My pearly 940X4 and a 4850X2 + 4850

Antman56

Distinguished
Apr 2, 2008
34
0
18,530
Hello everyone,

It is my first time starting a thread at this forum and I am going to share my results of my Phenom 940X4 with a trifire of 4850's. I have posted this is multiple forums (including Anandtech and ATI's Crossfire forum) to help people get their foot in the door. Finding info on success has been pitiful at best. It sounds simple, but it was actually terrible. Getting these things to tri-fire was one of the worst experiences I ever had... which is pretty over-dramatic (I did just graduate from GaTech... which was WAY WAY worse).



*IF YOU ARE USING ASUS 4850s (like tons of other people, not sure if this applies to 4870s), PLEASE UPDATE THE CARDS BIOS FIRST. IT DOESN'T SEEM LIKE THAT SHOULD BE THE PROBLEM... BUT TRUST ME IT IS!!!



Anyways, back to my setup. I currently have my 3.0 GHz CPU @ 3.53 (17@208). I have my integrated memory controller and L3 cache running @ 2288 Mhz (208 x 11) and I have my ram running at 1109 (208 x 5.33).



*note - overclocking my processors northbridge (integrated memory controller and L3 cache) did increase the IPC from what I can tell. I am scoring equal or slightly above most 3.7 Ghz Phenom 2's (atleast as far as friends are concerned). It also brought my memory bandwidth from 9.96 GB/s to 12.87 GB/s (pretty radical increase).*

My computer running 3DMarks 2006 at default settings:

AMD940X4353ATI4850X3700-1000-1.jpg


My computer running 3DMarks 2006 at 1920x1080 8xAA:

AMD940X4353ATI4850X3700-10501920-1.jpg


My computer running 3DMarks Vantage at Performance Settings:

AMD940X4353ATI4850X3700-1000Perf-2.jpg


My computer running 3DMarks Vantage at High Settings:

AMD940X4353ATI4850X3700-1000High-2.jpg



After looking at these, it is pretty obvious that my score in 3DMarks2006 is a lot more CPU bottlenecked than a lot of the top-end machines capable of mid 20000's and up. The SM2 Tests ( The first 2) and the SM3 Tests (The last 2) are only using two cores for the in-game portions of the benchmark, thus Vantage is a better benchmark in my case... as I can only manage to push this kind of GPU power with all 4 processors. My GPU Score (all that matters) is on par with a Geforce 295!

EDIT: Fixed Typos and Updated my 3DMark Vantage Score due to driver update (Now running Catalyst 9.1).
 

Antman56

Distinguished
Apr 2, 2008
34
0
18,530


Thanks man... it has been pretty incredible. I am going to post some of my benchmark results when I get around to it. In game is all that matters.
 

Antman56

Distinguished
Apr 2, 2008
34
0
18,530


Alright guys, I ended up staying awake all night to finish off these benchmarks to give you a taste of a 4850X2 and a 4850 in TriFire.

The benchmarks are taken from the following 4 games: Devil May Cry 4, Half-Life 2, Race Driver: GRID, and Far Cry 2. As much I would like to do this for a living, I did not have all the time in the world to get these benchmarks ran at various resolutions… thus I stuck with 1920 x 1080 with no less than 8xAA. However, they all reflect in-game play experience. Don’t focus too much on tests with close frame rates, there is naturally testing error involved with benchmarking (about + 2% to -2%)

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Devil May Cry 4 is running in DirectX 10 mode with all settings maxed out. I used the built-in benchmarking test for my results. The benchmark consists of four different scenes. Scenes 3 and 4 are less graphically intense, while Scenes 1 and 2 are more busy.

Scene One Average FPS:

Scene1AVG.jpg


This scene is a good match for a 4850X3. Regardless of Phenom II’s processing power, the video cards are almost scaling perfectly. One sees around 2.7 times of the performance of a single 4850 when running Tri-Fire. The 3.53 GHz Phenom II with the overclocked L3/IMC does a little better than “multiplier-only” 3.5 GHz Phenom II.

Scene Two Average FPS

Scene2AVG.jpg


Here we see a little of the same as Scene One, though it should be noticed that there was a little fall off on the 3x Overclocked 4850 setup until you reach the more-tweaked 3.53 GHz Phenom II.

Scene Three Average FPS:

Scene3AVG.jpg


Scene Three demonstrates what happens when you have video cards that are not being completely utilized. This scene is less visually stressful, thus the video cards pass the bottleneck onto the processor. As the processor becomes more powerful, the frame rate starts to increase. A straight overclock from 3.0 to 3.5 produces about a 4.62% gain, while going from 3.0 to an IMC-modified 3.53 gives an 11.86% gain. Pretty impressive!

Scene Four Average FPS:

Scene4AVG.jpg


Scene Four is not quite as simple as Scene Three, though it shows the same tendencies. The 4850X3 setup does well increase with additional processor power, but far less than before. This test is fairly balanced in its CPU/GPU combination. The Tuned 3.53 Phenom II with the overclocked TriFire does succeed in breaking the curse of having a lower frame rate than when paired with a non-overclocked TriFire.

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Far Cry 2 is running in DirectX 10 mode with all settings maxed out. I used the built-in benchmarking tool for my results. The benchmark loops through the “Small Ranch” sequence 3 times and the final average of the minimum, the average and the maximum is taken.

Far Cry 2 Minimum FPS:

MIN.jpg


At 1920x1080 with 8XAA and 16xAF, anyone playing with a single 4850 silently wishes to jump off a bridge. The processor clock cannot come to the rescue, not even an i7. But if you tag in a fellow 4850, you get some near perfect scaling and some decent minimum frame rates. When running the Phenom II @ 3.0 with two 4850s overclocked, you see an incredible 2.05 times the performance of a single overclocked 4850. As we move into TriFire, crossfire scaling almost seizes to exist. My video cards get starved and the Phenom II starts to show its weakness, but it can be helped. In these conditions, my Phenom II is the weakest link and overclocking shows that this processor still has a little fight left in it. The 3.53 GHz IMC-modded processor steps in and brings the minimum frame rate up by 153% (36.47 vs. 23.76) when running the 4850X3 configuration at their factory clock speeds. Anything over 30 frames per second is a respectable worst frame rate in a game like this or Crysis.


Far Cry 2 Maximum FPS:

MAX.jpg


Here we see the least meaningful frame rate, the maximum frame rate. It’s usually a burst and it’s usually not useful when gauging a in-game play experience. To further support the aforementioned need for CPU power, you can see the benefits of running a higher clock speed than the stock 3.0 GHz.

Far Cry 2 Average FPS:

AVG.jpg


The frame rates resulting from overclocking with 2-Way and 3-Way Crossfire see healthy boost when at any CPU clock speed. The test has many segments loaded with visual complexity, which can be better handled by overclocking the video cards. The Phenom II running stock doesn’t have the extra CPU cycles left over to efficiently feed a third card. So you can move up to different levels of processing power and see a noticeable increase in TriFire well. Overall, overclocking the 4850X3 setup and the processor led to about 1.5 times the performance had previously. A stronger processor (A higher clocked Intel Core2 or an i7) would allow for better crossfire scaling, though 246% performance of a single card isn’t too shabby.

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Half-Life 2 is running in DirectX 9 mode with all settings maxed out. It is running the Lost Coast add-on (free download on STEAM). I used the built-in Video Stress Test for my results. Unlike the other tests, I proceeded to use 24xCSAA. Otherwise, none of the in-game AA settings would bottleneck a single 4850. This way, I can atleast show “some” scaling. Obviously, I don’t have the processor power to even touch two 4850’s potential in this game.

Half-Life 2 Lost Coast FPS:

AVG-1.jpg


You can see that processor means everything in these tests. The transition from 4850X2 to 4850X3 does practically nothing to the frame rate, because 4850X2 was already sitting on its hands waiting for the CPU to give it something to do. Overclocking the processor better utilizes these video cards, but the second and third cards are greatly under-utilized and could benefit from more processor power (even though we humans can’t tell anything over 60, making more FPS unnecessary). It is noteworthy regarding the boost in performance seen from modifying the IMC on the Phenom 2. When the video cards are not overclocked, the IMC-mod brings an additional 10.06% performance gain over the “multiplier only” method of overclock the Phenom. As a simple observation, one could argue that the IMC overclock has led to performance boost similar to overclocking an additional 200-300 MHz.

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Race Driver: GRID is running in DirectX 9 mode with all settings maxed out. I used FRAPS to record the frame rates from my tests. The benchmark is basically me starting the very first race of the game, waiting for ten seconds after the starting light goes green and then totaling the car into the wall on the first right turn (right around the 20 second mark). The average of the minimum, the average and the maximum are recorded by FRAPS.

GRID Minimum FPS:

MIN-1.jpg


The more important frame rate, for those looking for a flawless experience, is a smooth sailing on any 4800 series graphics card. It’s very shader 3.0 heavy, thus making ATI cards have a field day with it. At 1920 x 1080 with 8xAA 16xAF with the Phenom II and a single 4850 running stock, you end up with a minimum of 49 FPS. Pretty ridiculous. There is a decent amount of scaling on the way to 2-Way Crossfire, though it runs out of processor power to really push a second 4850 to its full potential. The Phenom II brings you past the “flawless” barrier at stock clock speed once you run a 2-way or 3-way crossfire, but that’s about it. Get an i7 if you need a minimum frame rate of 100+ FPS (For those Nvidia 3D Glasses-wearing individuals).

GRID Maximum FPS:

MAX-1.jpg


Not quite as important of a test, though it shows us what good scaling looks like (something a stronger processor would make clearer throughout these GRID benchmarks).


GRID Average FPS:

AVG-2.jpg


You see, more or less, a similar scenario to the “Minimum FPS” results. However, the TriFire setup does make a more noticeable appearance, though it is clear that more CPU power enable higher FPS. Overclocking these 4850s results in NO additional performance, so don’t bother if this game is the reason you did it. The IMC-modded Phenom II makes it clear that there is extra performance to be had at the same clock speed as its 3.5 GHz counterpart (making the Phenom II more competitive with the Core2 architecture on a clock to clock basis).

EDIT: I fixed some typos...
 
Well crap.

The only game out of those ones that I own is Half Life 2 (I'll run the stress test on mine to see though). Do you happen to have Crysis?

EDIT: my HL2 fps on the stress test were 172.77 (standard clocked 4870x2, i7 965 @ 4.13GHz)
 

Antman56

Distinguished
Apr 2, 2008
34
0
18,530



That's pretty good. That was with you running the "Loast Coast" addon with settings @ 24x "Edge Detect" Anti-Aliasing, 16xAF and 1920 x 1080?
 
Yep. If your suspicion of CPU limitation was correct though, the i7 at 4.13 is probably the reason why. My bet is that on a more GPU limited title, yours might even have a slight advantage (depending on the crossfire scaling of 3 4850s vs 2 4870s). Oh, and I got Far Cry 2 today, so I can try that out and see how it runs.
 

Antman56

Distinguished
Apr 2, 2008
34
0
18,530


Yeah man, do that. I am interested in seeing how 2 4870s compare. Thanks!
 
G

Guest

Guest
This is very interesting. I have a Core i7 @ 4 ghz and currently have two overclocked 4850s. I'm debating on what I can do to upgrade before the 5xxx series Q4 09 (such a long time eh?). I was thinking of getting the 4850X2, which I can still use with my 4850s... (Tri SLI x58 mobo).


Have you benchmarked with Crysis or other games? I'm really interested in that with modern catalyst drivers... as inital reviews of 3-4 ATI cards only led to worse performance with drivers from 6 months ago.
 
OK, for a run near the beginning of FC2, I get the following (fully cranked, 1920x1080):

Stock clocked i7 965 (3.2) + stock clocked 4870x2:
Min: 37
Max: 61
Avg: 49

Oddly enough, upon reopening the game to run another test at 4.0GHz, I find that the 1920x1080 option has vanished, so here's a run at 1920x1200:

Min: 30
Max: 55
Avg: 44.6

I'll try to get a 4GHz run in at 1920x1080 though - if I get it to work, I'll post the results.

(these are all 60 second averages btw...)

EDIT: still can't get in a 1920x1080 run at 4GHz, but here's one at 1920x1200, 4GHz, 4xAA (rather than the 8x AA used above - all other settings identical):

Min: 47
Max: 108
Avg: 68

These may be the settings to use for actual gameplay - the 4x is a noticeable bump in FPS, without a noticeable drop in quality.
 

nerrawg

Distinguished
Aug 22, 2008
500
0
18,990
Cheers Antman56 for the graphs and system test results. I found them more informative than some of the articles that are officially posted by THG. Great to see so clearly how the processor speed and memory bandwidth directly affects the performance of CF and tri card CFX. It seems like a lot of the poor scaling results with 4870 X2 CFX might possibly reversed somewhat by integrated memory controllers, overclocking these and achieving higher memory throughput. Do you think that this is the case, or do think that other factors, i.e. drivers, will restrict the scaling more?
 

Antman56

Distinguished
Apr 2, 2008
34
0
18,530


Hey nerrawg,

I apologize for the delay. I have been in the ATL interviewing like mad!

I appreciate the input about my article. I put a lot of time into it and I figured that it would be very beneficial to others.

I do not think that the memory bandwidth has as much to do with the scaling directly as some people have come to believe. Keep in mind, when I overclocked the Integrated Memory Controller, I overclocked the L3 Cache as well. I believe that the faster Level 3 Cache is more responsible for the performance increase. Unfortunately, I can only speculate because I cannot clock the IMC and the L3 independently of each other. One thing that can back up my theory however... look at the AM3 processors. The added bandwidth of DDR3 hardly affects performance at all, which means the extra bandwidth picked up from modifiying my IMC can't be TOO responsible for my performance gains.

I believe that the additional speed of the L3 cache is responsible for making my Phenom 2 more effective per clock when compared to a normal one. Thus making my 3.5 GHz chip perform like a 3.7+ GHz chip.

I have 3 video cards that my processor is responsible for managing, thus explaining why modifying the IMC/L3 causes such a decent boost in performance. My processor is clearly restricting my scaling (bottleneck). To see a more dramatic CPU bottlneck, look at the dual core only games I benchmarked (Half-Life 2: Lost Coast and GRID). Only 2 cores of my processor are somehow responsible to doing A.I, sound, physics, and feeding the video cards... thus the overclocking of the processor in anyway shows an immediate increase.

Games like Devil May Cry 4 don't show near as radical of an increase because it supports more than 2 CPU cores, thus my processor at stock speeds is almost enough to make sure the video cards are being pushed.

EDIT: Fixed Typos!
 

spearhead

Distinguished
Apr 22, 2008
120
0
18,680
it might be that phenom II be a bit of a bottleneck tough but it preforms good enough for a radeon 4850x2 and all games are perfectly playable and is cheaper then a core i7 setup this might be a good reason for alot of people to go for it. expecialy now sapphire has attended that it will soon fit a much more silent cooler on its radeon 4850x2 which is great news, the noise problem kept atleast me from considering the card. Soon we shall see what happens great review though it showed me what to expact from a simular system config. i hope the triple cores will prove to be unlockable soon then i might stick with the 720 black edition for the system im considering to buy in may because that is the time i will get my vacation money :D. as for the screen i consider Acer V233Hb as the most affordable high resolution option. as for the case im considering Cooler Master Centurion 534 (Black) because it looks awsome for a price which is not to high. maybe if any of you have a good idea which components i should install please PM me. im considering www.zercom.eu as a good site for most parts since it is by far the cheapest here in the netherlands who offers a wide assortiment and some reasonable service.