AMD reveals its CPU & GPU roadmap for next five years

erek

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,875
Next year, AMD plans to bring two new CPUs to the market. One of these will be based on the ARMv8 architecture while the other will be based on x86 AMD64 architecture. Both of these chipsets will be targeted at the server segment, so not much new for the consumer market from AMD next year. However, AMD did reveal that the ARMv8 based K12 as well as the x86 Zen cores will offer support for “many threads”. This points to simultaneous multi-threading on the new chips instead of clustered multi-threading that AMD currently uses on its Bulldozer processor line.

http://vr-zone.com/articles/amd-reveals-cpu-gpu-roadmap-next-five-years/89825.html
 
Servers are what bring desktop in the market. And that 300w APU is CPU + GPU so look at today's TDP and add em together.
 
300w apu, are you an idiot. That could be a savior, I know I would buy a apu if it performed like a midrange card.
 
Well a GTX 970 is 164 watts, i7-4790K is 88 watts, so 252 watts, I'd like to see them put that much performance in one chip.
 
OmgitsSexyChase said:
300w apu, are you an idiot. That could be a savior, I know I would buy a apu if it performed like a midrange card.
Click to expand...

This without the personal attacks. So tempting to use for all the relatives. Good APU + AIO + SSD annnd you're set.
 
I do like the idea of mid-high end gaming APUs, I'd rather put in a tower heatsink, also you could make some tiny motherboards if you don't need any PCI-E slot for the GPU.
 
PcZac said:
I do like the idea of mid-high end gaming APUs, I'd rather put in a tower heatsink, also you could make some tiny motherboards if you don't need any PCI-E slot for the GPU.
Click to expand...

Sounds like that would be good for something like a Steam machine where it's just a console-sized thing.
 
Any APU that large makes no sense to me, if the majority of the power is in the GPU it's going to be cut off at the knees from a bandwidth standpoint.
 
PcZac said:
Well a GTX 970 is 164 watts, i7-4790K is 88 watts, so 252 watts, I'd like to see them put that much performance in one chip.
Click to expand...

I agree. For me I'm all about performance first. Once the performance their then try to improve efficiency over time.
 
Kor said:
Any APU that large makes no sense to me, if the majority of the power is in the GPU it's going to be cut off at the knees from a bandwidth standpoint.
Click to expand...


Makes sense under certain server/HPTC loads and maybe the low-end PC. Integrated GPU's are only going to continue to raise the lower bar of performance for the foreseeable future. Ain't gonna replace a 10" card in a tiny die.
 
if they could squeeze 290 level of performance into an apu by then i think it would sell well, especially if it had 8 cores as well.........some people only see the negative in things if its made by amd....there going out of business every year lol:D
 
Liger88 said:
Makes sense under certain server/HPTC loads and maybe the low-end PC. Integrated GPU's are only going to continue to raise the lower bar of performance for the foreseeable future. Ain't gonna replace a 10" card in a tiny die.
Click to expand...

It's fine to make that argument however I still question the merit of the design. A 300W APU says to me that 150 or more is dedicated to a GCN core, assuming we are in 16nm FinFet territory here were talking 280x/290 performance being hobbled by at best dual channel DDR4 (I can't imagine AMD is going quad channel for the APU market). The 25gb/s that sort of configuration provides isn't anywhere near enough to feed something that large no matter how much cache you bolt on.
 
primetime said:
if they could squeeze 290 level of performance into an apu by then i think it would sell well, especially if it had 8 cores as well.........some people only see the negative in things if its made by amd....there going out of business every year lol:D
Click to expand...

They basically are going out of business but like most companies it will be a slow death.
 
Hmmm...

I don't really get what AMD are up to. They haven't been able to compete on the high-end CPU field for ages now (since "Core" series Intels??). As for GPUs, that's kind of always gone back and forth but market share shows that nvidia are still way out in front.

Who knows what they're doing to be honest.

I think it would have made more sense if they took one of the console SoCs and turned that into a consumer SKU somehow.

Unified memory may have been the stumbling block here .
 
Kor said:
Any APU that large makes no sense to me, if the majority of the power is in the GPU it's going to be cut off at the knees from a bandwidth standpoint.
Click to expand...

High Bandwidth Memory ;)

An APU with HBM could see bandwidth of 640GB/s
 
while the 300w you also have efficient regulation of power so it wont be 300w in actual use.
 
Kor said:
Any APU that large makes no sense to me, if the majority of the power is in the GPU it's going to be cut off at the knees from a bandwidth standpoint.
Click to expand...

Kor said:
It's fine to make that argument however I still question the merit of the design. A 300W APU says to me that 150 or more is dedicated to a GCN core, assuming we are in 16nm FinFet territory here were talking 280x/290 performance being hobbled by at best dual channel DDR4 (I can't imagine AMD is going quad channel for the APU market). The 25gb/s that sort of configuration provides isn't anywhere near enough to feed something that large no matter how much cache you bolt on.
Click to expand...

OK TRY AND PAY ATTENTION TO THE NEXT QUOTED POST:

pandora's box said:
High Bandwidth Memory ;)

An APU with HBM could see bandwidth of 640GB/s
Click to expand...

Seriously guys, try reading a little. HBM blows every form of DDR GDDR out of the park in terms of bandwidth. And AMD obviously wants HBM not so much for their GPUs but APUS hence the acquisition of ATI and the years spent in development of HBM.
 
durquavian said:
OK TRY AND PAY ATTENTION TO THE NEXT QUOTED POST:



Seriously guys, try reading a little. HBM blows every form of DDR GDDR out of the park in terms of bandwidth. And AMD obviously wants HBM not so much for their GPUs but APUS hence the acquisition of ATI and the years spent in development of HBM.
Click to expand...

I have trouble envisioning how this will happen. On die? Buy a mobo with a set amount?
 
there are 2 forms.

3D HBM, which is on die.

and 2.5D which is on an interposer. This is the one AMD is using.

GPUS run way to hot to try to stack on-die, they would literally melt.
 
KazeoHin said:
Fermi.
Click to expand...

So in a few years from now, AMD will be where NVIDIA was 5 years ago. Sounds like a great plan.

APUs are meant for consoles and low end laptops. No one is going to want an APU with a video card sized heat sink and fan attached to it. :rolleyes:
 
PRIME1 said:
So in a few years from now, AMD will be where NVIDIA was 5 years ago. Sounds like a great plan.

APUs are meant for consoles and low end laptops. No one is going to want an APU with a video card sized heat sink and fan attached to it. :rolleyes:
Click to expand...

Nobody wanted an Intel chip with a GPU attached to it when it first came out. The general consensus was why are we paying for that trash to be attached to it. See how that worked out for Intel.

Maybe the companies that invent these technologies are smarter than we think. Look at how many companies are signed up for this technology in the HSA Foundation. I'd like to think that a few of the names on the front page are relevant in today's electronics business.
 
cageymaru said:
Nobody wanted an Intel chip with a GPU attached to it when it first came out. The general consensus was why are we paying for that trash to be attached to it. See how that worked out for Intel.
Click to expand...

I disagree. The enthusiast market (>2% of all PC sales) thought it was a terrible idea. Most PC's are pre built and 98% of them contain integrated GPU's.

Wed hardware enthusiasts are the minority. Intel knew exactly where the bread and butter was and is.
 
Most people couldn't tell you if they have an integrated or a add in gpu. Plus they don't care as long as it plays video and flash games.
 
I think a high end APU that can power most games would be great for AMD, especially if they can maintain their hold on consoles and make more inroads in consumer devices powered by an APU that fulfills the needs and wants of 98% of the market. That might mean fewer innovations in high end, enthusiast level stuff in the short term, but that may allow for a higher level of stability and potential for growth in the high end in the long term.
 
From what AMD has been speaking of for years, I think they want to push a new kind of efficiency. There are many that would argue, with a great deal of merit, x86 is terribly inefficient and simply by grunt we get thru. Much like DX and hence the move to DX12. And like many have mentioned, a simple look at the HSA foundation main page you see a strong desire to move away from the way we have been computing. APUs are the embodiment of this dream. The ability to split code to the necessary CPU or GPU to best and most efficiently do the work.

Look at the tests with Adobe( I think) using a 3960/3930(x or K) and the 7850K. Using the traditional code path, obviously the Intel won out large. But utilizing HSA, the 7850K pulled ahead. A $200 APU beating out the $1000 CPU. But what we need to take from this is that utilizing a much weaker CPU in the APU we can get a huge amount of work done. And this is only the beginning.
 
ocellaris said:
So AMD is dead in the performance desktop market? RIP :(
Click to expand...

Yep. We've known it for a while now but I guess this makes it definite. Sucks too. Guess I'll just hang on to my rig as long as I can then throw whatever Intel has in when I have to upgrade. :(

I guess the enthusiast segment is totally dependant on Intel now.
 
durquavian said:
OK TRY AND PAY ATTENTION TO THE NEXT QUOTED POST:



Seriously guys, try reading a little. HBM blows every form of DDR GDDR out of the park in terms of bandwidth. And AMD obviously wants HBM not so much for their GPUs but APUS hence the acquisition of ATI and the years spent in development of HBM.
Click to expand...

HBM will be unreasonably expensive in 2.5D form for some time to come, it's not cost effective for the APU market. Hell running alternate and more expensive memory types has proven to be a bad idea historically. Just ask Intel how well RDRAM worked out for the Willamette Pentium 4.
 
pandora's box said:
High Bandwidth Memory ;)

An APU with HBM could see bandwidth of 640GB/s
Click to expand...

that would have to be on the motherboard and off the core to prevent thermal problems.
Much like the old K3 days with onboard cache.

In this case I could not see anything less than 4 gigs onboard cache of HBM. And primary ram of quad channel DDR4
Even boards priced at $500 I would buy one.
 
As I recall a high performance (high wattage) APU is what AMD's been building toward for years. I'm glad to see it may see daylight.
 
You must log in or register to reply here.
Back
Top