The King is Back: Raja Koduri Leaves Apple, Returns to AMD
by Anand Lal Shimpi on April 19, 2013 2:21 PM EST
I remember back when AMD’s CTO of the Graphics Product Group, Raja Koduri, first quietly left the company for Apple. This was hot on the heels of Apple’s hiring of another AMD GPU CTO, Bob Drebin. At the time (2009) I didn’t understand why Apple would want so many smart graphics guys on staff, were they working on their own GPU? Mac OS X was hardly a gaming platform of choice back then so the idea didn’t make much sense to me. It turns out that Steve Jobs wanted to surround himself with the absolute best in the business. Today, the impact of the work of folks like Bob Drebin, Raja Koduri, Jim Keller and others is quite evident. Apple tends to ship some of the fastest GPU hardware in the mobile industry, and its work in bringing high-DPI displays to virtually all of its products is unparalleled. Apple also played a huge role in driving the performance of Intel’s processor graphics. It turns out, that’s what happens when you hire a bunch of crazy smart GPU folks.
Last year AMD announced it re-hired one of the folks who eventually landed at Apple: CPU architect Jim Keller. Keller was responsible for one of the more memorable AMD CPU architectures, as well as played an unknown role in the development of Apple’s own CPU roadmap (Keller likely had a hand in the planning for Swift).
Today, word of another major hire surfaced earlier than expected: Raja Koduri now works at AMD once again. For the past 4 years Raja has been working on all things graphics at Apple, but now thanks to a great offer and a fear of getting too comfortable at Apple, he's back at AMD.
Raja Koduri returns to AMD as a Corporate Vice President of Visual Computing, the implication of the title is more important than its indication of how highly AMD values its latest hire. In Raja’s previous role as graphics CTO at AMD, he oversaw GPU hardware but didn’t directly oversee AMD’s software development. For the first time in recent history, both GPU hardware and software are falling under the same man.
The timeframe for Raja’s influence depends on what you’re talking about. Raja’s immediate goal is to ensure that AMD has the best GPU architecture/hardware possible. Unfortunately, it will likely take 2 - 3 years to realize this goal - putting the serious fruits of Raja’s labor somewhere around 2015 - 2016. Interestingly enough, that’s roughly the same time horizon for the fruits of Jim Keller’s CPU work at AMD.
Raja believes there’s likely another 15 years ahead of us of good work in high-end discrete graphics, so we’ll continue to see AMD focus on that part of the market. However, Raja’s efforts will span everything from discrete GPUs to its own APU/SoC strategy. There’s no change there from the path AMD has been on for the past few years now, but it’s good to hear that as much as low power matters, high performance is still on the to-do list.
More immediately you’ll see bits and pieces of Raja’s influence emerge. Everything from hardware to drivers and developer relations now falls under Raja’s umbrella. Presumably, all should improve - with the latter two seeing impacts sooner than the 2 - 3 year timeframe it would take for us to see results in hardware. We recently spoke about AMD’s multi-GPU frame pacing issues and I wouldn’t be surprised if that’s one of the first bits of low hanging fruit that Raja gets his teams to address quickly.
Raja returns to a very different AMD than the one he left. I asked him what’s different and he responded by saying the AMD he left acted like a company that was 10x its size. Today, AMD is a much smaller and more agile company. Raja believes AMD is in a better position to take advantage of new opportunities vs. being in the hopeless position of never being able to catch up in mature markets. It’s difficult to see what the next big mover will be after smartphones. Intel is betting on the TV industry, but to be honest no one knows. What drives Raja is the belief that delivering good quality, high performance and low power graphics will continue to matter going forward. I agree.
For years, Intel held back appreciation of good graphics performance in the PC space but the current mobile revolution has changed all of that. Graphics performance is just as important as CPU performance, a fact that Intel will itself embrace with Haswell/Broadwell.
Raja views AMD’s challenges as being difficult, but not unsurmountable. AMD needs a great graphics architecture and it needs a great SoC. Raja’s scope will include making sure that, at least on the graphics hardware/software/dev-rel side, AMD is in the best possible shape. Jim Keller will do the same for AMD’s CPUs. AMD remains focused on this idea of servicing vertically integrated players, although it's still difficult to see how that strategy is going to play out given how today's vertically integrated players (outside of game consoles) aren't big on using AMD hardware.
I’ve known Raja for over a decade now, and he’s always struck me as a man in search of a challenge. Given the revolution that we’ve seen at Apple over the past few years, Raja’s attraction to Cupertino made sense. However, I don’t think you could ask for a bigger challenge than the position AMD is in today. Raja (and others at AMD) believe there’s still room for new hires, although they are very interested in bringing in new blood.
Update: Some have asked for clarification on Raja's LinkedIn, reflecting his position as Technical Advisor to Makuta (Raja is full time with AMD, Makuta is a side project of his). I asked Raja and he offered the following response:
"I always had the dream of building a Pixar like company in India and I got an opportunity to engage with a group of people who have the same mission. AMD allowed me to stay engaged on this endeavor and that's one of the reasons I chose AMD."
Whether or not the past several high profile hires are enough to save AMD is something we won’t know for sure until 2015/2016. One thing is certain, AMD is trying to assemble a group of folks crazy enough to try and turn this company around.
Facebook
Twitter
RSS
72 Comments
View All Comments
phillyry - Saturday, April 20, 2013 - link
Good to see someone with some sense to interpret the situation.TheJian - Saturday, April 20, 2013 - link
A one page love fest over a hire after cutting off 30% of the engineers? One or two people won't fix those kinds of layoffs and even in your own words won't mean a thing until 2015-2016 (if they survive that long). Are you planning on analyzing them losing their butts again this Q? They're showing 146mil loss but reality is it was around $310mil if you take out the one-time sale for $164mil of the Austin property.Regarding the MULTI-GPU driver issues, it only took you a year and 7 months to mention it and I'm sure we won't hear another thing about it until AMD has it fixed. It’s no surprise AMD talked to this site ONLY directly and even then didn't discuss the REAL multi-gpu issues...LOL. So basically you wrote 7 pages of CYA direct from AMD. Why not just let them write it for you? Whatever...But still, these are the cards Anandtech has been recommending for basically a year now; runt frames, stutter etc all included for your enjoyment :) Enduro still sucks too says notebookcheck (and others look around some forums):
http://www.notebookcheck.net/Review-Update-Radeon-...
80 pages of complaining can't be wrong :) That's on just ONE site.
"Although AMD promised optimizations, the improvements were only small. Despite several performance gains and bug fixes, the graphics card never worked completely as it was advertised. Only since Catalyst 12.11 Beta, which brought some innovations, did it begin to function reasonably well, but still not without its problems."
Well, shucks...So still not working YET as advertised. Hmmpf...
"Although Nvidia's Optimus technology still performs better (Optimus is more reliable and features more visual notifications and comfortable functions), Enduro is now fairly useful."
Well, at least its "FAIRLY USEFUL" and functions "REASONABLY WELL" after a year...ROFL. But they still note problems they had previously in FarCry3, Modern Warfare3, F1 2012, Borderlands 2 and Assassin's Creed 3. I guess that's why other than farcry3 these aren't in your gaming suite?
I can't wait to see Ryan's take on Adobe's OpenCL additions while leaving out the comparisons to Cuda in CS6 or Premiere Next (anandtech already ignores Cuda). No doubt he'll fan the OpenCL flames that are barely burning vs. Cuda, while acting like Cuda doesn't exist. Only an idiot buys an NV card and tries to do ANYTHING in OpenCL. You should seek CUDA apps for the same jobs and avoid OpenCL at all costs on NV hardware. I'm not quite sure why AMD was using a Xeon5530 @2.4ghz from 2009 for their tests on a 2013-2014 Adobe product...Fishy at best there? Why wasn't this a 3.5ghz+ I7 or something? Did they use a notebook driver 311.35? I'm thinking NV suddenly works better if using an actual state of the art cpu vs. a 2009 4yr old tech. Amazon recommends a i7-3930K 6 core with 12mb cache to go with the K5000 Quadro they sell, not some chip from Q1'09. Also note AMD hasn't been left in the cold all this time, OpenGL even according to NV's own site says any card that supports it would be accelerated in Adobe CS6:
"However, all OpenGL 2.0 graphics cards including NVIDIA GeForce, NVIDIA Quadro, AMD Radeon and AMD FireGL will accelerate Adobe CS6."
https://nvidia.custhelp.com/app/answers/detail/a_i...
Straight from NV site, even AMD works in opengl already
For a real read of AMD's gpu problems, start reading RYAN SHROUT at pcper instead of Ryan Smith here. All you get from here is a 7page AMD excuse in the form of an article supposedly covering stuttering.
"They’re already working on changing how they do frame pacing for multi-GPU setups, and come July we’re going to have the chance to see the results of AMD’s latest efforts there."
http://www.anandtech.com/show/6857/amd-stuttering-...
Translation: We'll put this talk off until AMD fixes it, then boast about the hard work they've done (at fixing a problem NV already fixed, hence FCAT). Scott Wasson reported on this stutter etc issue sept2011. So it only took you a little over a year and a half to catch up. WOW. So in July we'll finally get to see what our cards should have been doing all year (MAYBE?)?? Thanks for recommending these works in progress for a year over NV :)
"There are tools in development that meet our criteria for better measuring frame intervals, and hopefully in the not too distant future we’ll be able to discuss those tools to a much greater degree, and to use those tools to go about measuring frame intervals in the manner we’ve always wanted to. But that is a story for another day, so until then you’ll have to stay tuned to find out."
Tools are already developed. It's called FCAT, and I'm sure NV would be glad to get one in your hands for something more than that 3 pager you gave without any data a few weeks back. Was FCAT even mentioned in that glorious 7 page AMD CYA stutter excuse article above?...LOL. I wouldn't have to "stay tuned" if you'd have bothered to tell the truth of AMD's drivers situation for the last year (as hardocp, notebookcheck, pcper etc have pointed out) rather than doing AMD CYA all year. Let me know when their products actually work as sold. You gave a 3 pager on FCAT a few weeks back (mar 27th) but it's really more of the same. Delay until AMD gets a working solution rather than discuss the problems directly NOW before they fix it.
I have a lot of love for AMD too, but not when they suck. They look to be on track for yet another Billion dollar loss this FY ($310mil in the bucket already for Q1 if you add the Austin 164mil sale). I'm guessing you'll be ignoring NV profits again this quarter as usual. AMD still owes 240 mil if memory serves to GF (with 200mil due Q1'14, 40mil next Q I think), and the CFO just said they'll be needing loans to survive this year. No real revenue from consoles until next year, no arm until next year, and more free games to lower what little the gpu division makes even today (you can't give 5 games away and expect profits). Ryan Shrout has done probably a dozen articles on FCAT/Runts etc, but this place? I'm baffled by all the AMD love on this site. What we've been seeing for a year is the results of 30% of your engineers exiting in one way or another. Your drivers basically sucked for a year, finally got decent with never settle in Nov (with the included upping of the clocks to ghz - which NV answered with 4 monthly perf drivers since Dec), but you continue to have driver issues today. Now we're waiting AGAIN for july to get a working product? I'm not saying they don't function at all, but clearly they are not functioning as designed. Pcper just did another article on AMD's useless vsync claim they made at GDC13. It doesn't solve the issue and artificially caps your gpu in some instances. So their solution is to tell me to slow down my cards and pray a real driver fix is around the corner? Yeah ok...Just like Enduro, let me know when it works, maybe then you can justify claiming AMD is a good buy. Until then you've basically been selling a load of BS on anandtech.
At some point don't you have to call a spade a spade? How much longer will you try to prop up AMD with dumb benchmarks like Crysis Warhead, SniperV2 and Dirt Showdown? Nobody plays ANY of these games. Literally NOBODY according to a quick server check of warhead (0 players) and Dirt Showdown + Sniper sales don't even register on VGchartz (because you actually have to sell some units to make the chart-you may have just dropped sniper though from testing). Luxmark for compute? Why don't you run CUDA stuff? Fire up Octane or Iray etc for rendering instead of beating the dead horse that is OpenCL. Can't resists statements like this I guess "Moving on to LuxMark, we quite frankly transition into a more normal compute benchmark pattern for NVIDIA, which sees Kepler flopping." from your 650TI Boost article. Your articles all act like a normal person would fire up OpenCL on NV. That's moronic at best, as I'd find the cuda enabled version of whatever to do the same job. Luxmark is just showing a WORST case scenario for an NV card by forcing it to OPenCL when other options exist that are far better. So you can only say your snide remarks because you're acting like users would actually do something so stupid and NOT seek a CUDA app. For a real test, how about using LUXrender in Blender for AMD, and iray or octane in Blender for NV (or choose another, they both support 3dsmax, c4d, maya etc). That would actually prove something; Who really is the best when using their best scenarios that we'd all seek upon buying. I'd rather jump off a bridge before using OpenCL on NV cards...LOL. Still doing folding@home? Only 165,000 people even have this installed on their machines (350mil pcs sold each year). NOBODY CARES about folding@home. You openCL us to death, without a word about Cuda in every NV article. You should be testing the crap out of cuda vs. an equivalent app for AMD with OpenCL/OpenGL. You keep testing junk nobody uses to make money and saying NV sucks at compute. Total lies. They have CUDA in nearly every pro app for this. If RAYTRACING has become so important, why are you NOT USING CUDA for NV to do it? You are hiding the truth at best, or LYING at worst.
"Wrapping things up, we will be following up this article next week with part 2 in our look at FCAT. "
That was March 27 here...Still waiting for part2 nearly a month later...Keep kissing that AMD butt guys :) Just put it off until July, it's all good. Maybe you guys can help them make a profit finally ;)
silverblue - Saturday, April 20, 2013 - link
Sometimes, all it takes to shake things up is hiring a good manager, somebody with drive and the right priorities.Both Intel and AMD are promoting OpenCL; shouldn't NV try to join the party? I agree that there are viable alternatives but they're usually vendor specific. I think it'd be wrong to assume that AT doesn't run tests using either of those or that neither of them would be valid in any use case. The ideal situation would be to see how different methods work out, time permitting. However, if you don't get that here, you can always check another review to bolster your opinion as to how a specific card or architecture works at a designated task. If I was to make a purchase, that's certainly what my approach would be.
Regarding the FCAT issue... FRAPS was a good indicator but never quite accurate enough to point out exactly what was wrong where. FCAT has only really just arrived on the scene; that follow-up article can't come soon enough.
AMD have set a target of July for significant improvements. I'm not sure there's a lot of point in trying to see improvements until then, however do bear in mind that latency issues are far better than previously for single card users.
You make some very good points, however I'm not sure AT are working a second job as AMD apologists.
Will Robinson - Sunday, April 21, 2013 - link
Is that you Rollo?WaltC - Saturday, April 20, 2013 - link
"Whether or not the past several high profile hires are enough to save AMD is something we won’t know for sure until 2015/2016."Ah, yes--you can't be a tech pundit these days unless you have something smarmy and snide to say about AMD. I thinking knocking AMD is the mark of the half-literate tech pundit who is in over his head but desperately trying not to show it. Evidently, the people AMD is hiring feel the company has a definite future--and, obviously, one that goes far beyond 2015.
As has been noted, most of ill-informed, opinionated tech punditry is loathe to mention AMD's recent console wins except grudgingly, if at all. AMD will survive and prove them wrong once again, to their utter frustration, and it's always gratifying to see the pundits eating crow. Eating crow once again, I should say, for this is surely not an isolated occurrence...;)
HisDivineOrder - Saturday, April 20, 2013 - link
I find this article's hero worship (or bromance?) to be a little off-putting. Leaders are important, but when AMD just excised huge portions of its staff, including a lot of its R&D, I think there's a real question of how much talent they still have left and how much of it is quality talent that deserves the big bucks.I can't help feeling like AMD is hiring these big names to lead their divisions so they can PR spin their talent as better than it actually is. If you had the choice between no-name leaders who are good leaders plus lots and lots of talented engineers or a known name leader and very few engineers of indeterminate quality, which would you choose and why?
If you wanted great products, you'd go for the best combination of engineers, even if they were led by someone who didn't have the reputation for being a big idea guy. If you wanted to convince OTHERS you were fine and everything is great, you'd spend less and still have the big guy to sell the spin you're okay.
Looking at AMD's moves in the broader sense, it seems they're doing a lot of things that have short term spin effects (focus on fixing drivers/bundles rather than launch a refreshed GPU series, release Richland APU's instead of pushing on to Steamroller-based parts, hire two big hitmakers of old to smokescreen the loss of all the talent they've fired over the last year) rather than doing anything much in the mid-term or long term. These guys will have an effect if they can make it to the longest term.
If they can make it. At the rate they're losing cash even in quarters where they aren't even bothering to release new products despite the ones they're selling being long in the tooth, I think they're in real trouble and may not make it that far.
If that's the case, then having these guys come in for some big paychecks until AMD sets themselves up for a capital infusion, a merger, or buyout would make a lot of sense. Later, they can quietly announce these guys have moved on once they get through this rough patch and the positive PR only cost them a couple guy's big paychecks.
That seems much more realistic to me than the idea that one man or two men are going to show up and just magically turn AMD from the Titanic into the TARDIS.
Paulman - Sunday, April 21, 2013 - link
I can't help but think at least one of the reasons for the unbridled optimism is that "we" (as in the computing sector) need AMD to do well so that they don't go bankrupt and leave Intel with no competition. Otherwise, it's an Intel monopoly which is probably not the best for the consumer and arguably not the best for Intel, as well (because the US government may get more interventional in the interests of anti-trust/monopoly concerns).A healthy AMD is good for all the other markets they compete in, as well (graphics, smartphone/tablet).
lmcd - Sunday, April 21, 2013 - link
I'd disagree, it'd leave it with a desktop monopoly, but what about ARM? The A15-based Chromebook is a good basic platform for computers, and IBM + PPC is good for workstation and server as I understand. The unserved market here is then only gaming and high-end personal computer users, which hopefully ARM's next cores will move towards addressing.FearfulSPARTAN - Sunday, April 21, 2013 - link
I highly doubt even arm 64 bit processors will come close to intels and even then thats still one whole sector with one cpu company in control of pricingzed1 - Sunday, April 21, 2013 - link
If AMD don't get behind open source multimedia they are not going to progress. Maybe you should ask Raja why AMD are completely unprepared to back open source multimedia solutions and rather spend their rapidly diminishing resources on proprietary competitive solutions using third party companies run by their mates...