http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=1890557&highlight=

The AMD A-Series APU processor-in-a-box (PIBs) for the AMD A10-7850K and AMD A10-7700K, which started shipping in Q4 2013, are available starting today. The AMD A8-7600 will be shipping in Q1 2014. Additionally, the AMD Radeon™ R9 2400 Gamer Series memory is tested and certified for AMD A10 APUs, unleashing their full potential with AMD Memory Profile technology (AMP) offering speeds up to 2400MHz. For more information, please visit the Radeon Memory product page.

The AMD A-Series APUs are also available today in PCs from our partner system builders. For more information, please visit our product information page.

I am curious to see what Mantle can do for these APUs as well as for AMD's GPU line that supports it to…provided EA gets their head out of their ass with BF4. Or another Mantle title is released.

@Arinoth:

I am curious to see what Mantle can do for these APUs as well as for AMD's GPU line that supports it to…provided EA gets their head out of their ass with BF4. Or another Mantle title is released.

I fully agree with you and want to see what they can do to, think I will have to pick one up asap 🙂

We did not have to speculate on how BF4 will run on A10 Kaveri. I already seen it long time ago:

AMD Kaveri A10 APU Running Battlefield 4 Demo, Windows 8 JPEG Decode Acceleration with HSA – 01:57
— PC Perspective

😉 Sure, they picked up the corridor for fastest FPS, but still…  😉 Impressive, IMHO.

think I will have to pick one up asap 🙂

Oh, you lucky suxxka!  😉 And maybe Kaveri A10-7xxx won't be speed-throtled for SuperPi anymore? 🙂 You know the Bulldozer Conditioner, right? Simply spoken, just enable some legacy x87 registers and from 22sec get the SuperPi 1M test at 4.1GHz A-10 6800K to 19sec  😉

Bulldozer Conditioner R1.01B

Note: "x87 instruction (NRAC) block" -> Enabled means that the instruction is blocked (default on all 15h family APU/CPU/NPUs). Disabling it make the SuperPI "a bit" faster.
http://www.uloz.to/x3jsw1QT/bulldozer-conditioner-r1-02b-zip (version 1.02 and not v1.00, witch is linked into most articles)

🙂

Now if we could pick an AMD card at a decent price like pre-mining era  ;D

Well… you hardly can blame AMD for that 😉 Of course, this is a problem, but in the end, we should enjoy much cheaper graphic cards, when all this frenzy go by-by and then the miners will be looking for selling pretty huge amount of good cards (among other HW).

Yet this is about the new APU graphic speed and it seems to be surprisingly good - for APU. I never thought it can run BF4 at any decent framerate - despite the choosen example is biased and we should see this in multiplayer action, that will be real test 🙂

Still it looks like that at low detail settings it can run WOT pretty smoothly, witch it hard to manage with GeForce 8600 GT, so... that is great result for APU. If they show the performance on something less demanding, then it could be actually usefull.

It is confirmed, that the new AMD Kaveri got 4 channel memory controler:
http://support.amd.com/TechDocs/49125_15h_Models_30h-3Fh_BKDG.pdf
(BIOS and Kernel Developer’s Guide (BKDG) for AMD Family 15h Models 30h-3Fh Processors)

The processor includes two DRAM controllers (DCTs), named DCT0 and DCT3. Each DCT controls one 64-bit DDR3 DRAM channel as shown in the following table. Software must not attempt to configure “DCT1” or “DCT2”.

So the very first thing is to attempt to configure/use the DCT1 and 2 controllers ;D
Bad thing is, that FM2+ support only 2 channels, so… useless. But they could be meant to utilize the GDDR5M rams, but that idea was sadly canceled 😞 (and that could give us almost the performance of DDR5 for around DDR3 price 😞 )

So maybe AMD do some BGA version of Kaveri that then can bypass the FM2+ socket limit, or they planing to relase FM3 with 4 channel support and then activate these controlers?

So there are something that is not even unlocked in the A10-7850 and A10-7700 CPUs  😉

Photos of the core and article in czech:
http://diit.cz/clanek/kaveri-obsahuje-ctyri-kanaly-pametoveho-radice

Ah I love hearing/reading speculations  😄

Yea, well… me too. But there is a truth in this, and that is that in official AMD material they admit that there are 4 memory controllers. So that was what we know for sure. But why it was this way is only a speculation about the GDDR5M rams... or possible future support of 4 channel ram access.

Regardless of this, we both know, that any APU will benefit greatly from own dedicated memory or GDDR5M rams. AMD must know it as well, witch is why they integrated these transistors there. They will not be doing it for nothing, that will not make sense.
The big regret what is there is the end of these GDDR5M rams with the end of Elpida. Yes, their rams aren't have much overclocking headroom, witch is why people did not exactly like them, but the GDDR5M rams promised a soon-to-be performace increase and that is what we all are after, right?

Easpecially people trying to ditch out every % of performance from a bit old HW, now abusing friend mainboard with Opteron 148 at 2750MHz 😉

Kaveri - performance is directly tied to memory speed and scales accordingly:

This somewhat confirm my firm belief, that memory speed are the main problem and the faster the rams, the better. In Kaveri case, the difference is extreme - obviously the GDDR5M rams are greatly missed, as the performance could be much better with them… 😞

  • 11
    Posts
  • 5297
    Views