AMD Q4/2013 Desktop Roadmap

[AtenRa]

They will, those are the FS1B based APU, but dont even think for a moment they will be any good for gaming.

They will be fine with a low end GPU for casual gaming.

 

[Blitzvogel]

Any reason why AMD won't release a Jaguar/GCN combo for consumers? They've already completed most of the work with X-One and PS4.

Crappy performance per core and the cores themselves cannot scale that high. Trinity and Richland essentially have similar IPC to Jaguar but can hit 4 GHz which makes more sense, especially in a PC environment where plenty of programs do not scale beyond 1 or 2 cores.

Either console APU would be fine for servers for scientific computing using the CUs, though I don't realistically see the need for 8 CPU cores in that application.

Adapting the Xbox APU for example to a PC would mean making some major changes that make more sense for the PC market and software economics. For example I would replace the Jaguar cores with two Kaveri modules to get the high clock speeds single threaded applications should have access to. Also, I would enable the 2 extra CUs on the die that were disabled for improved yields for the console. I wouldn't hold onto the 32 MB of eSRAM unless it could be used as an L4 cache like Haswell as well as bumped up to 64 MB. The 4 DDR3 controllers are entirely useful with or without the eSRAM. Cheaper motherboards and lower end binned APUs can run dual channel with decent ones being four channel. Of course, the whole point of an Xbox One class APU isn't to be binned.

 

[inf64]

It all depends on the workloads of course. General purpose integer workloads won't be 100% faster (Excavator Vs Piledriver). Let us wait for Steamroller launch, I guess AMD will start talking about Excavator core in more details soon after they officially launch Kaveri.

 

[ShintaiDK]

They will be fine with a low end GPU for casual gaming.

Maybe some indie games. Since Kabini would roll around in the 15FPS range with low settings and low res on most tested games.

http://www.legionhardware.com/articles_pages/amd_a4_5000_kabini_the_mainstream_apu,7.html

 

[ShintaiDK]

It all depends on the workloads of course. General purpose integer workloads won't be 100% faster (Excavator Vs Piledriver). Let us wait for Steamroller launch, I guess AMD will start talking about Excavator core in more details soon after they officially launch Kaveri.

There is nothing that will be 100% faster. Its the same story time after time. Always the next core that comes out with a miracle. And funny enough that miracle never delivers, NEVER!

Even Haswell with higher IPC, AVX2, 256bit wide caches and such could "only" produce 80% faster in what essentially is a HPC load. And across the board we talk the usual 5-10%.

 

[AtenRa]

Maybe some indie games. Since Kabini would roll around in the 15FPS range with low settings and low res on most tested games.

http://www.legionhardware.com/articles_pages/amd_a4_5000_kabini_the_mainstream_apu,7.html

Those test are not with a discrete GPU.

 

[Fjodor2001]

There is nothing that will be 100% faster. Its the same story time after time. Always the next core that comes out with a miracle. And funny enough that miracle never delivers, NEVER!

Even Haswell with higher IPC, AVX2, 256bit wide caches and such could "only" produce 80% faster in what essentially is a HPC load. And across the board we talk the usual 5-10%.

Intel has already picked all the low hanging fruit, hence the low CPU performance improvements we've seen in later CPU generations. AMD has not yet, and therefore still has some low hanging fruit left to pick.

Having said that, I don't expect miracles from AMD either.

 

[ShintaiDK]

Those test are not with a discrete GPU.

Its not getting much better due to the very weak CPU. And with a discrete GPU, why even pick Kabini to start with.

Tomb Raider, benchmark:
Kabini 55FPS with a GTX680.
Haswell 800Mhz 74FPS.

Tomb Raider, ingame:
Haswell 800Mhz 22FPS.

Now what do you think Kabini would end at. :awe:

 

[ShintaiDK]

Intel has already picked all the low hanging fruit, hence the low CPU performance improvements we've seen in later CPU generations. AMD has not yet, and therefore still has some low hanging fruit left to pick.

Oh really, what low hanging fruits are waiting for AMD. That the entire engineering team at AMD havent discovered before?

 

[Ajay]

Oh really, what low hanging fruits are waiting for AMD. That the entire engineering team at AMD havent discovered before?

There is still a ton of work to be done in moving the BD architecture from a pure CMT design to something that resembles a traditional multicore design (ala Intel). There is still allot of work that can be done in improving IPC and ST. Carrizo could make a significant jump on 20nm (if it even comes out on 20nm), but after that GF isn't offering much (14XM - won't be a big gainer, it's optimized for LP, and what's after that?).

Really, 2015 is kind of a last stand for AMD as far as desktop CPUs are concerned, IMHO. And that's only if 20nm actually comes out of GF on time (and that would be a first). A 65 W Carrizo on 28nm would be dead in the water.

 

[jpiniero]

I have a feeling that Broadwell is only going to be released for the ultra high margin areas that AMD doesn't compete in anyway. So Carrizo wouldn't really be competing with Broadwell, it would be Haswell Refresh until Skylake comes out, and that might not be until 2016.

 

[AtenRa]

Its not getting much better due to the very weak CPU. And with a discrete GPU, why even pick Kabini to start with.

Tomb Raider, benchmark:
Kabini 55FPS with a GTX680.
Haswell 800Mhz 74FPS.

Tomb Raider, ingame:
Haswell 800Mhz 22FPS.

Now what do you think Kabini would end at. :awe:

I said with a lower-end Discrete, not High End like GTX680

For example, 25W TDP 2GHz Kabini with R7 250/260 will be fine for most of the games out there. Yes it will not perform good in BF4 or perhaps Crysis 3, but you could play most DX-9 and DX-11 games simple because your limiting factor will be the GPU in most of the cases.

Edit: And with Mantle games in the horizon, you could actually even pair Kabini with a higher-End Discrete like R9 270 or 280

 

[mrmt]

Really, 2015 is kind of a last stand for AMD as far as desktop CPUs are concerned, IMHO. And that's only if 20nm actually comes out of GF on time (and that would be a first). A 65 W Carrizo on 28nm would be dead in the water.

You mean as a market player or as a relevant player? Because they can stay on the market a little longer than that. Beema and Mulins won't be match for Airmont or Goldmont, but they can scavenge sales on the very bottom of the market.

As for their big core CPU business, that's another story...

 

[mikk]

I have a feeling that Broadwell is only going to be released for the ultra high margin areas that AMD doesn't compete in anyway. So Carrizo wouldn't really be competing with Broadwell, it would be Haswell Refresh until Skylake comes out, and that might not be until 2016.

Skylake for Desktop comes 2015.

 

[ShintaiDK]

There is still a ton of work to be done in moving the BD architecture from a pure CMT design to something that resembles a traditional multicore design (ala Intel). There is still allot of work that can be done in improving IPC and ST. Carrizo could make a significant jump on 20nm (if it even comes out on 20nm), but after that GF isn't offering much (14XM - won't be a big gainer, it's optimized for LP, and what's after that?).

Really, 2015 is kind of a last stand for AMD as far as desktop CPUs are concerned, IMHO. And that's only if 20nm actually comes out of GF on time (and that would be a first). A 65 W Carrizo on 28nm would be dead in the water.

20nm in itself doesnt do anything to IPC. Also 65W limit the frequency potential it might unlock.

 

[ShintaiDK]

I said with a lower-end Discrete, not High End like GTX680

For example, 25W TDP 2GHz Kabini with R7 250/260 will be fine for most of the games out there. Yes it will not perform good in BF4 or perhaps Crysis 3, but you could play most DX-9 and DX-11 games simple because your limiting factor will be the GPU in most of the cases.

Edit: And with Mantle games in the horizon, you could actually even pair Kabini with a higher-End Discrete like R9 270 or 280

I used the GTX680 to remove the GPU bottleneck. Kabini wouldnt even be able to push 20fps in Tomb Raider ingame.

Mantle is still nowhere to be seen. And even if we image a huge push. It might only be something like 5-15 games over the entire next year. BF4 is completely MIA. Mantle didnt happen in 2013.

 

[Sweepr]

Skylake for Desktop comes 2015.

Seems like only unlocked Haswell ''K'' models will be replaced by Broadwell ''K'' Q4/2014. The rest of the line up (Haswell Refresh) gets Skylake treatment 2015.

Broadwell-K will feature GT3 ''Gen 8'' (with 128MB eDRAM) graphics, huge upgrade over Haswell-K's GT2 ''Gen 7.5'' graphics.

 

[AtenRa]

I used the GTX680 to remove the GPU bottleneck. Kabini wouldnt even be able to push 20fps in Tomb Raider ingame.

Mantle is still nowhere to be seen. And even if we image a huge push. It might only be something like 5-15 games over the entire next year. BF4 is completely MIA. Mantle didnt happen in 2013.

How many times will you cherry pick a single game to support your agenda ??? Perhaps 2GHz Kabini will not perform ok in Tomb Raider but there are 100 more games where it will do fine.

As for Mantle, in those 5-15 games Kabini will be more than ok and will totally demolish any Intel ATOM. 25W TDP Kabini paired with 115W TDP R7 260X will do a nice Low Power Gaming system.
Intel has nothing in that price/TDP/Performance/Features range.
DX-11.2, True Audio, Mantle, Open CL, AES etc.

 

[SiliconWars]

You mean as a market player or as a relevant player? Because they can stay on the market a little longer than that. Beema and Mulins won't be match for Airmont or Goldmont, but they can scavenge sales on the very bottom of the market.

Pretty sure that they aren't worried about another 12.5% performance the next gen atoms will have over Fail Trail but AMD will struggle to match Intel's $1 billion tablet giveaway next year.

In the end they'll just keep selling them at the low-end notebook while the big guys see who can lose the least in low-end tablets. A bit like 2013 really.

 

[Arachnotronic]

Pretty sure that they aren't worried about another 12.5% performance the next gen atoms will have over Fail Trail but AMD will struggle to match Intel's $1 billion tablet giveaway next year.

In the end they'll just keep selling them at the low-end notebook while the big guys see who can lose the least in low-end tablets. A bit like 2013 really.

SiliconWars, please don't mislead people. Intel's $1b "giveaway" is to:

(1) Help OEMs port devices from QCOM/NVDA/TXN/etc. to Intel's platform
(2) In the mid-range to lower end, compensate for the higher platform bill of materials that BYT-T runs that some of the lower end chips don't.

It's pretty sad to see people like you mislead others. I want to chalk it up to you not doing your research, but I'm worried that this is just out of some irrational hate of Intel.

Basically, you're going to have to accept that any company that tries to compete with Intel is going to have their teeth kicked in sooner or later. You can mock them all you'd like and make snide remarks, but at the end of the day, anybody fighting Intel is fighting a company that generates so much cash that it can employ many tactics to "win" in new markets. It also has a massive capacity to spend on R&D and the company itself has a very well respected brand that adds value to its customers products.

In short, you - and many of the other Intel bashers - simply want to deny that over the next several generations, Intel will be the #1 merchant tablet apps processor vendor. Beginning in 2014, Intel will do the same thing in phones. Once Merrifield/Moorefield pave the way, Broxton/SoFIA and their successors will start doing some serious damage to Qualcomm/MediaTek/etc in phones.

Anyway, good luck to you!

 

[SiliconWars]

You can call it what you want but the bottom line is they are paying OEM's to take Fail Trail off their hands. But what's another $1 billion on an already $2 billion yearly mobile loss anyway?

As for the rest of your post, I'm not even going to dignify it with a response.

 

[mrmt]

Pretty sure that they aren't worried about another 12.5% performance the next gen atoms will have over Fail Trail

Last time I saw Rory "Predator" Read was turning tail and fleeing to a market decisivesly out of Intel's priorities. Maybe because he is confident that Beema and Mulins will be able to decisively take on Bay Trail and Cherry Trail.

But performance isn't the only issue here. I'm sure they are worried about the higher transistor density and lower power consumption of Intel 14nm, while they are stuck on 28nm on both foundries partners, one of them very subpar. It isn't performance alone that is killing AMD big core line, but a combination of extremely efficient chips and very small dies. We'll start to see the very same thing happening on the small core market. It shouldn't happen as fast as with their big core line, because their small core line doesn't leverage on a train wreck like their big core cousins, but it will happen.

 

[Arachnotronic]

You can call it what you want but the bottom line is they are paying OEM's to take Fail Trail off their hands. But what's another $1 billion on an already $2 billion yearly mobile loss anyway?

As for the rest of your post, I'm not even going to dignify it with a response.

All right, whatever you say

 

[ShintaiDK]

How many times will you cherry pick a single game to support your agenda ??? Perhaps 2GHz Kabini will not perform ok in Tomb Raider but there are 100 more games where it will do fine.

As for Mantle, in those 5-15 games Kabini will be more than ok and will totally demolish any Intel ATOM. 25W TDP Kabini paired with 115W TDP R7 260X will do a nice Low Power Gaming system.
Intel has nothing in that price/TDP/Performance/Features range.
DX-11.2, True Audio, Mantle, Open CL, AES etc.

BD was an Intel killer too, PD would fix everything, Hawaii destroyed nVidia and so on. Wait for Steamroller, wait for Excavator etc. Always the next thing will make AMD win.

Lets see where Mantle goes, if anywhere at all. The BF4 patch is already MIA. No performance numbers, just a lot of the usual hype and lack of delivery.

As soon as you need to pair Kabini with a dGPU, Silvermont is faster and got only 40% the TDP of the Kabini. But again, if TDP doesnt matter, then a 35W Celeron would just destroy Kabini as well, for a much cheaper price.

Kabini is a huge misfit.

 

[daxzy]

Well in all fairness, if the revealed specs of Kaveri are true, it would make a compelling product. I'm upgrading my current HTPC (2009 era) and can afford to wait the 2 weeks to see Kaveri benchmarks.