|
Bye raja, I still can't believe your smiled at that system running Vega and suckered me in. I believed in you, rear end in a top hat.
|
# ? Nov 8, 2017 00:17 |
|
|
# ? Jun 5, 2024 08:29 |
|
Linus dropping another graphics card just broke the poor man.
|
# ? Nov 8, 2017 00:18 |
|
Realistically speaking what does this mean for AMD/RTG in the short and medium term?
|
# ? Nov 8, 2017 00:27 |
|
Short term I bet their price goes down. The announcement came after hours but I bet there's a dip tomorrow morning, the stock market hates anyone getting fired. Long term I think they're going to rise, Su has been super successful all over the electronics industry and Raja has like a decade worth of unsuccessful projects. She was the one in charge of Ryzen and that turned out well.
|
# ? Nov 8, 2017 00:33 |
|
Nyaa posted:I hope he went straight to work with Nvdia. why
|
# ? Nov 8, 2017 00:43 |
|
Zedsdeadbaby posted:Realistically speaking what does this mean for AMD/RTG in the short and medium term? Realistically we're probably ascribing too much importance to the person at the top vs the team making the hardware. It's hard for me to believe that GCN has any gas left in the tank anymore, and Su has to know that as well. MCM-based GPUs are an additional increase in efficiency, but NVIDIA can make the same leap so they're not going to compensate for NVIDIA having >2x the performance-per-watt, and there's a lot of additional technical challenges that come with MCM designs. They need a new uarch next time around. If I were them, I would come out with a 750 Ti first - something to prove their uarch on a small scale. Ideally you would want it to have an appropriate Infinity Fabric interconnect so that you could start playing with MCM designs as early as possible, assuming that it doesn't consume an excessive amount of die space. If that's viable you stitch together a midrange product. Then you start taping out a bigger version for your flagship. But, this is all speculation, only AMD knows what they're going to do. I think they realize that while RTG may be struggling, they get a whole bunch of really important synergies from it. Although I guess nowadays Intel has the same synergies... I am actually still surprised about that. And it was Kyle Bennett of all people. Paul MaudDib fucked around with this message at 00:51 on Nov 8, 2017 |
# ? Nov 8, 2017 00:48 |
|
Kyle Bennett also said there was some friction between Lisa and Raja IIRC, seems Raja lost, but what the fight was about remains to be seen based on Lisa's moves. If Ryzen is an indication, Lisa wants to target the high end first, both in consumer and commercial, and let everything else trickle down from there. I think this makes some sense if we assume the reason Raja is leaving is Vega did not deliver on that target, while Raja wanted to go back to R700 and Cypress bang for buck days.Paul MaudDib posted:They need a new uarch next time around. If I were them, I would come out with a 750 Ti first - something to prove their uarch on a small scale. Depending on how Raven Ridge performs, they may already have their 750 Ti in Small Vega, it's just that it scales for poo poo. If 24CUs @ 1190Mhz can be done inside a 55W TDP then it kind of seems Vega can match Pascal in perf/watt (48CUs would be under 130W, 64CUs @ 180W), but hates going beyond 1300Mhz and bottlenecks like crazy past 48CU it seems. Maybe the new process solves the former issue but you obviously need a redesign of some sort to address the latter.
|
# ? Nov 8, 2017 01:07 |
|
Should have sold to samsung
|
# ? Nov 8, 2017 01:12 |
|
Nyaa posted:I hope he went straight to work with Nvdia. https://wccftech.com/exclusive-raja-koduri-will-seeking-new-horizons-intel/
|
# ? Nov 8, 2017 01:13 |
|
AMD literallly just hired former Micron CEO Mark Durcan a hot minute ago, one wonders if he's taking over Raja's old office. https://globenewswire.com/news-release/2017/11/01/1172200/0/en/AMD-Appoints-Mark-Durcan-to-Board-of-Directors.html
|
# ? Nov 8, 2017 01:32 |
|
SwissArmyDruid posted:AMD literallly just hired former Micron CEO Mark Durcan a hot minute ago, one wonders if he's taking over Raja's old office. didn't someone make that call in a thread about a week ago?
|
# ? Nov 8, 2017 02:00 |
|
Yes. Me. =P And everyone just kind of skimmed over it and nobody said anything. =P
|
# ? Nov 8, 2017 04:14 |
|
who i thought this thread was up to date
|
# ? Nov 8, 2017 04:42 |
|
FaustianQ posted:Depending on how Raven Ridge performs, they may already have their 750 Ti in Small Vega, it's just that it scales for poo poo. If 24CUs @ 1190Mhz can be done inside a 55W TDP then it kind of seems Vega can match Pascal in perf/watt (48CUs would be under 130W, 64CUs @ 180W), but hates going beyond 1300Mhz and bottlenecks like crazy past 48CU it seems. Maybe the new process solves the former issue but you obviously need a redesign of some sort to address the latter. Supposedly NGG fast path is not active yet, although this is probably just False Hope #3267 at this point. Frankly one thing I am concerned about is that a bunch of TBR/DSBR gains supposedly come from cache locality - i.e. having the thing you want cached into L2. Does that still happen if you have multiple chips? I wonder...
|
# ? Nov 8, 2017 04:48 |
|
Can I just stop here for a moment and mention how great the signal-to-noise ratio is on this forum? Y'all rock
|
# ? Nov 8, 2017 06:20 |
|
Wrong thread, raja sucks share price going nowhere but up
|
# ? Nov 8, 2017 06:27 |
|
FaustianQ posted:Kyle Bennett also said there was some friction between Lisa and Raja IIRC, seems Raja lost, but what the fight was about remains to be seen based on Lisa's moves. If Ryzen is an indication, Lisa wants to target the high end first, both in consumer and commercial, and let everything else trickle down from there. I think this makes some sense if we assume the reason Raja is leaving is Vega did not deliver on that target, while Raja wanted to go back to R700 and Cypress bang for buck days. I think the decision to go with HBM2 runs counter to that, unless you think Raja was naive enough to believe that HBM2 was viable in a midrange product and potentially in downmarket as the tech matured/production ramped. I do think AMD bet super hard on that... but they were moving on it as a high-end product. If Samsung had delivered 1.5 GT/s HBM2... if the whole primitive shader/DSBR gimmick had worked (whatever that ends up being in terms of driver/hardware fixes)... they probably would have smoked the 1080 Ti. This is a story where the design fell apart at every seam and is delivering 30-50% of what it was supposed to. They went flagship-first, they bet hard and had no plan B (in terms of overbuilt/"wasted" die), and they faceplanted. At the end of the day some of the stuff like "do we spend an extra 15mm^2 on spare memory controller" comes down to leadership calls. Somebody gets to fall on the sword when the performance is 1/3 of its design spec when your competitor is exceeding theirs. I'm sure they were super hard up for devs... and once you are behind you are behind, it actually temporarily saps your productivity to read even a few newbies in let alone increase your team by a factor of 10 or whatever. And Raja was kinda hosed given the budget. But he's the guy who's got to say "I need more". And... it probably doesn't speak well of his leadership overall that he wasn't able to pull it together even given an extra 11 months. Paul MaudDib fucked around with this message at 06:31 on Nov 8, 2017 |
# ? Nov 8, 2017 06:29 |
|
You know you hosed up when.
|
# ? Nov 8, 2017 14:20 |
|
I mean, is it really loving up when you leave AMD to go to Intel? Raja's experience will help Intel with Nervana _at least_. Semiaccurate weighs in on AMD + Intel: https://semiaccurate.com/2017/11/06/intel-gpus-officially-dont-work-amd-rescue/
|
# ? Nov 8, 2017 20:43 |
|
SwissArmyDruid posted:I mean, is it really loving up when you leave AMD to go to Intel? Raja's experience will help Intel with Nervana _at least_. "That brings us to the most interesting bit, what OEMs have been telling SemiAccurate that about Intel’s pitch for Kaby-G." That is the last paragraph, but I'm not even sure it qualifies as a sentence or a complete thought...
|
# ? Nov 8, 2017 21:32 |
|
Because that's where the paywall cut off the rest of the article.repiv posted:Have they given any hints as to the interconnect they're using? Is it just an on-package PCIe connection or something more elaborate? Finally found the question to reply to: That semiaccurate article says PCIe from CPU to GPU, EMIB from GPU to HBM. SwissArmyDruid fucked around with this message at 21:38 on Nov 8, 2017 |
# ? Nov 8, 2017 21:33 |
|
SwissArmyDruid posted:Because that's where the paywall cut off the rest of the article. Yeah but the second "that" isn't even correct grammar. They expect me to pay for that lousy editing?
|
# ? Nov 8, 2017 21:48 |
|
So Intel just got a license to use Polaris and poached Raja? Kyle Bennett said Raja was really pushing to sell RTG to Intel... which I guess is starting to sound a little more plausible right now.
|
# ? Nov 8, 2017 21:51 |
|
Paul MaudDib posted:I think the decision to go with HBM2 runs counter to that, unless you think Raja was naive enough to believe that HBM2 was viable in a midrange product and potentially in downmarket as the tech matured/production ramped. I do think AMD bet super hard on that... but they were moving on it as a high-end product. If Samsung had delivered 1.5 GT/s HBM2... if the whole primitive shader/DSBR gimmick had worked (whatever that ends up being in terms of driver/hardware fixes)... they probably would have smoked the 1080 Ti. This is a story where the design fell apart at every seam and is delivering 30-50% of what it was supposed to. They went flagship-first, they bet hard and had no plan B (in terms of overbuilt/"wasted" die), and they faceplanted. At the end of the day some of the stuff like "do we spend an extra 15mm^2 on spare memory controller" comes down to leadership calls. I guess? Kyle's article on the whole thing seems to paint a picture of Raja being this rogue who wanted ATI back and free reign to make his own design decisions. He also spoke of backwards thinking on Koduri's part as well, maybe in a nostalgia sense but also implying Koduri saw the CPU side as some albatross that had no vision, hopefully cutting the tether and sailing away with the GPU division and all its IP. Kind of indicates to me that he was sure of and/or satisfied in the design of Polaris and Vega, so he own their failure as products, and indicates to me that all of the GPU divisions decisions since Fiji have been his. So Either Greenland was some spectacular galactic dumpster fire that it took them an additional 11 months to launch or Raja put priority on trying to recapture the old HD 4870 magic. It all kind of backfired though; Raja is gone and Lisa has control of all of RTG now. Yeah, he's jumping ship to Intel like he threatened years ago, but with uh, what exactly? Intel hasn't gotten a single bit of IP, and Intel has no plans to release dGPUs. So what he's going to twiddle his thumbs on loving KL iterations? This isn't some prelude to RTG being bought by Intel because that is just not happening, that's suicide for AMD and they know it. Paul MaudDib posted:So Intel just got a license to use Polaris and poached Raja? No license at all, rather Intel is taking orders for custom Polaris chips (they're not Vega apparently, but a fork just like the XBOneX). That's it, no transfer of IP at all. EmpyreanFlux fucked around with this message at 22:17 on Nov 8, 2017 |
# ? Nov 8, 2017 22:15 |
|
FaustianQ posted:It all kind of backfired though; Raja is gone and Lisa has control of all of RTG now. Yeah, he's jumping ship to Intel like he threatened years ago, but with uh, what exactly? Intel hasn't gotten a single bit of IP, and Intel has no plans to release dGPUs. So what he's going to twiddle his thumbs on loving KL iterations? This isn't some prelude to RTG being bought by Intel because that is just not happening, that's suicide for AMD and they know it. You just missed where I said that Intel is poaching Raja most likely to put him to work on Nervana. You know, Xeon Phi, which are basically GPGPU compute cards in all but name? That's where Intel is probably going to put him to work fighting against Nvidia's compute monopoly SwissArmyDruid fucked around with this message at 22:57 on Nov 8, 2017 |
# ? Nov 8, 2017 22:22 |
|
I think, for Raja, this is just a situation where all the pieces have fallen into place for him to keep working and actually continue upwards, regardless of his actual performance at (ostensibly) running a PC gaming graphics brand. He's got the HBM, compute and integration know-how which Intel lacks - and in some sub-areas straight up desperately lack.
|
# ? Nov 8, 2017 22:33 |
|
This whole thing is some bizarre Game Of Thrones poo poo. So is this all basically a grudge from the ATI days? If I have the chain of events right...
Like... what? My brain is full of gently caress So has he been deliberately halfassing it/sandbagging, and/or sabotaging AMD from the inside? Does making a very strange deal with Intel and then joining Intel really not raise any questions of ethics/fiduciary responsibility? I kinda have a hard time believing he really had AMD's best interests at heart, one potential interpretation here is that he has been wrecking ever since he got back... I guess it's no wonder he got fired-ish then, Su has to be fuming mad about this whole thing. Paul MaudDib fucked around with this message at 22:54 on Nov 8, 2017 |
# ? Nov 8, 2017 22:38 |
|
cant wait for larrabee 2: koduri boogaloo
|
# ? Nov 8, 2017 22:52 |
|
Also, Raja has already been replaced by Scott Herkelman apparently. I have no idea who the gently caress that is, but Kyle Bennett says he's an excellent replacement, and well Kyle's crystal ball seems to loving work. EDIT: I don't think there was any kind of sabotage on Koduri's part. It'd make him extremely poisonous to hire, and I don't think Intel would want that kind of liability. EmpyreanFlux fucked around with this message at 00:28 on Nov 9, 2017 |
# ? Nov 8, 2017 22:53 |
|
Kyle Bennett went to the Himalayas to find enlightment after he transformed into a salt statue for Netburst
|
# ? Nov 9, 2017 00:06 |
|
Intel to Develop Discrete GPUs, Hires Raja Koduri as Chief Architectrepiv posted:cant wait for larrabee 2: koduri boogaloo dreams do come true
|
# ? Nov 9, 2017 00:30 |
|
repiv posted:Intel to Develop Discrete GPUs, Hires Raja Koduri as Chief Architect This has been the most buck wild week of GPU news
|
# ? Nov 9, 2017 00:34 |
|
This works in any situation, lol
|
# ? Nov 9, 2017 00:45 |
|
What the everloving gently caress is this timeline.
|
# ? Nov 9, 2017 00:45 |
|
repiv posted:Intel to Develop Discrete GPUs, Hires Raja Koduri as Chief Architect What the heck
|
# ? Nov 9, 2017 00:49 |
|
I don't know if I want a GPU with blue LEDs... they're really harsh at night time. Does this mean they might be able to put out their first discrete GPU products at 10nm?
|
# ? Nov 9, 2017 00:53 |
|
FaustianQ posted:It all kind of backfired though; Raja is gone and Lisa has control of all of RTG now. Yeah, he's jumping ship to Intel like he threatened years ago, but with uh, what exactly? Intel hasn't gotten a single bit of IP, and Intel has no plans to release dGPUs. So what he's going to twiddle his thumbs on loving KL iterations? This isn't some prelude to RTG being bought by Intel because that is just not happening, that's suicide for AMD and they know it. It's me guys, I'm the one who used up the monkeys paw. FaustianQ posted:Look man if I am stuck in this dumpster fire reality made only of the wrong choices on the probability tree I demand it at least entertain me.
|
# ? Nov 9, 2017 00:53 |
|
Zero VGS posted:I don't know if I want a GPU with blue LEDs... they're really harsh at night time. All LEDS are terrible and shouldn't be allowed on computers.
|
# ? Nov 9, 2017 00:57 |
|
Raja got a big pay day
|
# ? Nov 9, 2017 00:59 |
|
|
# ? Jun 5, 2024 08:29 |
|
I’m so confused
|
# ? Nov 9, 2017 01:10 |