The Kaveri Pre-Launch Thread (A10-7800 and A10-6800k @3,5 Ghz)

[inf64]

Sorry I had to vent there. I've seen it many times now in various topic (not posted by you, but by many users). You shouldn't have deleted it, just read the orginal post in the topic next time . Whenever you see a "news" piece on wccftect the chances are great it has been already posted on some forum and thoroughly discussed.

 

[Fjodor2001]

Sorry I had to vent there. I've seen it many times now in various topic (not posted by you, but by many users). You shouldn't have deleted it, just read the orginal post in the topic next time . Whenever you see a "news" piece on wccftect the chances are great it has been already posted on some forum and thoroughly discussed.

Yeah, I know. But sometimes WccfTech has the benefit of translating weird Chinese forum posts into English better than Google translate does at least. So then I think it serves a purpose, as long as the original source is credited too. You also have to admit they are quite quick to pick up news / leaked info, faster than many other sites. One just have to be cautious and critical when reading the info there.

 

[parvadomus]

Didnt expect such a big speed jump in wprime and superpi!. It looks like they shortened like 1/3 the pipeline or dramatically improved the branch predictor.
Its like 30% IPC gain on this types of loads, probably games too

 

[JDG1980]

Interesting, if it's accurate. These benchmarks are all over the board. Super Pi (typically an AMD weak point) sees ~30% improvement at the same clock speed as Richland, and wPrime scores are more than 40% better! But Cinebench only has about a 11% boost, and performance in the Excel benchmark (not sure what it is exactly - Monte Carlo simulation?) is virtually unchanged from Richland.

wPrime is a multi-threaded benchmark, so what we're seeing is partially IPC improvements and partially the removal (or at least reduction) of the CMT penalty. SuperPi is harder to explain; it has legacy x87 FPU code, single-threaded, and I can't imagine that this would have been a priority for AMD to optimize.

What kind of workload is Cinebench, exactly? Integer, floating-point, or a mixture of both? 3D rendering usually uses floating-point, so I'd guess that this is what's in play, and the lack of substantial FPU improvement is why Steamroller only sees lackluster gains. AMD presumably hopes that OpenCL+HSA will help fill that gap, but there are a lot of existing programs that can't or won't be rewritten. Excavator is supposed to increase FPU performance dramatically, but that's still a year or more away.

If the Excel benchmark is both single-threaded and FPU-bound, then that could explain the lack of meaningful gains there. But SuperPi remains a mystery; it should suffer the same problem, even more so because it uses legacy opcodes that they wouldn't be expected to bother with optimizing for. Wikipedia says the results can be dependent on memory bandwidth, so maybe that's what we're seeing? (But in that case, wouldn't the graphics core see bigger improvements as well? It seems to be bandwidth-limited in most games on both Richland and Kaveri.)

 

[Spawne32]

If the IPC improvements are truely this significant, I dont understand the logic behind leaving the FX lineup to the wasteside. Why not 6 and 8 core variants of these improvements? Or am i missing the fact that this is exclusive to the fact that APU's incorporate a GPU which gives them a computational edge. I am a newbie to the whole onboard graphics concept in a system where dedicated graphics is almost certainly used.

 

[monstercameron]

If the IPC improvements are truely this significant, I dont understand the logic behind leaving the FX lineup to the wasteside. Why not 6 and 8 core variants of these improvements? Or am i missing the fact that this is exclusive to the fact that APU's incorporate a GPU which gives them a computational edge. I am a newbie to the whole onboard graphics concept in a system where dedicated graphics is almost certainly used.

my guess is, fx brand is dead again, new branding incoming with traditional cpus launching later possibly all fm2+ based.

 

[Spawne32]

my guess is, fx brand is dead again, new branding incoming with traditional cpus launching later possibly all fm2+ based.

Seems like a logical assumption with the way FX motherboard support is.

 

[inf64]

my guess is, fx brand is dead again, new branding incoming with traditional cpus launching later possibly all fm2+ based.

FX brand is not dead .

 

[jpiniero]

If the IPC improvements are truely this significant, I dont understand the logic behind leaving the FX lineup to the wasteside. Why not 6 and 8 core variants of these improvements?

They don't have the money to justify the design and verification.

 

[Spawne32]

They don't have the money to justify the design and verification.

Yet they threw money into the 9xxx series which was a complete failure.

 

[monstercameron]

FX brand is not dead .

but they have ran out of numbers unless the restart at...lets say fx-51 :biggrin:

full circle.

 

[AtenRa]

Yet they threw money into the 9xxx series which was a complete failure.

Im sure you have financial numbers to backup that :whiste:

 

[jpiniero]

Yet they threw money into the 9xxx series which was a complete failure.

It probably didn't cost them that much to do the 9xxx series.

The other problem of course is that OEMs don't want a CPU without a GPU included, so you are talking about retail only. Once they decided to slowly withdrawal from the x86 core Server market, the justification of the expense of building the big dies went away.

 

[rtsurfer]

Yet they threw money into the 9xxx series which was a complete failure.

I am not sure how much it costs to bin your top CPUs which you are already manufacturing and selling them as rebrands.

 

[JDG1980]

If the IPC improvements are truely this significant, I dont understand the logic behind leaving the FX lineup to the wasteside. Why not 6 and 8 core variants of these improvements?

Basically, AMD is severely cash-constrained and is regrouping. They're putting money only into those areas where they currently have a real competitive advantage - consoles, discrete GPUs, and APUs. A Steamroller CPU with three or four modules would be superior to FX-8350, but it would also be a low-volume part with a questionable profit margin. The truth is that enthusiast chips don't pay the bills. Most of them are actually cut-down versions of server chips; this is true of both Vishera, and Intel's LGA 2011 "E"-series. So the relevant question isn't whether enthusiasts would like an 8+ core Steamroller CPU, but whether an 8+ core Steamroller CPU would be competitive in the server space. At this time, the answer appears to be no. That might change with Excavator, especially if AMD's profit margins get healthier and they have a bit more room to experiment.

Yet they threw money into the 9xxx series which was a complete failure.

The FX-9xxx series really just consists of factory-overclocked Vishera chips. The only engineering work they had to do was validation and binning. There were, as far as I know, no architectural changes. Despite low sales, the program probably turned at least a small profit, since they were able to squeeze out a few more bucks from the mature 32nm process.

 

[CHADBOGA]

There are no "more" Kaveri benchmarks. That is the very link from the OP that we are discussing... I hate I had to click on wcftech giving them a page visit and knowing up in front they are late and they've just ripped off some other news source who found it first. Also there is no dual graphics, those boons from wcftech cannot even read the simple charts . It's 7800+ discrete card to show how much behind the dGPU APU still is.

My confidently stated claim from long ago that Kaveri's IGP wouldn't match a HD5770 / HD7750 is looking pretty good, despite what the folks who chose to disagree with me had to say.

 

[frozentundra123456]

My confidently stated claim from long ago that Kaveri's IGP wouldn't match a HD5770 / HD7750 is looking pretty good, despite what the folks who chose to disagree with me had to say.

Very true. The same goes for intel as well actually. Seems like every new generation of igp is supposed to be the one that makes gaming without a dgpu viable at a decent level, but it never quite materializes. The only igp that so far has been able to solve the bandwidth problem and deliver on its promises is Iris pro for mobile, but that is such a niche product it hardly changes the landscape.

Edit: Kaveri *might* be attractive if AMD can finally work out the asymmetric crossfire issues and allow one to pair a HD7750 with it and get performance in the HD7850 range, but we have to wait and see if that ever happens.

 

[JDG1980]

I get the feeling that AMD's GPU improvements are geared more towards GPGPU acceleration than gaming. It's potentially a much larger market. With HSA (eliminating the need to shuffle data in and out of separate video RAM), the difference between 7750 and Kaveri on OpenCL should be much smaller than the difference on 3D gaming.

 

[monstercameron]

Very true. The same goes for intel as well actually. Seems like every new generation of igp is supposed to be the one that makes gaming without a dgpu viable at a decent level, but it never quite materializes. The only igp that so far has been able to solve the bandwidth problem and deliver on its promises is Iris pro for mobile, but that is such a niche product it hardly changes the landscape.

Edit: Kaveri *might* be attractive if AMD can finally work out the asymmetric crossfire issues and allow one to pair a HD7750 with it and get performance in the HD7850 range, but we have to wait and see if that ever happens.

I agree with this sentiment, kaveri might well be a disappointment in terms of raw perf gain but lets see how mantle, hsa and the new cf will work.

 

[Homeles]

Very true. The same goes for intel as well actually. Seems like every new generation of igp is supposed to be the one that makes gaming without a dgpu viable at a decent level, but it never quite materializes.

The same can be said for a whole slew of products.

"Y generation of products will make X generation obsolete!"

It's a goddamned meme.

 

[SiliconWars]

Some things are just out of your hands, especially when you are AMD. They are trying to push the envelope but are getting no encouragement. DDR4 should have been mainstream long before now but they aren't a memory company. It's no wonder desktop is dying with this level of myopia.

 

[PPB]

Some things are just out of your hands, especially when you are AMD. They are trying to push the envelope but are getting no encouragement. DDR4 should have been mainstream long before now but they aren't a memory company. It's no wonder desktop is dying with this level of myopia.

Aye, they already did everything they could regarding RAM when they started rebranding someone's else sticks as their own, just to make sure the RAM gimping on prebuilts by OEMs ceased.

 

[Shivansps]

the results are good if i take the A10-6800K vs A10-5800K numbers at launch in consideration.

 

[Tuna-Fish]

Woo! Look at that SuperPi score! AMD's best yet

In all seriousness, is that because of the 3 FPU stages, down from 4?

IMHO superpi is more about L1 cache than FPU. They supposedly fixed the write bottleneck on L1d -- that should be enough for that superpi improvement alone.

 

[DaZeeMan]

While I was googling Kaveri last night, I came across another discussion on the Linustechchips website. They said that, according to the text accompanying those benchmarks, that the 6800K had been underclocked (their word normalized) to 3.5 GHz, to match the Kaveri clock speed. Presumably to compare how much better or worse the new architecture would fare at the same clock speed.

Discussion here: http://linustechtips.com/main/topic/97361-first-kaveri-apu-benchmarks-surface/

If so, I can see why this can cause some confusion amongst this crowd, assuming this 'revelation' is correct. Especially since, if indeed the Richland was underclocked, i.e. how would the Kaveri fare against a fully clocked Richland?

I do not know how to read Chinese, though, so I can neither confirm or deny what the associated text says r.e. the 3.5 GHz normalization thing... just thought I'd share that tidbit for the collective here.

I am waiting, along with the rest of us, for the day that NDAs are lifted and we see more benchmarks from the 'usual suspects'. In the meantime, it's an interesting sneak peek, if nothing else. And who knows how 'reputable' this source is in any case? For example, were we dealing with 'alpha', or more mature, Kaveri drivers? So many things remain unclear...

AMD's CES press conference is in less than two days now. I am curious as to when the Kaveri NDA's will actually be lifted? I've seen multiple dates now, but Jan 6th seems like a 'natural' time to lift those. Of course, work on Kaveri's 'release' drivers might not be fully ready by the 6th... in which case we may be waiting a bit longer. We've seen Jan 13th, 14th, and 24th as other potential 'launch' dates.

Edit: Here's the associated 'translated' text from the baidu.com article:

Test Method: Due to the participation A10-7800 APU positioning evaluation is not the highest, so if it is directly related to the frequency of the big gap between the generation of positioning the highest A10-6800K APU comparison is not accurate, objective, so the processor performance tests, we will have two processor frequency is set to 3.5GHz, the same frequency testing by methods to understand the differences in performance relative to the roller structure for piling machine architecture. In the APU integrated graphics core performance testing, performance testing of heterogeneous computing, APU use the default settings.