AMD Bulldozer Core Patent Diagrams

the memory controller has been reworked and is now much better.
this shows off on llano already, you can see the results here, compared with an athlon II X4. (you may skip to the real world tests)

http://www.hardware.fr/articles/837-2/architecture-cpu-nouveau-controleur-memoire.html
huge wins from faster memory with the new hardware, especially at ddr 1600.

 

We already knew Llano is a bandwidth-starved architecture. BD on the other hand, does not have a graphics core (yet) embedded.

 

Just a question, from the look of things will FX-8150 be able to outperform the i7 970 with the two extra cores ? I was going to go with BD but with all the delay, I was thinking of picking up i7 970 instead, since the price of that has come down somewhat.

 

It will highly depend on what you plan to run on that CPU. BD should be pretty awesome at running non-AVX code but loose quite a bit of power (in terms of perf per int core vs i7) once you start using it.

 

I wouldn't call that huge wins, except the synthetic scores. But yes there's definitely a difference, I exaggerated a bit. (In theory the difference could be smaller with Phenom II cause of the L3 cache.)
In any case, I don't get why people are so interested in memory overclocking. So you spend another 100$ over regular ddr3-1600 so you can have a 50% higher memory speed for another 3% performance? The ROI just isn't there. (Now if you've got a IGP scaling is much better but buying premium memory is still not worth it as you could just get a discrete card instead.)

 

Is AVX going to matter during BD's or i7's lifetime in the consumer world? I'm guessing this will be another SIMD set that is useful in consumer enthusiast world with apps like video encoding but not so much elsewhere.

 

That's most likely true but if you do want to encode videos then that's something you should consider

 

How do you find out if apps has AVX supports ? Is there a list somewhere ?

 

At this moment I'm pretty sure there are very few apps that have any kind of support for it but the number will surely change as AVX-supporting CPUs get more common.

 

I'm pretty sure BD will destroy i7-970 in AVX considering i7-970 doesn't support it

 

Surprisingly, the current version of AVX doesn't support 256bit packed double precision FP dot products - at least it's not mentioned in the AVX intrinsics listing

--edit--
However, I do see a vdppd instruction that should (judging by its name) perform this very operation. Does anybody know why this instruction is curiously omitted from the intrinsics?

 

No 256 bit integer operations in AVX... Makes me sad

But 1024 bit wide SIMD in future AVX sounds really promising. 32 parallel floating point calculations... is exactly the same as NV SIMD warp width. Now if you add some eight way SMT (like Niagara 2) and we have pretty much reached GPU level of parallelism in a CPU. But 16 cores with HT (32 threads) should be ok as well

 

True, older i7's don't support it indeed. I was thinking of the sandy bridge based ones

 

http://www.xbitlabs.com/news/cpu/di...eries_Microprocessors_Show_Up_on_the_Web.html

still needing some info about how turbocore2 works

 

How does Turbocore 2 function again ? Will it always use near max TDP regardless of the amount of threads that are going on ?

So if I am getting this right BD will be faster in non-AVX applications? I am not aware any of my apps having or will have AVX support. Beside apps, will there be many games that support AVX ?

 

Versus the i7-970? I'd expect the 970 to beat BD across the board.

 

You might be disappointed

 

So far there's nothing to suggest that I will be, what with the lack of quantitative information, delays and the absolutely terrible performance/watt of LLano's CPU component not giving much hope that GloFo's 32nm process will be spectacular.


And the 970 is an extremely formidable opponent, with more than 30% better single threaded performance than a K10.5 core at the same clockspeed while still having the throughput of about 9-10 K10.5 cores.

 

What does the pricing tell you? Factor in that AMD is really not a charity, nor is it in a position where it can truly afford to play pricing tricks.

 

First off all what pricing do you have in mind? Supposed leaked pricing from DH or Chinese website taking preorders on SKU's which will never be available in retail such as FX8130P? I haven't seen any official pricing structure from AMD for upcoming Zambezi models other than slide from last year with product positioning placing FX8 anywhere between i7 2500K and i7 2600K+. If we take it as a reference point then in worst case scenario FX8 should be somewhat on 2500K level of performance which by itself would put it in at least some tasks above i7 970. We can agree on that I think?

Secondly if we believe any of the 'leaked' pricing we also should believe some of leaked scores. In this case we had ES sample of BD scoring 27k-28k in CinebenchR10 MT x64 at 3.2GHz. From Anand table I see i7 970 scoring in same test 25k. This alone puts Accords post to rest.

Besides Accords wording just begs for this kind of answer, because I can even now find a task where Phenom II X6 will perform on par or even slightly better than i7 970. Of course majority of applications is performing much better on i7 970, but I feel it's silly to come here and say BD will loose to it across the board.

Now I will raise my hand and say: If I'm wrong I will sincerely apologise to Accord1999 on this forum, but if he's wrong I would expect him to do the same to everyone reading this thread.