AMD "Piledriver" refresh of Zambezi - info, speculations, test, fans

[informal]

Considering Trinity is shipping from OEM's, there will be no magic fairy dust with performance improvements from newer BIOS.

No need for magic dust or miracle BIOS. Piledriver is faster than Bulldozer at the same clock by very solid 10-15%. This is with no L3 cache. So Piledriver core does things more efficiently while clocking 10% faster. A very nice follow on to bulldozer in my opinion. Raise the clock and increase the IPC= success. SR core just continues on this road,more IPC and more clock with more cores.

[Smartidiot89]

Richland is the Trinity core, dual core A4 part...

I am corrected, no wonder the name sounded all too familiar Well at least he got the manufacturing technology and CPU cores right

No need for magic dust or miracle BIOS. Piledriver is faster than Bulldozer at the same clock by very solid 10-15%. This is with no L3 cache. So Piledriver core does things more efficiently while clocking 10% faster. A very nice follow on to bulldozer in my opinion. Raise the clock and increase the IPC= success. SR core just continues on this road,more IPC and more clock with more cores.

I didn't imply so either, I think Trinity is great for it's designated market. L3 cache did very little for Bulldozer, during the briefing AMD explicitly stated it was "a bonus" for the consumer market and that it were there simply cause the same die was used for servers.

[Olivon]

http://www.expreview.com/20232-1.html

[Oese]

there's an inconsistency looking at pov-ray... Test in forumdeluxx show number close to llano (600/200).

Rest indicates better integer+gpu perf, fpu is still not used correctly because of no fma software. Single thread fpu is faster then multicore cause of higher clock, but only 2 fpu units...

I see mostly optimizations in frontend, where integer was limited before. FPU perf wasnt limited by frontend (needs fma to be fully used) and doesnt improve much therefore.

[EniGmA1987]

Anyone know if the Vishera FX-8350 still expected to launch in October? Or has it been delayed now?

[Oese]

to my knowledge, there's no official delay.

i wouldnt expect them on 1st october tho :p

[EniGmA1987]

Ya I figure a launch in October would mean retail availability late October or early November. I just keep getting the itch to buy something new but the only thing I would buy is a 2011 socket system and I really dont want to spend that kind of money for no real reason. SO I keep trying to find a reason to just hold off and keep the system I have now, Vishera would be what I am looking to play with on my current system

[andressergio]

[cesariuth]

Kaveri with 4 GHz, 512 SP (8CU) ~200 mm² and ~1.8B transistor...

Tabla.jpg
http://translate.google.com/translat...ity%2F&act=url

[tbone8ty]

^^^^ this...nice read

[Smartidiot89]

Old
http://www.nordichardware.com/news/6...n-hd-7750.html

[FlanK3r]

OK, my simulation of FX-8350, if will be clock at 4 GHZ/4.2GHz stock...

Cinebench R11.5: 6.64b


x264 HD 4.0 (no patch for FX) 40.6 FPS


Superpi 1M

[Oese]

explain?

this looks strange xD

[The Stilt]

Hmm...
Can you explain how did you come to this conclusion?

[FlanK3r]

I know, Vishera will have better IPC a bit (maybe 5% clock to clock), default clocks of FX-8350 looks from internet sources (but I dont know it) 4GHz and 4.2 GHz for turbocore. So, I simulated it at 4070 MHz with NB 23xx MHz for a bit better perormance. But as I said, it is only simulation and speculations about final clocks

[fakhrain]

i wonder if the L3 cache sizes have significant impact on max performance on the upcoming Vishera

the quad core parts have different L3 cache sizes (FX4350 12mb, FX4320/FX4300 8mb)

[FlanK3r]

this is wrong...all FX has the same L3... 12MB is L3+L2 for this model

[Olivon]

For those who missed ... 1T test with same freq. :





Almost 15% better, very good !

[FlanK3r]

3DsMax...hm, big diference! and 4c/4T vs 8C/8t!

[Olivon]

3DsMax...hm, big diference! and 4c/4T vs 8C/8t!

By running a single-threaded workload like iTunes, we could neutralize the difference in core count (though, if anything, FX could have benefited from its 8 MB L3).

Turning off two of FX-8150's Bulldozer modules gives us the opportunity to run a threaded workload like 3ds Max without slanting the result toward Bulldozer. And once again, the Piledriver-based APU wins by roughly 15%.

http://www.tomshardware.com/reviews/...0k,3224-2.html

[BeepBeep2]

Its a start. 15% Performance / Clock gain is great. Not exactly Phenom II levels but closer. Hopefully upwards of 20% improvement over BD after L3.

[Oese]

Its possible, at least. Combined with lower consumption and higher clocks it will be a nice upgrade for BD users xD

I cant await to test it by myself, anyway, still, i didnt run into cpu limited scenarios with my 4,6ghz turbo bd, and consumption is less then with pII x6, so i'm still satisfied (as long as you dont play super pi and cinebench, that is )

Nevertheless AMD needs to show that they do not only have good ideas, but also "execution".

Other fact is that with haswell, FMA will be used more frequently, so hopefully, at last, the shared FPU will start to show its muscles some time...

[haylui]

after so long finally, Phenom II at 4.0GHz stock..

[fakhrain]

this is wrong...all FX has the same L3... 12MB is L3+L2 for this model

sorry, i misquoted. was actually referring to the total cache mentioned in the article on fudzilla http://www.fudzilla.com/home/item/27...parts-in-q4-12

[FlanK3r]

New Athlons II based at Piledriver core (Trinity with iGPU disabled). Of course are this all for FM2 socket...

Athlon II X4-730: quad-core, 2.80 GHz, 4 MB total cache, 65W TDP
Athlon II X4-740: quad-core, 3.20 GHz, 4 MB total cache, 65W TDP
Athlon II X4-750K: quad-core, unlocked BClk multiplier, 3.40 GHz, 4 MB total cache, 100W TDP

http://www.techpowerup.com/169558/AM...2-Package.html