Was AMD holding back on their fusion line?

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
Honestly they're CPU's are somewhat underwhelming, let's not get into the debate of VFM for now, but their (i)GPU's are top notch & this is where I think (& hope) that HSA/HUMA plus OpenCL will put them in a decent position provided the software developers do their job properly.

The problem with hsa/huma is software support, not hardware. And AMD software support is fourth class in most cases. What solution would you adopt, something that Intel is backing up, providing tools and support, or something AMD would be undertaking these tasks?

There is a reason for the price gap between Nvidia and AMD professional cards, and if AMD can't even take on Nvidia in a niche market, give up your hopes of them being able to take on Intel in the entire market.
 

norseamd

Lifer
Dec 13, 2013
13,990
180
106
The problem with hsa/huma is software support, not hardware. And AMD software support is fourth class in most cases. What solution would you adopt, something that Intel is backing up, providing tools and support, or something AMD would be undertaking these tasks? There is a reason for the price gap between Nvidia and AMD professional cards, and if AMD can't even take on Nvidia in a niche market, give up your hopes of them being able to take on Intel in the entire market.

why is it that with any company that the hardware seems ten steps ahead of the software
 

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
why is it that with any company that the hardware seems ten steps ahead of the software

Because not everything the hardware companies develop are good ideas. Itanium, Bulldozer, Denver and others are example of this, and for a software company betting the farm in one wrong idea is a life or death question, so they wait.

As we speak now multiple paradigms compete for supremacy on the market, AMD HSA is just of them, along with CUDA, MIC and whatever arm and mips have in their wings.
 

norseamd

Lifer
Dec 13, 2013
13,990
180
106
As we speak now multiple paradigms compete for supremacy on the market, AMD HSA is just of them, along with CUDA, MIC and whatever arm and mips have in their wings.

cuda and open cl are not exclusive of heterogenous computing

thing is cuda is propeitary and open cl and can work for everyone so we should go with that

open cl can be used on apus, gpus, and compute cards
 

R0H1T

Platinum Member
Jan 12, 2013
2,582
164
106
The problem with hsa/huma is software support, not hardware. And AMD software support is fourth class in most cases. What solution would you adopt, something that Intel is backing up, providing tools and support, or something AMD would be undertaking these tasks?
The same argument can be made about Intel's graphics drivers & Quick Sync support but it's never mentioned in the same breath because assumedly no one games on Intel's iGPU, even though Steam says otherwise, or that Quick Sync is the magic elixir that's better than OpenCL for hardware accelerated video tasks.
There is a reason for the price gap between Nvidia and AMD professional cards, and if AMD can't even take on Nvidia in a niche market, give up your hopes of them being able to take on Intel in the entire market.
I don't follow the professional graphics arena much but from what I know CUDA & Nvidia's ecosystem based on it is excellent, however AMD was less than forthcoming with its professional GPU offerings in the recent past but they may have turned over a new leaf with their latest Hawaii based products, only time will tell whether they can get critical mass behind them once again & with the same innovations they're pushing in the general computing space.
 
  • Like
Reactions: marees

mrmt

Diamond Member
Aug 18, 2012
3,974
0
76
cuda and open cl are not exclusive of heterogenous computing

thing is cuda is propeitary and open cl and can work for everyone so we should go with that

open cl can be used on apus, gpus, and compute cards

Whatever. I just want to point out that it doesn't matter if something gives X extra performance, if they put a disproportionate burden in the development process of the software house, and this is currently the crux of the issue for GPGPU: Too much work for the benefits you get.

Software support and development tools aim to cut this "too much work" to something more palatable to the software companies, but they must be developed by the hardware companies themselves. And this is where things get dirty for AMD. They don't have the experience and the financial muscle to develop the tools.

The same argument can be made about Intel's graphics drivers & Quick Sync support but it's never mentioned in the same breath because assumedly no one games on Intel's iGPU, even though Steam says otherwise, or that Quick Sync is the magic elixir that's better than OpenCL for hardware accelerated video tasks.

Intel is improving, and they have the material conditions to improve faster than AMD or Nvidia in this arena. Financial muscle does matter here. Just look at how Intel iGPU was talked in 2010 and look at how people talk about it now. Even the AMD resellers are having a hard time tapdancing around the issue.

I don't follow the professional graphics arena much but from what I know CUDA & Nvidia's ecosystem based on it is excellent, however AMD was less than forthcoming with its professional GPU offerings in the recent past but they may have turned over a new leaf with their latest Hawaii based products, only time will tell whether they can get critical mass behind them once again & with the same innovations they're pushing in the general computing space.

Why do you think they turned a page here? Because they said so? Because Semiaccurate said so? What really changed with Hawaii?

When you buy a professional card, it's not the hardware you are buying, but the hardware, the drivers, access to the ecosystem and, much more important, the software support. Hawaii just changes one link in the chain, but we have the others that are very weak: It's the same crappy driver, it's the same software support team that leaves bugs open without answer, it's the same crappy driver team that leaves bugs open for MONTHS, it's the smaller app ecosystem.

And don't even start talking about OpenCL, because Nvidia cards ALSO support OpenCL, and they did that before AMD. Don't forget the drivers, because Nvidia implementation of OpenCL is less buggy than AMD.

This is a relatively small market, with a lot less players to interact with and with a lot less demand of OPEX than, let's say, HSA and HUMA in the consumer and enterprise market. If they can't get something that small right, what are the chances of AMD changing the entire programming paradigm of the entire x86 market?
 

Insert_Nickname

Diamond Member
May 6, 2012
4,971
1,695
136
The same argument can be made about Intel's graphics drivers & Quick Sync support but it's never mentioned in the same breath because assumedly no one games on Intel's iGPU, even though Steam says otherwise, or that Quick Sync is the magic elixir that's better than OpenCL for hardware accelerated video tasks.

I can say that Intel's drivers have some compatibility problems with older titles. To put it mildly. They're getting better of course, but Intel does seem to drop driver development for their IGPs rather quickly.

Compare that to almost 8 years of mainstream support on Nvidia cards (G80/8800-series).
 

marees

Golden Member
Apr 28, 2024
1,467
2,057
96
The endgame for AMD - ATI acquisition
No more stand alone gaming gpu chips starting RDNA 5 in 2027 (2 decades after ATI acquisition)

full RDNA 5 lineup
  1. AT0 = 10090xt is primarily for AI
    1. (>5090) — 144 CU to 192? CU (384 bit gddr7 to 512 bit gddr7)
  2. AT1 = 10080xt is scrapped
  3. AT2 = 10070xt is (gddr7 192bit) shared with xbox
    1. (~5080) — 72? CU (192 bit gddr7 — there could be cut down versions
  4. AT3 = 10060xt is (lpddr6 384bit) shared with medusa halo
    1. (~5070?) — 48 CU (384bit lpddr6)
  5. AT4 = 10050xt is (lpddr6 128bit) shared with medusa premium
    1. (~3060 12gb +) — 24 CU (128bit lpddr6)
 

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
31,876
32,043
146
There is a reason for the price gap between Nvidia and AMD professional cards, and if AMD can't even take on Nvidia in a niche market, give up your hopes of them being able to take on Intel in the entire market.
Greetings from the future, where this aged like milk. Less than 3 years after this seemingly easy to make prediction so many shared, Zen launched. The framing of Nvidia being easier to compete against aged even worse.
Intel is improving, and they have the material conditions to improve faster than AMD or Nvidia in this arena. Financial muscle does matter here. Just look at how Intel iGPU was talked in 2010 and look at how people talk about it now. Even the AMD resellers are having a hard time tapdancing around the issue.
Future here again, and man are you going to 💩 bricks when you see how the bold played out.
If they can't get something that small right, what are the chances of AMD changing the entire programming paradigm of the entire x86 market?
I surmise this dude didn't ride the AMD stock from a couple dollars a share to where it's at now. :p
 

Farfle

Member
Jan 10, 2006
85
2
71
Greetings from the future, where this aged like milk. Less than 3 years after this seemingly easy to make prediction so many shared, Zen launched. The framing of Nvidia being easier to compete against aged even worse.

Future here again, and man are you going to 💩 bricks when you see how the bold played out.

I surmise this dude didn't ride the AMD stock from a couple dollars a share to where it's at now. :p
My bold prediction to be completely shredded apart in another 10 years hence:
  1. An external influence will be necessary to bring down the titan NVIDIA and its choking hold on the AI industry. Think Standard Oil in early 1900s after the Oil boom...
  2. ARM architecture is the future. Apple hast showeth the way. But the PC will live on, whether it's OS is Windows or some Linux variant (Chrome OS, Steam OS, etc). So, who takes Intel and AMD's current place in providing those PC chips? Qualcomm? Mediatek? Or some other unmentioned entity? A deluge of ARM CPU players is possible!
Good luck AMD! You're certainly in a better financial position now than back in 2014, but the challenges ahead are possibly even more severe! At least back then you didn't have (quite yet) an existential crisis to deal with for the future of your company's primary product. And Intel ... well, I feel its only hope lay with Intel Fab, as demonstrated recently by the US government's acquisition of 10% stake in the company. I don't see how a spin-off of - *not* their Fab, but rather their products division - is not already just about completely written in the books.
 
  • Like
Reactions: DAPUNISHER

DAPUNISHER

Super Moderator CPU Forum Mod and Elite Member
Super Moderator
Aug 22, 2001
31,876
32,043
146
My bold prediction to be completely shredded apart in another 10 years hence:
The boldest prediction you made, is that the forums will be here in a decade.

crying-laughing.gif
 
  • Haha
Reactions: gdansk and Farfle