K8 vs K10 on 790FX - Results

[mAJORD]

Firstly. Big appologies for the differences in RAM timings between the Windsor and Agena in the following.

Due to problems with this board I could not run CAS4 on the K8 on this bios at least. Also I should have by rights used a 10x multi to keep RAM frequency the same.

Instead though I've clocked the K8 20Mhz faster in an attempt to keep things as even as possible. In the end though the aim of the exersize was find areas were K10 majorly outperforms K8 not nitpick at anything less than 2%. Anyone who's spent time with Phenom on the current BIOS will no what a pain the arse it really is atm to run overclocked. Hence someitmes it's just too hard to get things exactly how you want them. I will keep at it and RERUN THE BENCHMARKS with identical settings.

Now, the focus here is on 1-2 threaded Apps only, with the aim of giving people a better idea whether an X4 is for them at the current clocks and rev's, especially those with current X2s I

THIS IS NOT AN INTEL VS AMD THREAD. Especially given it's single-dual thread nature, where we allready are pefectly aware K10 does not have the IPC of Core for MOST desktop app's (but not all, I will say that) Trolls, please See the troll thread, not this one. That goes for anyone

Also if someone with a good brain can have a check over my % calcs, we all make mistakes.




Apart from the minor RAM timing differences, all tests are on the same machine, same install. All run on the Phenom first. Some repeated at the end to ensure no degredation of the OS had occured during the install of some programs.

Some I sitll have to do for K8 Include Superprime and the new release Wprime. stay tuned for that.

Setup:

Mainboard: Gigabyte MA-790FX-DQ6 Bios F2H
RAM: Crucial Ballistix 1066
Phenom: 4-4-4-12 487Mhz
Athlon: 5-4-4-12 440Mhz
Video: Gecube HD3850 256mb @ 750/1000

CPUs:

K8 Windsor 1Mbx2 L2 @ 2695Mhz
K10 Agena @ 2680Mhz

OS
Windows XP32 SP2 +crit updates


On with the results...


Super Pi 1M




Phenom: 27.687s
Athlon:30.328s
K10 +9.5%

*Note, Winrar bench not run single threaded- discard. Will rerun along with others

-------------------------------
Nuclearus MC:



Phenom:
ALU: 4973
FPU: 5116

Athlon:
ALU: 4356
FPU: 5067

ALU K10 +14%
FPU K10 +0-1%

--------------------------

Lightmark 2007



Phenom: 269.3
Athlon: 233.7

K10 +15%

--------------------------------
LAME 3.7 MP3 192Kbs - Standard Compiler




Phenom: 2:06
Athlon: 2:27

K10 +16%

-------------------------------

LAME 3.7 MP3 192Kbs - Intel Compiler



Phenom: 1:42
Athlon: 2:01

K10 +18%

------------------------------
FLAC CDex 1.70



Phenom: 4:17
Athlon: 5:00

K10 +16-17%

-------------------------------
CineBench R10 32bit


Single Thread:
Phenom:2372
Athlon:2314

K10 + 2.5%

---------------------------------

3Dmark 06:



Phenom:
SM2.0: 4309
SM3.0: 4780

Athlon:
SM2.0: 4142
SM3.0: 4578

SM2 K10 +4%
SM3 K10 +4%


*important to note this Test is not overly CPU bottlenecked. more interest factor


-----------------------------------

Aquamark 3



Phenom:
GFX: 22566
CPU: 13422

Athlon
GFX: 19980
CPU: 12971

GFX: K10 +13%
CPU: K10 +3.4%
Overall score K10 +8%


TO BE CONTINUED..

STILL TO COME

*Video Encoding

*Archving

*SSE heavy Stuff

HOPEFULLY some real games if I can get some help on DEMO's I can use, as I don't have much unfortunatly


*** - UPDATED: 7dec 07 ***


Notes:

RAM timings vary a bit on these ones, I had initially got them both running at CAS4 finally, but then during a CPU swap I've managed to end up with Phenom on Cas 5 for some of these tests, Frequencies now identical though. So Phenom slightly disadvantaged this time around. Again though we're talking a percent at the most. K10 especially much less affected by the odd timing difference.

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

x264 Benchmark



Phenom:
PASS1:89.00
PASS2:17.58

[106.58]

Athlon:
PASS1:75.13
PASS2:17.97

[93.1]


K10 +14.4% overall

-------------------------------------

Systool



Phenom: 263sec
Athlon: 237sec

K10 + 10.9%



notes: CAS4 on K8

----------------------------------------

]Wprime 1.55



Phenom: 29.562
Athlon: 32.14

K10 +10.3%

-----------------------------------------
POV-RAY 3.7 single thread



Phenom: 394.8
Athlon: 370.5

K10 +6.55%

------------------------------------------
ScienceMark



Phenom: 1687
Athlon: 1512

K10 +11.6%

-------------------------------------------



***Lightman's Results****


Lightman has also kindly let me merge his results from the thread into this post, Enjoy:



I did some tests for you guys simulating Phenom X2 2200MHz with DDR-800 4-4-4:


CPU Queen 4395 vs 4262 of X2 4200+ 2.2GHz


CPU PhotoWorxx 26375 vs 17902 of X2 4200+ 2.2GHz


CPU ZLib 29487KB/s vs 26238KB/s of X2 4200+ 2.2GHz


FPU Julia 4660 vs 3742 of X2 4200+ 2.2GHz


FPU Mandel 2308 vs 2114 of X2 4200+ 2.2GHz


FPU SinJulia 5671 vs 5872 of X2 4200+ 2.2GHz


Cinebench 2003 x64 + SuperPi




POST #2

[Brother Esau]

Good work brother Nice to see some people posting up some AMD stuff Too bad these 6400+ don't seem to O.C much but the bandwidth on the new AMD platform ROCKS!!!!!!!!

[Morais]

[Aussie FX]

That's where you've been hiding then...

Good work mAJORD.

BTW I just killed my mobo!

[EnJoY]

It makes you wonder what AMD was doing that whole time while they were "developing Barcelona"... It seems Intel achieve more in going from Conroe to Penryn than AMD has going from Windsor to Barcelona. This is what AMD brought to the next-gen fight?

[KTE]

Thank you for the work mAJORD.

Some details you left out I'd like to now please:

What volts were you running, HT, NB, CPU?
Whats max ref. HT, HT speed and NB speed you reached?

My HT works at half the speed of that and NB is much lower too.

Can you run some Sandra, EVEREST, Systool 16M Pi, ScienceMark and POV-Ray please. I'll need to compare later.

Games would be good for those who are interested because I have none here but HL2 and UT as I haven't played since coming here.

My setup at those speeds are giving lower results because I can't run dual channel, high NB or high HT. Bugged out very badly.

[JS1234]

Nice post! Look forward to more!

[mAJORD]

Aussie: What the hell? how?

KTE, will do.

Volts are as shown in CPU-Z (reporting correct) 1.31 on air.

243 is the max HTT I can reach, it's a hard wall. F3A doesn't help. I hope we can (all) get around this issue as I'd love to put all this cooling to work As it is i have no reason to use anything more than the dinky little Arctic64 pro..

[Brother Esau]

It makes you wonder what AMD was doing that whole time while they were "developing Barcelona"... It seems Intel achieve more in going from Conroe to Penryn than AMD has going from Windsor to Barcelona. This is what AMD brought to the next-gen fight?

I know its quite disappointing isn't it Alex?

[Aussie FX]

Aussie: What the hell? how?

PM sent.

[neo2005th]

But C2Q is about 10% faster clock for clock than Phenom and more cheap.

And K10 is about 15% faster clock for clock than K8.

Penryn performs about 15% more clock for clock than K10, and 30% than K8.

Where is the AMD classic price-performance?

[duploxxx]

It makes you wonder what AMD was doing that whole time while they were "developing Barcelona"... It seems Intel achieve more in going from Conroe to Penryn than AMD has going from Windsor to Barcelona. This is what AMD brought to the next-gen fight?

here we go again.........

not that difficult to increase that much performance after so many years of netburst crap and from con to pen its just the apps that benefit from additional cache and were already constrained in conroe on fsb. (and the sse4 offcourse)


for the k10 part
k10 first original design was killed because the market was not ready for it.

second k10 design (this one) has some advanced memory control, shared cache and native but for the rest the overestimated the increased performance by that i think. if l3 would be as fast as and as big as intel cache you would see a much better int performance, lets see what shangai brings.

and it was late to the market because they had to delay 65nm because of first soi issues (long time ago, yields ) and native design.

maJORD thx for the info, but aren't you the first to report k8 issues on the 790fx chipset? or did i miss some posts...... i know there were phenom bios issues.

[mAJORD]

duploxxx

the CAS 4 not sticking is the first problem i've had with it on K8.

[Lightman]

Fantastic work mAJORD! I'm jealous of your CPU Mine can do 2.5GHz stable

I will throw a bit as well if you happy with that:


You can see Clock difference between exactly same score of 508 points in Single threaded test for Phenom and K8 Brisbane

[swedkg]

k10 first original design was killed because the market was not ready for it.

care to elaborate?

[knightwolf654]

f l3 would be as fast as and as big as intel cache you would see a much better int performance

AMD cpus do not need the extra cashe becuse the have a on-board memory controler. but it would be nice to see it a bit faster. as well as the hole thing with SOI fixed.

[AliG]

agreed, this is ridiculous! Conroe had huge gains over netburst and k8 alike and that's what a new architecture should offer, and from what we've heard, nehalem will be to conroe as conroe was to netburst so another massive gain, but only 1-20% faster for k10 over k8? That's pathetic! Especially with all the delays we experienced.

I don't care what amd does to fix this, but something needs to be done. If bulldozer fails, amd might as well force Hector to stay and focus on how they can lose even more market share and stock value

[Nazu]

Netburst->Conroe was a bit different cos Netburst sucked so much that it was easier to get huge performance gains. K8 was much better than Netburst so it's not so easy to get that big performance gain. Though I agree that K10 isn't exactly a succes.

[64dragon]

great test majord

[AliG]

Netburst->Conroe was a bit different cos Netburst sucked so much that it was easier to get huge performance gains. K8 was much better than Netburst so it's not so easy to get that big performance gain. Though I agree that K10 isn't exactly a succes.

still, I know an intel engineer and he's been saying for a long time now the increase of conroe-->Nehalem will be similar to that of netburst-->conroe. I know intel has an awesome amount of resources, but still, amd should easily be able to pull 20% across the board over k8 especially with the 4 years it took for them to design k10, and if they wanted that market share and actually be able to call their product an innovation, k10 should have been at least 15+% superior to conroe across the board, and a lot stronger in areas where amd shines like fpu

[Motiv]

very good work majord. Finally get to see some interesting result. A 15% gain over the K8 looks about the main in single/double threaded stuff.

Now, if they can get the speeds ramped and keep the prices down it could be a decent cpu on the desktop. It certainly looks like AMD went out of their way to produce a server chip though in my opinion (which at higher speeds seems to be very competitive).

[darkorb]

clock for clock its faster, but its also adding 2 more cores. k8 vs k10 that is

maybe try rebenching both at stock speeds?

[NickS]

Indeed, I'll be sitting back to see the results that follow! *excited*

Nice work mAJORD

[informal]

still, I know an intel engineer and he's been saying for a long time now the increase of conroe-->Nehalem will be similar to that of netburst-->conroe. I know intel has an awesome amount of resources, but still, amd should easily be able to pull 20% across the board over k8 especially with the 4 years it took for them to design k10, and if they wanted that market share and actually be able to call their product an innovation, k10 should have been at least 15+% superior to conroe across the board, and a lot stronger in areas where amd shines like fpu

First,there is almost no chance that that Nehalem will be faster than Conroe as much as Conroe was faster than Netburst,in genereal "average" case.
Nehalem is a Core based uarch. with some (yes radical) changes to system arch as IMC and QuickPath.Conroe/Penryn already have great ALUs and i can see Nehalem some 15% faster there.In FP there could be some greater gains since it will be "2 threads per core" capable design (HyperThreading),but this could be probably seen more in some server applications rather than client(desktop) ones.Intel will lower mem. latency and improve Core uarch overall,but to get anywhere close to 50% faster(in what?SSE?int?) as it was general case in Netburst Vs Core2,there is little chance.Like i said, in some mem. constraint server applications where mem. bandwidth is crucial and where good scaling to more cores is needed,Nehalem could be that much better than Core2.Spec_fp(rate) is a good example where "40% better than Core2" could be achieved.But in this case,we have K10 beating Core2 by as much as 40%.

@mAJORD

Impressive work man .
What are your observations wrt the NB speed(ie. 2 cases where we have same clock for the cores but with different clocks for NB)?Does L3 latency drop (happens with NB clock increase) help in any measurable way?

[trueblue]

still, I know an intel engineer and he's been saying for a long time now the increase of conroe-->Nehalem will be similar to that of netburst-->conroe. I know intel has an awesome amount of resources, but still, amd should easily be able to pull 20% across the board over k8 especially with the 4 years it took for them to design k10, and if they wanted that market share and actually be able to call their product an innovation, k10 should have been at least 15+% superior to conroe across the board, and a lot stronger in areas where amd shines like fpu

Actually, Barcelona/Phenom is quite good at floating point. Take a look at this link:

http://www.cisl.ucar.edu/dir/CAS2K7/...torricelli.pdf

A 2 GHz Barcelona is beating a 2.33 GHz Xeon by 30-50% in floating point when using multi-core, heavy memory type engineering applications such as CFD (Fluent) and structural mechanics (LS-DYNA). Take a look at the differences when all four cores are being utilized. SuperPi fits into the cache of Core-2 Duo, hence it will always have better SuperPi scores.

Funny, I have not seen SciMark 2.0 benchmarks in any of the website reviews.

In the type of applications used by servers and engineering analysis, AMD is quite competitive.

The 10% difference in integer performance is nothing. I recall when different steppings of K8 (Venice vs Winchester) gave almost this much improvement. The big advantage of Core 2 Duo is their excellent over-clocking ability. I am sure that AMD will improve in this area, but I am not sure they will ever beat Intel in overclocking.