from Ian Cutress' 2021 interview:

​

IC: Correct me if I’m wrong, but for the time inside AMD, it kind of sounded like Jim’s way or the highway?​

​

JK: I wouldn't say that! The funny thing was, we knew we were kind of at the end of the road - our customers weren’t buying our products, and the stuff on the roadmap wasn’t any good. I didn’t have to convince people very much about that. There were a few people who said ‘you don’t understand Jim, we have an opportunity to make 5%’. But we were off by 2X, and we couldn’t catch up [going down that route]. So I made this chart that summarized that our plan was to ‘fall a little further behind Intel every year until we died’.​

​

With Zen, we were going to catch up in one generation. There were three groups of people - a small group believed it (that Zen would catch Intel in one generation); a medium-sized group of people that thought if it happens, it would be cool; then another group that definitely believed it was impossible. A lot of those people laughed, and some of them kind of soldiered on, despite this belief. There was a lot of cognitive dissonance, but I found all kinds of people that were really enthusiastic.​

​

https://www.anandtech.com/show/1670...storrent-ceo-ljubisa-bajic-and-cto-jim-keller​

sometimes a narrow vision near the top can restrict or setback everything.

The above example shows how crucial it is to have good leadership. A bad leader can definitely have the effect you describe.

... with the Mantle API, Raja had pretty much given AMD everything he had.

Wast Mantle his baby? His return to AMD was announced in late April, 2013. This was published just 6 months later. It seems to me that work on Mantle must've started before he took charge.

https://www.anandtech.com/show/7431/amd-blog-post-the-four-core-principles-of-amds-mantle​

I assume Mantle arose in the course of AMD's participation in Khronos' OpenGL working group, as they struggled to define "OpenGL Next", which would later become known as Vulkan. I think they wanted to show what was possible, if you deconstructed the existing GPU APIs.

 

[bit_user]

Looking bleak. I smell the same destiny as Optane and the modem division: they'll be gotten rid of out of nowhere with no notice.

The key difference is that Intel needs iGPUs for its CPUs. If it doesn't design them internally, then they'll have to license IP from someone else. That makes their graphics division somewhat safer than the other things you mentioned. Especially if tGPU come to more closely resemble dGPUs in their design and operation.

 

[bit_user]

AMD being bad at AI didn't really stop being an issue after he left,

But Vega was neither very good at rendering nor AI! At least RDNA has enabled AMD to be more competitive on raster performance.

You're right about their AI challenges, including the whole ROCm aspect. That's something they're still struggling to come to terms with, now potentially made more complicated by the involvement of Xilinx.

 

[DaveLTX]

Don't think too hard about "marketing logic". If a name sounds good, they're likely to use it.

FWIW, HD 5970 wasn't called Hemlock - the GPU's codename was Tahiti. That's a fine code name, but I don't think it works as well as Polaris, with respect to marketing and selling GPUs.

Tahiti is 7970 my boy