The AMD Kabini Review: A4-5000 APU Tested

[NTMBK]

what about the xeon phi, couldn't that make an awesome graphics + compute system for consoles?

There's a reason it seems so much like a graphics card- read up about Larrabee some time.

There were random rumours that Intel actually was shopping Larrabee for the next-gen consoles several years back, but between the tech not being ready and the global economic downturn the whole thing got canned. Totally unsubstantiated rumours, of course. http://www.theinquirer.net/inquirer/news/1048101/intel-console-deal-larrabee-ground

 

[Roland00Address]

Dont know for TSMC , for GF AMDs Trinity has about 5.3 millions transistors/mm2, going to 28nm should increase density by 30% to 6.9M/mm2.

As a comparison Intels IVB has 8.75M/mm2 density.

Trinity is silicon on insulator you can't compare that to bulk which jaguar would be.

 

[galego]

You honestly don't think Intel has the technology to make an APU? Where have you been for the last 3 years?

There are APUs and there are apus, and please don't miss half of the equation. I am sure that investing enough brute resources Intel could do something close to what will be included in next gen consoles, but Sony and Microsoft did want them for yesterday and without the overprice.

 

[Vesku]

Microsoft already tried the Intel route with the original Xbox. After that they realized they needed more control over production decisions and pricing mechanisms than Intel was willing to sign up for.

 

[NTMBK]

There are APUs and there are apus, and please don't miss half of the equation. I am sure that investing enough brute resources Intel could do something close to what will be included in next gen consoles, but Sony and Microsoft did want them for yesterday and without the overprice.

They've had the fundamental tech since Sandy Bridge and refined it with Ivy Bridge- and they have the process lead which means they can bruteforce the problem with less efficient GPU parts and still wind up with a reasonably sized part.

But yes, the pricing would be way outside what the console makers want to pay.

 

[R0H1T]

They've had the fundamental tech since Sandy Bridge and refined it with Ivy Bridge- and they have the process lead which means they can bruteforce the problem with less efficient GPU parts and still wind up with a reasonably sized part.

But yes, the pricing would be way outside what the console makers want to pay.

By APU if you mean something like trinity or richland then Intel has had the tech for what ~4yrs now ? But going against something like Kaveri(HSA based) Intel will need a more competent & refined IGP, not something like a makeshift effort because the GPU inside that chip shouldn't hold back the CPU, ideally that is, but Intel is a good 2~4yrs behind AMD's current GPU tech at this point in time so the gap will only widen with the ever improving bulldozer microarch based cores !

 

[NTMBK]

By APU if you mean something like trinity or richland then Intel has had the tech for what ~4yrs now ? But going against something like Kaveri(HSA based) Intel will need a more competent & refined IGP, not something like a makeshift effort because the GPU inside that chip shouldn't hold back the CPU, ideally that is, but Intel is a good 2~4yrs behind AMD's current GPU tech at this point in time so the gap will only widen with the ever improving bulldozer microarch based cores !

It's not the fundamental tech that is really the issue, it is Intel's reluctance to accept lower margins. Their Ivy Bridge dies are considerably smaller than Trinity- Trinity is 246mm^2, 4 core IB with GT2 is 160mm^2. With Haswell GT3 Intel significantly crank up their die size (no precise measurements yet, but the photos show it as pretty big) and what do you know? Their graphics performance goes way up. GT3e will probably beat Richland in graphics performance- although it will probably also cost significantly more.

Intel have the building blocks for a DX11 GPU, and to get good performance all they need to do is duplicate it more times on the same die. They have the tech to deal with bandwidth issues, whether it is via a GDDR5 bus (see Xeon Phi) or via EDRAM. But they don't have the will to accept the lower margins they would need to compete with AMD on price.

 

[CHADBOGA]

Intel have the building blocks for a DX11 GPU, and to get good performance all they need to do is duplicate it more times on the same die. They have the tech to deal with bandwidth issues, whether it is via a GDDR5 bus (see Xeon Phi) or via EDRAM. But they don't have the will to accept the lower margins they would need to compete with AMD on price.

The Haswell GPU EU's are essentially the same as Ivy Bridge's, but they have major EU architecture revamps coming for the successor chips that come out on 14nm.

At least the first revamped GPU on 14nm, will only have to contend with AMD on 28nm.

 

[R0H1T]

The Haswell GPU EU's are essentially the same as Ivy Bridge's, but they have major EU architecture revamps coming for the successor chips that come out on 14nm.

At least the first revamped GPU on 14nm, will only have to contend with AMD on 28nm.

Maybe you're being too optimistic, I'm not sure when/where the next die shrink APU will come from, as in GLF or TSMC, but I don't expect the 14nm Intel desktop nor mobile(atom based) parts to be on shelves anytime before the third quarter next year, even that is an optimistic outlook. Anyways I expect a die shrunk APU from AMD within 6~12 months of the Intel 14nm parts & perhaps with GCN 3.0 :awe:

 

[ShintaiDK]

Maybe you're being too optimistic, I'm not sure when/where the next die shrink APU will come from, as in GLF or TSMC, but I don't expect the 14nm Intel desktop nor mobile(atom based) parts to be on shelves anytime before the third quarter next year, even that is an optimistic outlook. Anyways I expect a die shrunk APU from AMD within 6~12 months of the Intel 14nm parts & perhaps with GCN 3.0 :awe:

Its pretty obvious that Kaveri on 28nm will have to fight Broadwell on 14nm with a new GPU uarch.

 

[AtenRa]

Its pretty obvious that Kaveri on 28nm will have to fight Broadwell on 14nm with a new GPU uarch.

If Kaveri will launch in Q4 2013 it will compete against Haswell not Brodwell.

 

[CHADBOGA]

If Kaveri will launch in Q4 2013 it will compete against Haswell not Brodwell.

And what happens when Broadwell comes out. :whiste:

 

[NostaSeronx]

And what happens when Broadwell comes out. :whiste:

Then Carrizo launches in Q3 2014.

 

[Abwx]

And what happens when Broadwell comes out. :whiste:

At least a full year after Kaveri is launched...

 

[Idontcare]

And what happens when Broadwell comes out. :whiste:

My guess is that come Broadwell we'll all find out that we pine for the days of 5GHz 32nm SB OC'ing all the more than we do now.

 

[BallaTheFeared]

My guess is that come Broadwell we'll all find out that we pine for the days of 5GHz 32nm SB OC'ing all the more than we do now.

Your pessimism wounds me deeply sir.

 

[Idontcare]

Your pessimism wounds me deeply sir.

I'm a realist.

The glass is definitely there, maybe. And it just might happen to be half-full, but full of what exactly?

Is it half-full of piss or vinegar? No one will ever say, and this intentional omission of facts regarding the molecular nature of the contents of said half-full glass troubles me at a fundamental level

If that makes me indistinguishable from a pessimist then all I can conclude is pessimists and realists must be bosons...versus the fermions that I have long hoped us to be

 

[ShintaiDK]

My guess is that come Broadwell we'll all find out that we pine for the days of 5GHz 32nm SB OC'ing all the more than we do now.

If you just want frequency and not actual performance, then there are other alternatives

 

[Gikaseixas]

The fun factor has been getting less and less attention. That's a negative in my book.

 

[AtenRa]

And what happens when Broadwell comes out. :whiste:

First lets see $130 and lower Haswell CPUs and then we can talk about Brodwell. Do you honestly expecting a Core i3 GT3 haswell for socket 1150??

What scares me is that a 6-core Kaveri(if they'll make one) could be that good that it would be a $150+ and a 8-core SR FX at $250.00

 

[inf64]

I don't wanna post a new topic since nothing is verified yet,so I'll leave it here(and IDC can move it if needs be).
Fellix_bg dug out a pic of supposed SR die shot.

Compare to PD/BD side by side(thanks goes to sdlvx on SemiAccurate forums who made it):

Notice the big difference to key areas AMD highlighted on HC last year : front end, instr. and data caches, decoders(doubled), souped up integer cores, branch prediction and finally (makes it too good to be true) : doubled up FP unit. The FP unit improvement might be related to latest update to AMD's 15h SOG manual than had an update in May (this month ).
http://www.planet3dnow.de/cgi-bin/newspub/viewnews.cgi?id=1368122313

Old manual:

New manual(Kaveri2.0??):

New manual states native 256bit FP pipelines for some next AMD 15h core iteration. It has usually been the case in tha past that it's related to the next iteration of the core that is yet to launch (meaning successor to PD=>Kaveri).

 

[Enigmoid]

If Kaveri will launch in Q4 2013 it will compete against Haswell not Brodwell.

Seriously doubt a Q4 launch. Thinking more likely a Q1/Q2 2014 launch.

Richland is just being released now (looking like june, same as haswell for official launch) and seems like they will have inventory problems like they did with llano if they release kaveri too soon. Not to mention that a 6 month lifespan for richland would be pretty pointless

 

[SiliconWars]

Richland is just a stop-gap though, at least on desktop and it's a drop-in FM2 replacement so it's not like the OEM's have to commit to a year with it.

 

[erunion]

Richland is just a stop-gap though, at least on desktop and it's a drop-in FM2 replacement so it's not like the OEM's have to commit to a year with it.

Same thing as Vishera then. Vishera didn't get a half year life span.

AMD has been really sly about release dates lately. In january they said Richland was shipping, leading people to claim it was weeks away for the past 5 months.

If kaveri is launched in 2013, it will be a paper launch.

 

[CHADBOGA]

At least a full year after Kaveri is launched...

I very much doubt that Kaveri is going to have that much breathing space.