Eposide 5: Retail Version PhenomX4 Got Tested(Cine & 3DMark06)

[cookerjc]

Episode 1 ：Phenom X4 with SuperPi

Episode 2：Phenom X4 with Crysis Time Demo

Episode 3：Phenom X4 with 3DMark06

Episode 4: Phenom X4 with Cinebench

Episode 5: Reatail Phenom X4 with Cine and 3Dmark06

These days we manage to take some tests with an OCed engineering sample Phenom X4, And received lots of feedbacks. Some of you think an ES version Phenom X4 can not represent the upcoming retail version so today we put some score of two retail versions here.

http://www.expreview.com/news/hard/2...83d6863_1.html









================================================== =============================

Why CPU-Z show 5-5-7-18?

We set all the memory to "By SPD"(by SPD= 5-5-5-18). But when we are using the Phenom X4 platform, CPU-z wrongly detected the timing as 5-5-7-18. Please wait for the CPU-Z's further update.

Why CPU-Z shows Phenom X4 only have single channel?

That's because PhenomX4 have two memory controller, one controller controls one memory module. In this way when system performing multiple task, the performance may boost little higher.

[Epsilon84]

funny, you have a phenom but you cannot find a good pair of memory modules

I think finding such bad memory is way harder than finding the actual phenom so congrats

single channel ddr2@336Mhz 5 5 7 18

wow

Lighten up dude, at least we're getting retail stepping results.

Phenom utilizes dual 64bit memory controllers, hence CPU-Z reports it as single channel per module.

Also, Cinebench does not scale much with timings.

I'm sorry if the truth is too hard to bare.

[nemrod]

funny, you have a phenom but you cannot find a good pair of memory modules

I think finding such bad memory is way harder than finding the actual phenom so congrats

single channel ddr2@336Mhz 5 5 7 18

wow

look at this one, it's far better memory...
http://www.oohashi.jp/c-board/file/3....2G-220x10.png

[cookerjc]

updated with some FAQ

[Gambit_2K]

Seriously give us some game benchmarks other than Crysis and memory @ 1066 4-4-4-12. Can't be that hard if you can get the Phenoms that easily. Who cares about 3mark? And maybe som screens of the Phenom retail vs. Phenom ES?

[knightwolf654]

i like it! now GIMME!!!

[EvlUndrWareNome]

not too bad I guess.

[GoThr3k]

memory at 1066 wout be better,but its preview stuff,so i like every info i can get
about a week we will have reviews enough

[rozzyroz]

thanks for sharing cookerjc!

[LowRun]

The infos you bring here are appreciated but i don't see the need for a new thread in the news section every time you bench a new app, why not just make a thread in the AMD section and have all the other threads merged

[happychappy]

How long till the ing starts?

[hardrail]

How long till the ing starts?

Pretty soon.

Why did AMD abandon the idea of a large L2?

[freeloader]

Maybe we'll see a larger L2 cache when they transition to 45nm. Who knows?
I was just thinking about how large the die size is at 65nm and maybe that's why the cache sizes are what they are.

[Kingcarcas]

How long till the ing starts?

It already has

[justapost]

Thank you for the screens, now it's devinite these are B2's

<archived>How about a comparison between ganged and unganged mode for us non believers? </archived>

[nemrod]

How about a comparison between ganged and unganged mode for us non believers?

Non believers of what? Do you expect so different results between those one (phenom) and those obtain by kyosen on K10 (B1) barcelona? Or you don't believe to that kyosen has a barcelona?

[justapost]

[nemrod]

I was just kidding so i will not take that into any deep.

Sorry, I have use the gun to quick. I'm just a little angry by people claiming it'is fake when someone share kindly such kind of results we all waiting.

Source


So using 4 modules should result in two DCT's with an 128-bit interface each. That's more interesing than ganged mode.

I have already seen your post in the Kyosen thread

[stm]

SM 2.0 Score 4208 Marks
SM 3.0 Score 4725 Marks
CPU Score 1954 Marks
are my results with a 4200+ x2 @ 2.5mhz and stock 88gtx. is that right for the phenom to have around the same SM scores, but way better CPU score? i always thought the 88gtx was way limited by my CPU, but the phenom doesnt seem to show improvement in that area?

[justapost]

Sorry, I have use the gun to quick. I'm just a little angry by people claiming it'is fake when someone share kindly such kind of results we all waiting.

np

I have already seen your post in the Kyosen thread

I thought one DCT with two DRAM's will act at 128bit width, but as I reread the manual i post i found that it will only be 64bit.
I guess kyosen and stephen have a better bandwidth because they can use node interleaving on they dual-board.

[SADS]

SM 2.0 Score 4208 Marks
SM 3.0 Score 4725 Marks
CPU Score 1954 Marks
are my results with a 4200+ x2 @ 2.5mhz and stock 88gtx. is that right for the phenom to have around the same SM scores, but way better CPU score? i always thought the 88gtx was way limited by my CPU, but the phenom doesnt seem to show improvement in that area?

yes, SM 2/3.0 are Graphics card tests, SM = ShaderMarks i think?
and thats why you are getting the same score there as you have the same card as the one that was used in testing the phenom

and the CPU score is obviously just that

[zanzabar]

SM2/3 are shader model 2 and 3 and they are rendering instruction sets

[TigeriS]

Can anyone compare Phenom's multi tasking performance with Conroe's? And I don't mean multi threading but more like running 2 different benchmarks at the same time.

[Inteleron]

3400+ in 3DM06 at stock settings. Sure beats my E6300 @ 3.5GHz

[nemrod]

I guess kyosen and stephen have a better bandwidth because they can use node interleaving on they dual-board.

If you think they have better bandwith due to cinebench score:

On the first kyosen board:
http://222.151.153.254/c-board/file/...t2350_2.4G.png
Kyosen has at 2.4GHz for cinebench:score 1 cpu 1984 and multi: 7709
here for 2.3 GHz 1cpu: 1930 multi 7391
As frequency scaling is great on cinebench:
1 cpu: 1930*2.4/2.3=2013 multi cpu 7391*2.4/2.3=7712 a little bit better for single equal for multi

I have not seen 32 bit cinebench results on the dual-board, only 64-bit cinebench results. (And there is huge difference beetwen 32 and 64 bits, about 14% improvment for classical cpu like Q6600)

So I don't see where you see better bandwith?