Intel Q9450 vs Phenom 9850 - ATI HD3870 X2

[JumpingJack]

Very interesting..... the memory performance rises very nicely, a 10ns reduction in latency + a nice increase in bandwidth just by increasing the NB speed 20%.

Can you run any gaming tests comparing 2.0GHz / 2.4GHz NB speeds?

I really wish review sites were as comprehensive as you

I will repeat Xplitz info below.

[Boschwanza]

i already said that.

absolute min fps is worthless, same as max fps, what really counts is the avg. fps.

If you record a certain duration, you could try to get the median min fps.

Or just use V-Sync at 60 FPS/60Hz to limit overall high FPS to get a good average of all the median/low fps. Thats the way i would do that.

[JumpingJack]

And another strange thing.

I overclocked from 3100Mhz to 3316Mhz and my MIN FPS jumped from 20 FPS to 85 FPS???????????
Lol strange.

Edit: seems to be a windows issue. did every test after a fresh windows boot and wait 3 minutes. Repeated the tests without reboot later
And enabling the red dot in AOD gave me boost of min fps between 5-8 fps which is alot.

What graphics card? And I cannot go over 3.0 GHz, could you redo a run a 3 GHz to check, I will post the same shortly.

EDIT: I think there might be a windows memory leak somewhere, after 4 or 5 runs through the benchmark, I had issues to.

[JumpingJack]

here is a quick and dirty Phenom/Crysis clockspeed/NB speed study....

I used the benchnark tool found at Guru3D, the setup of the system has been documented elswhere, if you want a link to full system specs let me know (I posted that several pages back). In short, the data collected here was done on a Phenom 9850 BE + Asus M3AMVP-Deluxe + DDR2800 C4 + ATI Radeon 4870 X2 with catalyst 8.8 drivers (EDIT: oops, my bad... looks like the SPD is defaulting it to C5 on the DDR2). The benchmark was ran to focus on the CPU, 800x600, low quality settings. I ran at 2.5 GHz and NB of 2000 and 2400, then again at 3.0 GHz with NB of 2000 and 2400 MHz. The results (screen shots) and output are posted below:

9850BE@2.5 GHz, NB=2000 MHz ave = ~ 112-113 FPS

9850BE@2.5 GHz, NB=2400 MHz ave = ~ 115 FPS

9850BE@3.0 GHz, NB=2000 MHz ave = ~ 125 FPS

9850BE@3.0 GHz NB = 2400 Mhz ave = ~ 128


I have also attached the log/output below (this provides all the details that screen shots cannot capture).

For a 400 Mhz increase in NB speed, the increase in frame rate for this configuration is 2-3 FPS, the minimums are random, take note there is some texture loading occurring mid game as can be seen by the first run always being lower on the first loop. A 500 Mhz increase in core clock speed improves FPS by 10-15 FPS on average, gain in this configuration. Remember that this is running the GPU test in all low settings. I am looking into setting all the CPU related settings (physics quality and particle quality to high). Finally, a combination of 400 Mhz NB and 500 Mhz core clock will yield 15-16 FPS improvement which is > 10%.

I will turn my attention next to a 2.5 GHz clocked 9650 and 3.0 Ghz clock 9650....

Jack

[JumpingJack]

Ok, here is the QX9650 clock experiment... like the Phenom above, I ran the same crysis run (3 runs of 3 loops demo's on the stock CPU bench, all at low settings). The configuration is as identical as I can get it -- but, of course, MB and CPU are different -- Asus Rampage Forumla + QX9650 + DDR2-800 C5 + 4870X2, details of the config have been linked previously.

QX9650@2.5 GHz ave = ~ 152-158


QX9650@3.0 GHz ave = 177-183


So for a 500 MHz increase in clock, the QX9650 gains roughly 25 FPS ...

125/112 for Phenom is roughly 1.116 or 11.6% scaling for Phenom 2.5G->3.0G (NOTE: Comparing stock NB frequencies)
177/152 for QX9650 is roughly 1.164 or 16.4% scaling for QX9650 2.5G->3.0G

This needs to be qualified under the understanding of the configuration, these results are not going to be indicative of every CPU/4870X2 as chipset, mem speed, etc. change this scaling as well ... not to mention we are forcing an odd environment by running everything at very low res and low settings.

As with the Phenom experiment above, I attach the raw log file output below.

Jack

[MomijiTMO]

Wow thanks for those. Very interesting.

[xPliziT]

Jack i am going to re test now with 3GHz.

I dont have a 3870x2 or 4870x2 so my results are hardly comparable. I still have a HD2900XT 1GB.

Since you have a Intel quad and willing to test with could you please test this program for me on your intel?
If you run it write down the GigaFlops indicated on top of the window.
First after the program runs and then click in the middle of the screen until you only see stripes and then note the gigaflops again.

Maybe running it at diferent speeds also to see the scaling.

[xPliziT]

Ok here are the 3GHz tests.
I also used the AOD red dot feature to see what effect it has.

First picture: 3GHz, NB 2 GHz, no AOD
second picture: 3GHz, NB 2GHz, AOD red dot
third picture: 3GHz, NB 2.4GHz, no AOD
fourth picture: 3GHz, NB 2.4GHz, AOD red dot

[JumpingJack]

^^^^ We are pretty darn close .... nice job. Thanks. It's the reproducibility that I am always wanting to see. Seeing your run get within 1-2 FPS of my runs is good.

Yeah, will run the program for you shortly.

EDIT: Get warning -- application configuration is incorrect, try re-installing.

Jack

[xPliziT]

Oh you need visual c++ 8 runtime installed

[JumpingJack]

Oh you need visual c++ 8 runtime installed

Ha.... I will download it and install it. I think I have it saved somewhere

Nope, downloaded and installed VC++ 2005 SP1 redistributable, and same error.

[Hornet331]

Or just use V-Sync at 60 FPS/60Hz to limit overall high FPS to get a good average of all the median/low fps. Thats the way i would do that.

i dont like v-sync espeical with lcds, in most cases it increases input lag.

[Boschwanza]

i dont like v-sync espeical with lcds, in most cases it increases input lag.

Yes, but for comparision of median low FPS its perfect. You dont have it to use anyway.

*still waiting for the CoHOF benchmarks*

[JumpingJack]

Yes, but for comparision of median low FPS its perfect. You dont have it to use anyway.

*still waiting for the CoHOF benchmarks*

Working on it....

[Boschwanza]

Working on it....

[JumpingJack]

Ok... the reason this has taken me so long is that COH:OP is a new game to the benchmarking suite -- COH opposing force (I have the original COH installed) installation was tricky ... aside from not installing over COH correctly, it also is the most patched game on the planet.

If I cannot get an identical install on each rig (the Phenom and QX9650), then I start over and make sure that everything is identical. After a few attempts, I had to wipe the original COH and install new and fresh.

What I did: This is the first in several runs -- feel free to request specific items. Also, all the runs here are geared toward a CPU perspective, I will also be doing high quality/high AA setting runs as well. The configuration has been linked previously in this thread, briefly Phenom 9850 @ 2.5 GHz (stock)/QX9650 @ 2.5 GHz, Asus boards are the M3A32-MPV deluxe/Maximus Formula (X38), DDR2-800 C5, 4870X2 with Cat 8.8, WD Raptor 150 Meg, Windows XP SP2 ....

The runs were done at two resolutions. First, 1280x1024, with all low setting except physics which was set to high (CPU duty) and effects density was set to high. Then I ran the same settings but at 1920x1200. These were done using the -novsync command line which disables the 60 FPS limit, I also ran with the Vsync on by removing the -novsync command line. Finally, I collected FRAPs for the 1920x1200 run....

Phenom 9850 @ 2.5 Ghz
1280x1024, all low physics high, effects density high

1920x1200, all low physics high, effects density high

1920x1200, all low physics high, effects density high -- VSYNC thus 60 FPS


QX9650 @ 2.5 GHz
1280x1024, all low physics high, effects density high

1920x1200, all low physics high, effects density high

1920x1200, all low physics high, effects density high -- VSYNC thus 60 FPS


Note: In a fit of exhaustive stupor .. I mislabeld the JPG filenames for the QX9650 1920x1200 runs, it is is 1200 not 1280 ..

Finally, the 4 fraps traces for the 1920x1200 runs -- NOTE: the above screen shots were not running fraps as fraps induces a few FPS hit, these were separate runs -- and were within 3-5 Ave FPS of what is shown above.


A few notes to make ....

* Though at low settings graphically, 1920x1200 is still CPU bound -- this was not the case for my 8800 GTX which started becoming GPU bound at these graphical settings at 1600x1200. Demonstrating the power of the 4870X2.

* COH: OP is not a great game to use as a bench in this case, it way favors Intel architecture ... so it is not a generalized representative statement of the population of games.

* The most important point: Either CPU is capable of supporting > 60 FPS in the capped (synced) regime ... as expected. The take away is that the CPU will not matter to the overall quality of game play for this particular game (and I have done enough to say that as a generalized statement as well).

I will be following up with CPU utilization and high quality settings too, stay tuned....

Jack

[Boschwanza]

First of all thanks for this, but i have to add some critical points.

CoHOF has the great advantage that it shows you which option in the graphic setup affects graphic card, processor and system memory (simply scroll over the options) .

So far physics only affects the cpu, but there are several options which affects cpu and graphic as well, like shadows, water reflection, landcape details, model details, effect playback, effect density and object shifting.

So i would recommend to add this options to your benchmark. Again, thank you. (A run with all settings high and AA would be nice too )

[JumpingJack]

First of all thanks for this, but i have to add some critical points.

CoHOF has the great advantage that it shows you which option in the graphic setup affects graphic card, processor and system memory (simply scroll over the options) .

So far physics only affects the cpu, but there are several options which affects cpu and graphic as well, like shadows, water reflection, landcape details, model details, effect playback, effect density and object shifting.

So i would recommend to add this options to your benchmark. Again, thank you. (A run with all settings high and AA would be nice too )

Yeah, I knew of this and when designing the processor centric settings I experimented with each to find the ones that impacted the performance most when changing CPU speed.... those were the two biggies, everything else convolutes the CPU vs GPU which I was trying to separate out....

Nonetheless, the more interesting stuff for most people are the settings that make the game fun and full of eye candy... still working on that, I should have it in the next day or so.

Jack

[xPliziT]

Thx for the comparsion JumpingJack,

could you please add your comparsion for both processors running the mandelbrot program?

And btw. not quite sure but running with physics high might be a double sided sword since all those effects need to be rendered also.
Best would be with physics low i guess although it lowers cpu activity.

[Tonucci]

Thanks jack, interesting results

I didnt read all the thread, but have you compared phenom and c2q in crysis on demanding settings ? To see if the phenom's I/O performance thing is true etc ?


PS: Just as an side note, some reviews like [H], and a few other sites, utilizes "average minimum" FPS, not absolute min. Absolute min FPS is not very usefull, but avg min is, imo at least, by far the most important thing when buying an card.

[Hornet331]

PS: Just as an side note, some reviews like [H], and a few other sites, utilizes "average minimum" FPS, not absolute min. Absolute min FPS is not very usefull, but avg min is, imo at least, by far the most important thing when buying an card.

problem with that: what is avg min fps?

there is no general definition of that, for avg fps on the other hand there is, same goes for absolute max and min.

[Tonucci]

problem with that: what is avg min fps?

there is no general definition of that, for avg fps on the other hand there is, same goes for absolute max and min.

Thats true...no general definition. And I may have exaggerated a bit about its importance

I dont know how they calculate...I could not find an answer... but looking at graphs, like [H]'s apples-to-apples graphs, I can see how much time each card stays at very low FPS, or how many dips.

Its not accurate at all, but still is an better indication of overall smoothness than avg or max fps imo.

an card can spend half of the game below 20 FPS, but if it gets 400fps for a few minutes, the avg fps may be high. In this example, gameplay would be horrible, despite the high avg and max FPS.

Reviews can be misleading in that sense...so we have to look at the overall picture.

i dont want to de-rail this thread, but i felt like i had to explain myself, as I was very vague

[Hornet331]

an card can spend half of the game below 20 FPS, but if it gets 400fps for a few minutes, the avg fps may be high. In this example, gameplay would be horrible, despite the high avg and max FPS.

actually no, cause if a game spends most of the time at 20fps a few minutes won't increase the avarage fps dramatically.

[gosh]

Sample calculations for one minute:

20 FPS for 30 seconds, 100 FPS for 20 seconds, 400 FPS for 10 seconds
Average = (20 * 30 + 100 * 20 + 400 * 10) / 60 = 110 FPS

25 FPS for 30 seconds, 80 FPS for 20 seconds, 200 FPS for 10 seconds
Average = (25 * 30 + 80 * 20 + 200 * 10) / 60 = 72,5 FPS

20 FPS for 1 second, 100 FPS for 59 seconds
Average = (100 * 59 + 20) / 60 = ~99 FPS

[Tonucci]

actually no, cause if a game spends most of the time at 20fps a few minutes won't increase the avarage fps dramatically.

The example may not be accurate but i hope you got the point

Thanks for taking the time gosh, shows what I meant.