AMD Presents Kaveri in Taipei - Memory Overclocking Still Key, So Far OC Not As Good As Richland

Those following us on social media will probably already know this, but yesterday AMD came to Taipei, Taiwan to present the new Kaveri micro-architecture. Since this was the last stop of their business trip, there's not much new to tell. From a performance perspective HSA seems to be interesting, although it's still too early to see the real effect. I enjoyed seeing references to overclocking on the marketing slides, especially highlighting the benefit of increasing the memory bandwidth. With HSA, system memory bandwidth is again a bit more important.

They didn't mention any early overclocking results, but during our chat behind-the-scene it became clear that AMD too was a bit underwhelmed by overclocking potential of the Kaveri APUs. Currently the maximum frequency on LN2 (internally) is about 7GHz, which is far away from the 8GHz the Richland APUs could do. From what I gather, the potential problem might be the new 28nm silicon. Because the CPU part is using a different process than the GPU part, mixing them together on one die is complicated. Early Q3 (June/July) they expect to have better yields after optimizing the process. Maybe the Kaveri refresh will bring us more overclocking joy.

So far the highest CPU-Z frequency result in our database is Bboyjezz's 5498.72 mhz with A10-7850K at 5498.7MHz (coldbug at -65°C). I believe AMD's been holding off sampling APUs to overclockers because of the current poor overclocking capabilities. From various sources we've heard it's difficult to get ahold of samples. I'm sure this will change in the near future, though!

In any case, for those who are interested in seeing what AMD tells the press, here are the two press decks for your information. There are three slides related to overclocking I've highlighted for you.


Belgium Massman says:

Might as well go buy one in the store :)

Netherlands rsnubje says:

Don't we all just miss the days of Llano, when the memory would actually be able to run really fast.

Germany der8auer says:

APU - no thanks.

Indonesia Lucky_n00b says:

From the 5 chips I have tested in the office, don't waste the LN2 for CPU-Z validation. Better spend it on IGP 3DMarks, Kaveri is strong there ;)

FlanK3r says:

der8auer said: APU - no thanks.

Roman, if there will be 7 GHz sample, you will be surprised...Target 32M AMD WR...I broked my personal air 32M record with Kaveri and with better RAM I broke my 5400 MHz 32M with FX-9590 :)...Kaveri needed only little over 4800 MHz for it :)

Germany Luebke says:

...with Kaveri and with better RAM...
maybe because of the better RAM? ;)

FlanK3r says:

with better RAM at Kaveri than in first try :) (singe sided vs dual)

Belgium Massman says:

Do you still need the bulldozer patch?

FlanK3r says:

I dont know, if some new unpublished version will working with Kaveri (richlands day version not), but for Richland I need it (F2A85-V Pro)

Belgium Massman says:

Yeah, of course you need it for Richland. But I was wondering because you mentioned Kaveri is so much faster, if it also needed the patch.

FlanK3r says:

Kaveri is much faster exmaple in wprime too. There is nice boost more than 100s :) - both stock. Clock to clock more :)

Please log in or register to comment.