AMD CPU speculation... and expert conjecture

[BeastLeeX]

Correct me if I'm wrong, but hasn't AMD struck a deal to supply the next gen consoles with AMD CPUs and GPUs?

Your not wrong, their cpu and gpu are supposed to be in both of the next gen consoles, and the gpu is already in the Wii U.

 

[gamerk316]

Go back and read what I wrote. I was referring to an update to the windows scheduler for programmers where windows wont put 2 locked threads on the same core if that ever happens. Obviously it wont happen because programmers don't want the ability to hand schedule anything even if its done in a way that improves performance.

EDIT: Keep it civil, guys...

Ok, fine. Another example then: What happens if the number of programs using locked threads exceeds the number of available CPU cores? Oops, you schema blows up, again.

Seriously, you want to go back to coorporative multithreading, which sucked. It never worked well, was frail, and ate performance. Thats why no OS uses it anymore. Its not the 80's, and we shouldn't code like it is either.

If you honestly think you are smarter then someone who's handled thread logic for over two decades, worked on several million SLOC programs, and has legitimately worked with programs that use hundreds of threads at a time, go ahead. At the end of the day: THERE IS NO PERFORMANCE ADVANTAGE, EVER, TO USING MANUAL THREAD MANAGEMENT.

I can't make it any more clear then that.

 

[gamerk316]

Your not wrong, their cpu and gpu are supposed to be in both of the next gen consoles, and the gpu is already in the Wii U.

I still chuckle that the WiiU GPU is essentially a slightly modified 4670. Can't imagine the margins on continuing to produce that part justify its production.

Seriously, in 5 years, when everyone is on the 10nm node, AMD is going to be stuck continuing to produce a 4000 series chip. Does that sound like a good idea to anyone?

 

[de5_Roy]

"Global Foundries lays out their FinFET plans"
http://semiaccurate.com/2013/02/08/global-foundries-lays-out-their-finfet-plans/
they also sorta admit that they fxed up. i've been blaming them for a while... too little too late, jerks!

AMD to Focus on Radeon HD 7000 “For Quite Some Time”.
AMD Likely to Refresh Desktop Product Line in Q2 2013
http://www.xbitlabs.com/news/graphics/display/20130207235000_AMD_to_Focus_on_Radeon_HD_7000_For_Quite_Some_Time.html
"Rambus to Help GlobalFoundries' Clients to Quickly Transit to 14nm-XM FinFET Tech.
Rambus to Design Silicon-Proven Memory and Interface Blocks for 14nm-XM FinFET Tech"
http://www.xbitlabs.com/news/other/display/20130207234115_Rambus_to_Help_GlobalFoundries_Clients_to_Quickly_Transit_to_14nm_XM_FinFET_Tech.html

 

[blobby91]

I still chuckle that the WiiU GPU is essentially a slightly modified 4670. Can't imagine the margins on continuing to produce that part justify its production.

Seriously, in 5 years, when everyone is on the 10nm node, AMD is going to be stuck continuing to produce a 4000 series chip. Does that sound like a good idea to anyone?

It will probably get a die shrink or two during that time like the CPU's and GPU's have done in consoles for the past two generations.

 

[mayankleoboy1]

All this talk of 22nm, 20nm, 14nmHP, 14nmLP,14nmHLP, 14xm etc etc is making my head spin.
What remains to be seen is the actual products that emerge.

 

[mayankleoboy1]

I still chuckle that the WiiU GPU is essentially a slightly modified 4670. Can't imagine the margins on continuing to produce that part justify its production.

Seriously, in 5 years, when everyone is on the 10nm node, AMD is going to be stuck continuing to produce a 4000 series chip. Does that sound like a good idea to anyone?

by the time 10nm comes, AMD could be making 10nm HD4670 chips, no ?

It would be very interesting to revisit older architectures that were EPIC fails at their original nodes, in a newer node. A Fermi arch at 10nm would be delicious

 

[sarinaide]

Most people will generally accept that the 6800K will deliver 22-27% more performance than the 5800K. I was reading speculation on Kaviri using a HD8660 in dual graphics to produce performance close to a HD7870, considering the chip and GPU will come to $220 that is extremely good performance. 55% stronger iGPU than that on Richland, which I assume is mostly down to metal level changes and a far faster GPU arch and doubling the amount of stream processors. I was thinking of buying a 6800k but will just save money and hold out for the Kaviri parts

 

[-Fran-]

Ok, this is a little off the track, but...

Since now Intel and AMD are using numbering in the thousands and the letter K for some products, let's start using the company name for the sake of context, please. It's getting in my nerves the name orgy they're creating.

And regarding the 28nm node... I think Fabs are desperately trying to close the gap with Intel's, so they're negotiating with all clients to let them switch to 20nm or under faster. That could explain some "long periods" to take a product out. Where "some" are GPUs, hahaha. I could be totally wrong here, but I'll throw it anyway

Cheers!

EDIT: Forgot to ask you sarinaide. Can you edit your post? I need context for the numbers, hahaha.

 

[gamerk316]

by the time 10nm comes, AMD could be making 10nm HD4670 chips, no ?

It would be very interesting to revisit older architectures that were EPIC fails at their original nodes, in a newer node. A Fermi arch at 10nm would be delicious

But the point is, that would require AMD spend a lot of cash to fit that old arch to a new node. So how much money, over the lifetime of this part, will they actually make?

 

[m32]

Most people will generally accept that the 6800K will deliver 22-27% more performance than the 5800K. I was reading speculation on Kaviri using a HD8660 in dual graphics to produce performance close to a HD7870, considering the chip and GPU will come to $220 that is extremely good performance. 55% stronger iGPU than that on Richland, which I assume is mostly down to metal level changes and a far faster GPU arch and doubling the amount of stream processors. I was thinking of buying a 6800k but will just save money and hold out for the Kaviri parts

I hope it has that kind of performance, but I doubt it.

 

[JAYDEEJOHN]

But the point is, that would require AMD spend a lot of cash to fit that old arch to a new node. So how much money, over the lifetime of this part, will they actually make?

They wouldnt have to have it on the cutting edge node, and one shrink, if they did do so, would be all itd need, as older nodes are dirt cheap comparatively

 

[noob2222]

Ok, fine. Another example then: What happens if the number of programs using locked threads exceeds the number of available CPU cores? Oops, you schema blows up, again.

Seriously, you want to go back to coorporative multithreading, which sucked. It never worked well, was frail, and ate performance. Thats why no OS uses it anymore. Its not the 80's, and we shouldn't code like it is either.

If you honestly think you are smarter then someone who's handled thread logic for over two decades, worked on several million SLOC programs, and has legitimately worked with programs that use hundreds of threads at a time, go ahead. At the end of the day: THERE IS NO PERFORMANCE ADVANTAGE, EVER, TO USING MANUAL THREAD MANAGEMENT.

I can't make it any more clear then that.

I was actually referring to hybrid threading.

What I am trying to understand is why you believe so hard that programming in its current state is at its apex and will never advance any further than it is right now.

Do you really believe that nothing will ever come along and change the current situation, or is it just too difficult and expensive to even try?

If thats what you believe, then in theory the HSA foundation will fail. Sorry, but I am a bit more optimistic that things will continue to change.

 

[JAYDEEJOHN]

We will see how the consoles do things, it may give us a glimps as to which way things are heading.
FF HW is always better than a SW solution anyways

 

[jdwii]

Correct me if I'm wrong, but hasn't AMD struck a deal to supply the next gen consoles with AMD CPUs and GPUs?

Amd expects 20% of their income to come from the consoles. This year anyways

 

[BeastLeeX]

Amd expects 20% of their income to come from the consoles. This year anyways

Sounds good to keep AMD afloat, and won't console makers expect AMD to survive another 7 years to keep creating chips?

 

[iceclock]

i was referring to performance desktop chips, i know theres a future for arms and low powered mobile chips, but i meant for mainstream performance doesnt look good.

 

[JAYDEEJOHN]

Its a shrinking market for sure, but this can mean also different ways to use them as well

 

[esrever]

Sounds good to keep AMD afloat, and won't console makers expect AMD to survive another 7 years to keep creating chips?

Yes. The way the x86 license works. If AMD does not live, they can't supply the chip and nobody can fab those chips and sell them. Both microsoft and sony expects AMD to pull through. If Microsoft has to, they will probably loan AMD money just to keep making xbox chips.

AMD is expected to make a turn around in a couple of years maybe with some more downsizing. The worst of their financial troubles looks to be over.

 

[amdfangirl]

I'm still wondering if a court can force Intel to give out more x86 licenses.

Maybe if Intel jacks up the prices after the demise of AMD, nobody buys x86 anymore .

 

[JAYDEEJOHN]

I will say, the current licenses for x86 is precedent, with conditions.
If another company wanted such a license, under conditions,, then the value would have to be close or the same due to the precedence already set.
Im sure VIAs license isnt the same cost as AMDs, thus a few of the conditions.
Im not familiar with them, but it would have to happen, as the courts would possibly look at a denial as a monopolistic move

 

[anxiousinfusion]

It seems pretty silly to me to license architecture types. Imagine if you could claim license to the truss design of bridges and any other company that wanted to build bridges had to go through you first for the rights to the design type.

Also did anyone else catch this on VIA's site? http://www.via.com.tw/en/initiatives/spearhead/glory/
I figure it has to do with the whole 'living room PC' push being made by valve.

EDIT: Hold on, I'm an idiot that news may be extremely old as it includes DDR1 and Windows XP.

 

[iceclock]

yes the steambox.

 

[Cazalan]

well see jaydee. amds doing bad financially. best thing for them might be arm processors and low powered mobile processors.

There's way more competition in the ARM space. They'll have to compete with Samsung (which makes Intel look like a dwarf), Qualcomm (which is the size of Intel but without tons of debt), Nvidia, AMCC, Calxeda, Texas Instruments, to just name a few. These companies have been doing custom/semi-custom ARM cores for 5+ years. AMD has zero chance to catch up there anytime this decade.

AMD needs to execute better with x86 or they're toast. Steamroller should help a lot there if the IPC gains hold true.

 

[sarinaide]

I hope it has that kind of performance, but I doubt it.

So far every claim AMD have made on iGPU gains have been accurate, AMD want mainstream on Excavator so they need to be aggressive in the iGPU progress to achieve that.