eDRAM and Optane. Legendary Intel advances but they dropped the ball in an even more legendarily bad way on them.
eDRAM and Optane. Legendary Intel advances but they dropped the ball in an even more legendarily bad way on them.Adding huge amounts of cache was niche in the past, like what Intel did with edram in the past, now it's normal with V-Cache.
AMD doesn't have the $$$ and know-how to build and support gaming GPUs on the same level as nV. nV has been the better & the advanced tech for more than 11 years.This is more proof that they think gaming is unimportant to their bottom line or they have little confidence in their teams to change the status quo.
But let's be realistic. AMD could do a LOT better. They have both CPU and GPU. How come they can't figure out a way to create more synergy between the two and make Nvidia/Intel combo look unattractive? Their constant focus on enterprise sucks. This is more proof that they think gaming is unimportant to their bottom line or they have little confidence in their teams to change the status quo. Whether this is the fault of the management (hiring 2nd rate employees coz the best end up at Nvidia or Apple or they don't want to pay enough to attract top talent) or the employees themselves (failing at execution due to whatever reasons) is something we may never know.
On the bright side, some of those employees could get picked up by Qualcomm? They sure need help
Wasn't HSA part of the reason why they did the shared floating point with Bulldozer? From what I remember they hoped that most floating point operations would be moved to the GPU. Interesting to see that turned on it's head. AMD's CPUs have the biggest baddest floating point performance, and their support for GPU compute (in client GPUs at least) is bad.Uhm, "The future is Fusion?"
They tried with HSA.
After years of switching between SDKs they settled on ROCm and they have been pushing it for 8 years and ... well, the UX still lags horribly behind CUDA. The CPU support is way easier.Wasn't HSA part of the reason why they did the shared floating point with Bulldozer? From what I remember they hoped that most floating point operations would be moved to the GPU. Interesting to see that turned on it's head. AMD's CPUs have the biggest baddest floating point performance, and their support for GPU compute (in client GPUs at least) is bad.
Novel Idea, but why? A couple hundred unit extra sales? Not even worth a meeting rather on a head.They need to have a spokesperson here. By interacting with us, he/she could get at least a "little" idea of why people like AMD and what more they expect from AMD to keep liking them even more. If they can wrap their heads around why some people get disappointed about their launches or why certain avid users are deciding to skip a generation, it will help them that much more to create a more compelling product that will be much harder to resist for the enthusiast. And that trickles down to normal users because most enthusiasts love to express and spread their excitement when using a product and recommend it to anyone and everyone in their circle. Some may even be good writers and do a very public blogpost, giving AMD free publicity.
Totally agree with this.No, people here are AMD's worst customers. Loudest critics. Biggest value seekers. Declining market size.
Frequently, large companies use a good excuse to thin the herd and getting rid of some acquired fat.Someone suggested it is mostly due to excess employees picked up from acquired companies.
It would also make sense if AMD were to purchase Intel .MLID claims they are cutting from gaming and semi division, which was my guess yesterday. Also that it is to make AMD look more attractive to investors by having high margins. It's probably a little of all of it. I hate all of these tech layoffs being announced during the holiday season. Not cool. The employees deserve better.
Yeah they do. But ROI is not there.AMD doesn't have the $$$ and know-how to build and support gaming GPUs on the same level as nV
that's cope.There is only one way that AMD could hope to really compete with NVIDIA with good success, and that's a comprehensive cross licensing and development partnership with Intel for GPUs. It would take Intel agreeing to provide AMD low cost foundry services, and AMD to assist Intel with it's own GPU efforts. It would allow them to attack NVIDIA in it's margins.
It makes sense considering that they will only have UDNA going forward. They don't need two teams creating CDNA and RDNA. Before they announced the change, I wondered if they would use CDNA consumer chips because of AI. I bet Raja pushed to have two teams since he complained about doing both with Vega before he left.MLID claims they are cutting from gaming and semi division, which was my guess yesterday. Also that it is to make AMD look more attractive to investors by having high margins. It's probably a little of all of it. I hate all of these tech layoffs being announced during the holiday season. Not cool. The employees deserve better.
Cope you say? Where is AMD supposed to get this magic performance from? NVIDIA out spends AMD by twice on research on an annual basis with FAR more spent on GPUs and related tech than AMD. AMD caught up with Intel because they were stumbling badly. NVIDIA is hitting on all cylinders right now.Yeah they do. But ROI is not there.
that's cope.
Microarchitecture.Where is AMD supposed to get this magic performance from?
How is this relevant.NVIDIA out spends AMD by twice on research on an annual basis with FAR more spent on GPUs and related tech than AMD
They in fact did not "catch up" to Intel. they smashed 'em.AMD caught up with Intel because they were stumbling badly.
But they did, RDNA2 did well in laptops. Then the oopsie happened.Right now, AMD can't even get OEMs to use their dGPUs in laptops at ANY price.
Must be location dependent since those basically didn't exist at least here in Finland.But they did, RDNA2 did well in laptops. Then the oopsie happened.
Well yeah, need more unit and revenue share to get to more inhospitable places.Must be location dependent since those basically didn't exist at least here in Finland.
"Well" is a drastic overstatement...Microarchitecture.
How is this relevant.
Intel was spending more on fab R&D than $TSM yet fell hopelessly behind.
They in fact did not "catch up" to Intel. they smashed 'em.
But they did, RDNA2 did well in laptops. Then the oopsie happened.
JPR unit share is a very real thing."Well" is a drastic overstatement...
They did non-zero, which was a massive percentage increase. You saw a tiny handful of models for anything over the 6500. I know two people who purchased 6800 models, and both went back due to motherboard failures during the first 6 months.
Right now? yeah. Don't let your execution slip is the lesson.AMD has near zero OEM cooperation when it comes to dGPUs.
it's a Very Expensive Thing.but it remains to be seen how well it can be priced.
Sure they can do quad channel. But just for a low end dGPU performance that can't be upgraded? You want people to buy a $300 board for a $150 dGPU equivalent at best? And you can't even get some money back by reselling? And you think the CPU is going to be cheap somehow? It's a $300-400 CPU + $300 board versus any CPU + $100 board.If AMD really cared about total domination, they would introduce quad channel RAM to consumer desktops, give hex or octa channel to TR platform and reserve 12 channels for their servers. Being complacent while ahead never turns out well, as can be seen with Intel. Mobos are already $300 or more. Respectfully, please don't give me any crap excuse that they can't do quad channel in that price range.
This doesn't help with the "Chungus APUs are niche" comment by @adroc_thurston.Adding huge amounts of cache was niche in the past, like what Intel did with edram in the past, now it's normal with V-Cache. Quad channel is niche because vendors want it to be niche so you pay much more for it.
A big iGPU will only be lower power if both of them are superior to competition. If the GPU part is behind, then you end up same as the dGPU.I also think big APUs are really niche. CPU + dGPU will be cheaper but consume a bit more power.
Apple gets away with it, somehow. Plus if people are going to continue to disrespect AMD dGPUs to the point that they can't crack 10%+ marketshare, it may be the only way to expand their graphics business would be to refocus on iGPUs in some fashion.I'm with him that big iGPUs make little to no sense.
Apple gets away with it because they have the niche of people who work on videos and not much else. The encoding/decoding engines and big GPU do work because Apple has their own video editing software and the market is so Apple-focused that other software companies also support it.Apple gets away with it, somehow. Plus if people are going to continue to disrespect AMD dGPUs to the point that they can't crack 10%+ marketshare, it may be the only way to expand their graphics business is to refocus on iGPUs in some fashion.
Apple also does things the rest of the world dreams of. In both Smartphones and PCs they take majority share for premium devices. AMD can't do that.Apple gets away with it, somehow. Plus if people are going to continue to disrespect AMD dGPUs to the point that they can't crack 10%+ marketshare, it may be the only way to expand their graphics business would be to refocus on iGPUs in some fashion.
Apple gets away with it because they have the niche of people who work on videos and not much else. The encoding/decoding engines and big GPU do work because Apple has their own video editing software and the market is so Apple-focused that other software companies also support it.
AMD's big iGPU is near useless because the software won't work with it properly. And it won't be a good gaming GPU either. Strix Halo, even if it's a technically good product, will sell worse than even Navi 31/32. But at least they can justify the R&D as it should lead to their future Zen.
People are buying MI300 because it's a good accelerator. MI350/X looks even better.Oh and pivoting to AI when the only people buying MI300 are those who want to hedge against Nvidia is a bad bet. As soon as the AI market turns down it'll be near zero sales, no need to worry about Nvidia allocation anymore. AMD needs to diversify its revenue streams. But I guess you can hardly call Navi 31/32 a revenue stream...
APU BoM is lower due to unified memory.CPU + dGPU will be cheaper but consume a bit more power.