AMD Q4/2013 Desktop Roadmap

[Arachnotronic]

28nm or 20nm is the big remaining question. Also it seems they aren't going for the maximum possible performance, otherwise they wouldn't lower the max TDP to 65W. I'm also surprised that AMD doesn't bring a Beema successor till 2016.

My sources say 28nm.

 

[Abwx]

Im kind of worried about Beema until 2016... Intel gona launch 2 products with the 3rd in the works in that lifetime, Beema its either too good, or they are just not trying anymore.

Double the perf/watt of the current jaguar line up,
numbers extracted from real silicon have been published
and are quite good.

 

[Fjodor2001]

Wow, not even DDR4 for excavator. And 65W TDP confirmed. No chipset update either.

And FX completely dead. AMD really is the new VIA.

Do you think Intel's offering for mainstream desktop CPUs in 2014-2015 will be that much more exciting compared to its previous generation CPUs? After all both Steamroller and Excavator is expected to bring quite an in crease in CPU performance, compared to the increases Intel has been delivering lately.

 

[Arachnotronic]

Double the perf/watt of the current jaguar line up,
numbers extracted from real silicon have been published
and are quite good.

LOL, you actually believe that tripe? What magic do you think AMD concocted to get TWICE the performance per watt on the same process node? Black magic? Voodoo?

 

[Abwx]

LOL, you actually believe that tripe? What magic do you think AMD concocted to get TWICE the performance per watt on the same process node? Black magic? Voodoo?

They published numbers from real silicon.

The new 2014 AMD A-Series mainstream APU platform, codenamed “Beema,” is expected to deliver up to 104 percent better productivity performance per watt when compared to the previous generation “Kabini” platform. Testing conducted by AMD Performance Labs on optimized AMD reference systems. PC manufacturers may vary configuration yielding different results. PCMark 8 - Home score divided by TDP (W) is used to simulate productivity performance per watt; the Beema platform (15W) scored 2312 while the Kabini platform (25W) scored 1861. AMD "Larne" reference platform system used for both APUs. Kabini-based AMD A6-5200 quad-core APU with AMD Radeon™ HD 8400 Graphics, 2x2GB of DDR3-1600MHz RAM, Windows 8.1, 13.200.11.0 - 03-Sep-2013 driver. Pre-production engineering sample of “Beema” quad-core APU with next generation AMD Radeon graphics (model number TBD), 2x2GB DDR3-1600MHz RAM, Windows 8.1, and unreleased reference driver.BMN-3

The new 2014 AMD A-Series low power APU platform, codenamed “Mullins,” is expected to deliver up to 139 percent better productivity performance per watt when compared to the previous generation “Temash” platform. Testing conducted by AMD Performance Labs on optimized AMD reference systems. PC manufacturers may vary configuration yielding different results. PCMark 8 - Home score divided by TDP (W) is used to simulate productivity performance per watt; the Mullins platform (4.5W) scored 1809 while the Temash platform (8W) scored 1343. AMD "Larne" reference platform system used for both APUs. Temash-based AMD A6-1450 quad-core APU with AMD Radeon™ HD 8250 Graphics, 2x2GB of DDR3-1333MHz RAM (running at 1066MHz), Windows 8.1, 13.200.11.0 - 03-Sep-2013 driver. Pre-production engineering sample of “Mullins” quad-core APU with next generation AMD Radeon graphics (model number TBD), 2x2GB DDR3-1333MHz RAM, Windows 8.1, and unreleased reference driver. MUN-3

 

[Sweepr]

Double the perf/watt of the current jaguar line up,
numbers extracted from real silicon have been published
and are quite good.

Bay Trail-T already beats 15W TDP A4-5000 CPU-wise and I can buy tablets based on it right now (no need to wait till H2/2014), Cherry Trail-T will only improve here.
4.5W TDP Beema scores close to A4 5000 in 3DMark11, which would put it around 2x as fast as Bay Trail-T in actual games according to AnandTech's preview. I'm pretty sure Cherry Trail-T can easily close that gap with 16 EUs (Gen 8), four times as much as the current generation + new architecture.
And we might see Willow Trail-T in 2015 (Cherry Trail-T successor) before Beema's sucessor arrives if this roadpmap is accurate.

 

[Arachnotronic]

They published numbers from real silicon.

Yeah, it's slimy as hell to do performance per watt calculations with "TDP" numbers. Intel is just as guilty of that garbage with its SDP and claiming "Core performance in a tablet", to be fair.

 

[Abwx]

Yeah, it's slimy as hell to do performance per watt calculations with "TDP" numbers. Intel is just as guilty of that garbage with its SDP and claiming "Core performance in a tablet", to be fair.

The 4.5W TDP mullins is 2W SDP according to AMD , so we
have a valuable comparison with BT for power usage
in said so called "most probable scenarii"...

Edit : Kabini and BT have Core perfs by the virtue
of 4 cores against 2....

 

[OCGuy]

They threw in the towel on the performance market ever since the Ph II X6 stomped all over Bulldozer. :/

 

[JDG1980]

If Excavator can catch up to Sandy Bridge in IPC (quite possible, given the leaked die shots we've seen), then AMD would be foolish not to go back after the server/high-end desktop market. The margins there are much higher, and Intel has slowed down the pace of innovation enough that they could reasonably hope to catch up.

It doesn't necessarily have to be AM3+; they could create a new socket that would be shared between high-end desktop and server, and would replace AM3+, C32, and G34. Basically an equivalent to Intel's LGA 2011. The biggest thing hurting AMD in the high-margin server market is their weak IPC. If they fix that, there's no reason why they should not try to take away some of Intel's lucrative business here, especially since Intel keeps their high-end server CPUs one release behind state-of-the-art.

I suspect nothing is really set in stone at this time. If SR and/or EX are greeted with a collective yawn, there will be no FX or server release. On the other hand, if all the reviews love it and say how awesome it would be if there were full-fledged 6-core/8-core parts, then there's a good possibility that AMD will try to devote some spare resources to giving people what they want. Announcing new FX CPUs now would mean they would be locked into releasing them whether or not the demand is there. It would also kill sales of Vishera almost immediately, especially if a new socket was announced. Who would buy AM3+ and Vishera now, if the roadmap said that AM4 and Steamroller FX was coming in June? Better to keep the options open and say nothing.

I can't imagine AMD actually still selling Visheras at the end of 2015. How is this even remotely viable? The manufacturing cost is high (large die), and premium prices are out of the question since they are fairly decisively beaten by Intel on the high end. The FX-8350 is already down to $199 (less at Micro Center). Who would want to buy that in two years? The only thing I can think of is that Hector Ruiz's crazy contract with GloFo commits them to a certain volume of 32nm wafers, and they figure they have to do something with them. The processors might be sold at a loss, but less of a loss than if they paid the penalty and got nothing in return.

 

[SiliconWars]

They threw in the towel on the performance market ever since the Ph II X6 stomped all over Bulldozer. :/

99% of Intel's CPU sales (not including servers) have integrated graphics. That pretty much says it all for the high end desktop.

The industry went to APU's a long time ago and it's just not worth making these high end chips without the server market to support it.

 

[JDG1980]

99% of Intel's CPU sales (not including servers) have integrated graphics. That pretty much says it all for the high end desktop.

The industry went to APU's a long time ago and it's just not worth making these high end chips without the server market to support it.

But high-end desktop parts pretty much are server parts, just with a few of the advanced features fused off. Intel develops the LGA 2011 series as server chips first and foremost, then disables ECC and various other features to make the enthusiast-oriented CPUs. The server chips continue to be manufactured because they are needed by the industry and because they command high margins. And as long as that's the case, the enthusiast market can continue to be served without too much additional effort needing to be put in. is.

 

[SiliconWars]

Yep but that's the thing, AMD is in full retreat out of classic server space and is going all-in with ARM microservers.

If you add it all up it was quite obviously going to happen. It'll cost ~$100m to design an octo-core Excavator and the chances of AMD getting that back in profit isn't very high.

These numbers might be a bit off but Intel sells what, around 250 million CPU's so 1% of that is 2.5 million "high end". Even if they made $200 profit on average that's still only $500 million bucks in profit. AMD would likely get $50 profit on each, if lucky, and with only 1/5th the market. That's 25 million bucks profit on a CPU that would easily cost double if not quadruple that amount to design.

 

[OCGuy]

99% of Intel's CPU sales (not including servers) have integrated graphics. That pretty much says it all for the high end desktop.

The industry went to APU's a long time ago and it's just not worth making these high end chips without the server market to support it.

You mean it isn't worth making high end chips if the performance and/or public perception doesn't justify high enough margins to support it.

 

[SiliconWars]

You mean it isn't worth making high end chips if the performance and/or public perception doesn't justify high enough margins to support it.

No it's not worth making high end chips if they won't bring in the profit to cover the design costs. With the move to microservers that was pretty much that for AMD's high end, imo.

 

[Abwx]

If Excavator can catch up to Sandy Bridge in IPC (quite possible, given the leaked die shots we've seen), then AMD would be foolish not to go back after the server/high-end desktop market. The margins there are much higher, and Intel has slowed down the pace of innovation enough that they could reasonably hope to catch up.

Expect the margins to collapse in this market , cheap multicore
parts will undoubtly take a big chunk of the pie.

I can't imagine AMD actually still selling Visheras at the end of 2015.

For 2014 they ll have a 20% better perf/watt part for servers,
that s not negligible and they ll surely use 8C parts for the AM3+
plateform.

 

[SiliconWars]

Don't forget that Mantle will breath a lot of new life into Vishera as well. If AMD spends all the ~$100 million saved (by not designing an octo-core Excavator), on paying devs to adopt Mantle instead that will be money way, way better spent.

I believe this will be their tactic for the next few years.

 

[Abwx]

Don't forget that Mantle will breath a lot of new life into Vishera as well. If AMD spends all the ~$100 million saved (by not designing an octo-core Excavator), on paying devs to adopt Mantle instead that will be money way, way better spent.

I believe this will be their tactic for the next few years.

What a lot of people seems to forget is that with current
softs Vishera is still under utilized , the beast has a lot of
cores that render it quite future proof with perfs increasing
as softs are more MThreaded...

 

[Centauri]

Very cool to see AMD sticking with FM2+ for a while. Definitely gonna grab a board and Kaveri CPU when they're available.

 

[PPB]

Someone explain to me why when Intel drops the rated TDP of their processors is good because all what matters now is perf/watt, but when AMD does it its because they are becoming the new VIA?


Oh yea, because they were trolling. Forgot about that possibility :awe:

 

[monstercameron]

Someone explain to me why when Intel drops the rated TDP of their processors is good because all what matters now is perf/watt, but when AMD does it its because they are becoming the new VIA?


Oh yea, because they were trolling. Forgot about that possibility :awe:

this is what via did before dropping out of the x86 consumer market, rgwt dropped tdp, went mobile then bowed out and also went arm...

 

[blastingcap]

Wow, not even DDR4 for excavator. And 65W TDP confirmed. No chipset update either.

And FX completely dead. AMD really is the new VIA.

Freaking A, there go my plan of using my AM3+ mobo as a cheap NAS. I was hoping for a 28 nm AM3+ CPU I could drop in, but noooo. And their FM-whatever garbage doesn't support ECC so that is not an option. UGH. A big fat ugh for those of us waiting on FX.

 

[AtenRa]

I wonder if this slide is legit since a few time ago AMD published
official DT roadmap that show Beema clearly replacing Jaguar while
this roadmap show Kabini being produced for the whole of 2015,
that is 18 months after Beema is released....

I knew about the FS1B socket Kabini for a while now but i was told it was going to be a Q4 2013 part. Apparently it has been pushed back in Q1 2014 and that is too close to Beema release.

 

[monstercameron]

I knew about the FS1B socket Kabini for a while now but i was told it was going to be a Q4 2013 part. Apparently it has been pushed back in Q1 2014 and that is too close to Beema release.

what is socket FS1B, qiuck srech shows FS1 as llanos socket...

Bay Trail-T already beats 15W TDP A4-5000 CPU-wise and I can buy tablets based on it right now (no need to wait till H2/2014), Cherry Trail-T will only improve here.

top of the line bay-trail hardly beats the middle of the range a4-5000 in cpu benchmarks and is completely beaten by the amd part in gpu perf. sure cherrytrail will be better than baytrail...but so will beema/mullins.

also you can get temash in a tablet now, although baytrail has better offerings...cut out the crap fanboi hyperbole

 

[AtenRa]

what is socket FS1B, qiuck srech shows FS1 as llanos socket...

FS1 socket is used in the AMD embedded APUs. FS1R2 is the equivalent of Desktop FM2 for embedded R series.

The FS1B will be for the Desktop Jaguar products. Unlike current Desktop BGA Jaguar APUs, with the FS1B you will be able to change the APU like Trinity and Kaveri by keeping the same Mobo.