4850 Xfire Benchmark suite - what you guys after?

OH, for anyone wanting to do a comparison with the 3870Xfire setup, I just reran the SF4 1920x1200 8xAA benchmark. Didn't take a screeny but the average FPS was a little over 110.
If I remember right, PmanUKs setup got about 85 with the same settings, which would indicate; if CPUs werent a limitation in either setup, a difference of 25fps approx.

This would indicate at stock clocks, the GPUs in the 4850 setup are about 22% faster than the 3870s, which sounds moderate, but then consider the 3870s were clocked at a little over 600Mhz on the core and these are clocked at 500. If we can overclock the cores to compensate, then we could be talking in the region of 30-50% faster GPU clock per clock (110/5=27.... 110+27=137 [aka 4850@600Mhz]. 85 vs 137 = 52fps difference = 44.2% faster clock per clock...Im not sure on the exact clock of the 3870 cards to equalise smaller maths ). This is all theoretical though right now

 

what was your score on farcry2?

 

Glad to see the reseating helped so much .

 

10-15 degrees doesnt make my machine as cool as some of the others; but its a hell of a difference. Shame its not easier with this machine! Probably the most awkward CPU fit I've had since...AMD Athlon XP period.

@Don, I haven't had a chance to run that yet; I've had a lot of long work days so been run a bit ragged on that sort of stuff this week, especially when most of my time in the last few days has been trying to get this X9100 sorted.

Im off this weekend though so hoping to get a full 're-benchmark suite' run this weekend, still need to reinstall half the bloody things. Crysis benchmark also seems to be a no go as I mentioned in another thread as it just kicks me onto level 1, and doesnt actually benchmark.

Im not the only one with this issue so guess it's a bug?

Will try and get the 'suite' uploaded by Sunday afternoon UK time

 

oh let me see that results and we'll tell you what's up with it. i miss that one completely.

crysis and are you patched and what drivers are you running.

 

Latest drivers, Crysis fully patched (1.21 IIRC), when you go to run the benchmark it just throws you into the 1st level instead of running a demo loop. In believe EvilCorsair was also having the issue. Will happilly get the rest of the restests done this weekend though!

Am looking forward to seeing the machine put through its paces at a more even level to a lot of the other specs floating around!

 

oh yes i have had this error also. try running it in 32Bit mode(x86)

 

Well guys as promised; here's a full suite of updated benchmarks, run on an x9100 @3.325Ghz, and the new 9.8 Catalysts (mobility modded)

I have updated the OP with these

3DMark 06 (default settings)

Cinebench R10 X64

CrystalDisk

DMC4 DX9 1280x720 (720p)

FarCry2 1.02 Benchmark (DX10, Very High, 1920x1200)

HDTach

HDTune

Last Remnant (Default settings; 1280x720 Full Screen)

PCMark Vantage

Resident Evil 5 (Default Settings, 1920x1200)

Street Fighter 4 Benchmark 720p

Streetfighter 4 Benchmark (1080p)

Tropics UniEngine (Default settings, Full Screen)

3DMark Vantage

 

can you run your benchmarks like dmc4 and SF4 at 1920*1200 maxed out then 2aa and 8aa.

you can also use setfsb and you should be able to hit 3.57ghz on it stable.

i'm liking your fc2 scores you beat me and i was using a quad.

 

^^^^^^^i agree with the moo!^^^^^^

=b thats abit shocking!!!!

 

AFAIK theres some sort of driver bug with RE5 benchmark atm; I've seen it mentioned elsewhere in regards to the 4 series cards and the RE5 benchmark.

@Don, unfortunately no my x9100 is not stable at 3.5Ghz linpack, and I will not run it at this speed if its not; it's senseless, given the heat produced by this and some other X9100s recently, I'd hazard a guess to say some of the later X9100s are not quite as cool or as voltage light as some of the earlier ES.

Will go get some DX10/AA screens for you

 

no don't run occt at 3.5ghz you'll kill your system. you run it for benchmarks. and then reboot when your done.

 

Yeah I've run mine at 3.5 for a benchmark but slowed it down to 3.3 for normal run. I posted this in another thread but here is an awesome cooler

http://www.amazon.com/ACC-NT-CRYO-B...1?ie=UTF8&s=electronics&qid=1250424131&sr=8-1

 

You have to play the game some until "contact" is not at the top of the list for saves within the game. Once you are able to displace "contact" then you can run crysis benchmarking tool.

 

Im not interested in unstable benchmarks, I dont see the point. Intel use linpack as a guage of thier processors; so I stick with the same methodology. If its linpack stable; I'll use it, if not I wont, Im not into 'suicide' runs, Im after replicable, everyday performance, which is what my benchmarks are designed to demonstrate, my 3.325Ghz clock is 'everyday' stable.

No point in benchmarking at x clocks if you can't use your machine that way

@Sgilmore
Thanks for the tip


Im in the process of running the NoAA/2xAA/4xAA/8xAA/8xAA maxxed out benchmarks atm...slow going but hey.

 

Agree with you here alex, whats a fast running system if its not stable.

 

Thanks, I'll give that a go and see if it helps.

(He was right about me having the same issue)

 

ok. I don't want to argue with you on that topic.

I can't wait to see the results of you dmc4 runs.

 

Right guys here we go,some extended benchmarks from DMC4 DX9, DX10, SF4 and also a re-run of FarCry2 in Ranch Long, set to Ultra High mode

I did try to retest RE5 under DX10, and it appears ATI *HAVE* corrected the poor performance in the benchmark there; however there appears to be some instability with the driver at the moment; as my graphics/the benchmark froze up part way through each time. THAT SAID, in the phase 3 where I was running at 30-25 fps in DX9, I was running 45-60fps in DX10...MASSIVE difference, so in a driver revision or two; I will try again, and they might have fixed/optimised the DX9 performance by this point too

These ATI Cards are VERY good at allowing high levels of AA at playable framerates!


Far Cry 2 (1920x1200, Ranch Long, DX10, Ultra)




Street Fighter 4 Benchmark 1920x1200

No AA



2xAA



4xAA



8xAA



8xAA Settings Maxxed out




Devil May Cry 4 Benchmark DX9 1920x1200

No AA



2x AA



4x AA



8xAA



8x AA Maxxed Out Settings




Devil May Cry 4 DX10 mode, 1920x1200

No AA



2x AA



4x AA



8xAA



8x AA Maxxed Out Settings




Updated the OP.

 

as excepted using aa brought up performance. no aa, it's on the same level as the 3870 crossfire both at stock clocks 4850(500/850) and the 3870(660/850). (however the 3870 suffers from aa )

what driver are you using to get these scores?

 

Hey alex what did everest tell you for the maximum temp after you ran the DMC4 tests?

 

Unfortunately Im not too familiar with everest since they went pay. Im only using the trial version and that doesnt *seem* to be logging the maximum temps, only current; unless this info is left somewhere else.

Crysis still won't benchmark properly either even though Im now on the second level.

Driver wise Im using MM 9.8s, I see no point in using outdated drivers, these 9.8 seem to have fixed the GPU-Z crashing issue as well.

 

Cant imagine what the numbers would be like if you overclocked the 4850s to 660 .

 

in order for everest to log the max temps you just need to go to tools, system stability test, and then tab over to statistics and it will log the max, min and average

if you could run it again I would be very much obliged! I need to know if my system still has issues or if it's normal for this rig to run so damn hot

i'm idling at 63-66 now that's AFTER as5 application..

 

Performance or heat?

The 4850s are pretty nippy little cards; rather hot though.

Should continue to outperform the 3870s as games become more shader heavy too; that's where a majority of the additional power of these cards comes from (along with AA Hardware resolve). Vantage kinda demonstrates that one

 

Maximus, Im idling around 58, AS5 has a 'cure/burn in time'. It's one of the reasons I prefer Tuniq MX-2 and OCZ Freeze, as they work on the ball providing they're installed properly; and are pretty much on par with AS5, gimme 5 mins and I'll go check DMC for you, currently have some OCZ Freeze on my CPU and some TX-2 on my GPUs

 

Oh ya they should stomp on the 3870s under certain situations especially when AA is involved . The reason again why AA takes no hit and even adds performance under the Capcom benches is because they use DX10.1 which is far far better then DX10 and also why Nvidia tends to take a decent hit .

 

Thanks man Really appreciate it

we'll see, if your temps really are much lower than mine I might replace the AS5 I just applied 2 days ago

at least my x9100 doesn't get hotter than 70C, even after wprime for 20 minutes

 

Right just ran DX10, 8xAA, maxxed settings, as I figured that'll probably use most grunt out of the DMC tests I can run, maximum temp recorded on any of the 3 sensors on each card was 94 degrees. My max idle temp was 57 though, so ambient could be coming into it.

37 degree difference between hottest idle and hottest load.
My TX2 was applied using the 'credit card' spread method (thin line on one end of the cores rubbed along the surface to make a thin layer with the excess removed as you spread it)

Shame they get so hot or I'd try going for gold on the overclocks, but with temps that high I'd rather wait and see what happens before I consider it, as I'd have to basically blind flash some modified drivers; even 550 both cores would probably yield some pretty nice results given the scaling of this series.

 

ok that's good news, highest mine gets was 96 so I guess I'm in the clear

Thanks a ton! I think these cards just run hot, despite the dodgy cooling in the m17

not too worried about it anymore if everyone is running them this hot

 

Haha yeah, oh I peeled off the dodgy yellow plastic glued onto the top of the GPUs as well, as I realised it was retaining TIM cleaner, which I didnt want! Took me about half an hour yesterday to get all the crud off...half tempted to go try OCZ freeze on them now and see if the load temp is cooler but...well I CBA atm lol

Its a shame the M17n doesnt use 2 smaller cooling loops for the two GPUs (seperate fan/heatpipe) rather than a larger connected single cooler; as I think thats reducing the effectiveness, especially having felt the temps of the heatpipes after taking the back off. The positioning of the cooling vents is a little bizarre as well!

 

I wonder if adding some copper heatsinks to the top of the gpu sinks would reduce temps by another 5-10C

that would be awesome

 

the 3870 at the same settings in dmc4 with 8aa is 60/57/62/55. aa takes big hit. with 16aa the game drps even more but that's a different story. but nice.

also different drivers helps different games, if it'old or new. 8.57 best for 3dmark06
8.10 last driver to overclock on 3870. I think changed now.
8.12 big increase in the stream cpu's active in games like fc2. ground breaking for ati.
9.1 optimzed alittle for dx10
9.2 leans of the cpu wasn't great
9.3 leans more off the cpu norhing great either.
9.4 very stable and one of the bes drivers I ever used.
9.5 used to get my fc2 record I posted before.
9.6 didn't try sold whitebook. but people reported that you can overclock with them
9.8 apparently it great for some and not for others.
alex, you don't have lost planet it's a old game but it's from capcom. you should have some good numbers there as well. it's forgoting bemchmark but I never broke 100 fps so it still counts in my head. there's a benchmarking tool in it.

all the snow in it and it's a very heavy gpu game. dx10 also improves performance over dx9.

 

did you read project freeze????????
everything is talked about with heat sinks on the gpu. I a making a power point for it but I hae to resize the pictures and I don't have the motivation because I a stuck at 1280*800 so pic at 1600*1200 gets me mad.

 

Unfortunately no Lost Planet; its one of those games I looked at a few times but never massively felt the draw.

The drivers used in my test were the Alienware (9.1s) and 9.8s, so gives you a little idea on some of the progression

I did actually buy a few basic ram heatsinks the other day on the cheap; made my CPU temps worse though IMO, as I dont think they had the airflow or heat dissapation to help. Will take a look at Project Freeze properly at some time and see whether its worth throwing any somewhere on the GPUs, but again I think lack of airflow would kill the utility a bit.

 

my mistake on 500/850 it's 550/850 as you pointed out.

the last driver I used was 9.5 so I dnlt know how well the other ones are.(9.6-9.8) but they seem promising in some games but not in others.....

 

When I bring up the max temps page on everest I dont have any GPU temps there. Only have them under the computer tab and then sensors. Weird as this only shows current temp and does not record highest. Another issue is has anyone managed to get powerplay to work with these cards. Checked GPUZ and the cards do not clock down when switched to power saver under vista. This would explain the high idle temps.

 

Not sure if you meant me there; the stock on these cards actually is 500/800... I was implying that if I could clock up to 550/800 given the scaling, the architecture would probably give a nice performance boost; and might actually directly scale in performance.
I mean compare a 4870/4890 vs the atomic edition...basic difference is the clocks; but the performance increases massively in some cases, just from an extra 250Mhz core over the 4870 1GB.

Marshall - upgrade to 9.8s, the GPU-z crashing an non-powerplay is fixed...both my cards happily sink down to 250/400 after being left alone for a moment, in GPUz when you first load they'll typically jump up, but will return to the idle speeds after a few seconds left on the sensor tab

Also, in everest, I got the same, you have to go to the stability tool as mentioned, and go into preferences. It'll probably have cpu/CPU1/CPU2/HDD selected at first; change this to the 3 sensor readings for both GPUs (using the drop down menus next to the different coloured lines), along with CPU1/2 and you're there. When you save these new preferences, it should carry over and show the temps

 

so cunfused. I hve been getting mixed numbers from the w90 4870's, the whitebook 4850ls/3870's and the msi gt725's 4850.
I think I got it know though.
3870 is 660/850
w90 is 500/850?
msi is???/???
your 4850 is 500/800(which says awhole lot in terms of potentional) do you have ati overdrive in your ccc?

 

just read over the whole project freeze, looks like people are seeing 5-8C drop in GPU temps from adding copper sinks

definitely going to do this, thanks dondadah

where can I get them in a store? I only have sundays off from work so I definitely want to get them today and try it out

Best Buy? Fry's electronics?

 

Unfortunately no CCC overdrive so I'd have to Bios flash..

W90 "4870"s are 550/850 (they're basically just BIOS clocked 4850s, as they retain GDDR3, not GDDR5)
MSI 4850s are 500/800 (512/1GB variants about)
M17 4850s are also 500/800

The M17 4850s use Qimonda RAM chips which might be important if we go down this route.

Id LOVE to be able to overclock these a little for everyday usage; unfortunately the heat and steps required make this a bit of an a**e (especially for testing) lol

The 4 series chips really do have massive potential if they could be clocked without massively raising our already high temps, but with the temps as they are we'd have to be very careful not to raise the temps more than about 5 degrees as they run close to 100 degrees already, and theres nothing more certain to kill a graphics card quick than raising its high temps further

 

Have the heatsinks on the GPUs changed from the 3870s ? or are they same ?.

 

Im not entirely sure; but they look fairly generic to me

 

at max
I would try them if it's not to far. but places like pcw computers would have them most likey. newegg and frozen cpu have them but they are online. i'll update the first post with links to everything on project freeze and a point will be nice.

at alex
I fl yo on that one. because of that I never really pushed my gpu's pass 708/900

but mod it like scook did in project freeze and use a crylo. he got stable at 735/1165 which is a big difference.

 

alex
does it look like this
http://i42.tinypic.com/2116gyc.jpg

 

Yeah exactly the same.

Don't suppose you can give me a headsup and let me know where the heatsinks were most effective (ie, 1 on each extention over the metal heatspreaders and 1 on each heatpipe etc)? I've got 8 little heatsinks here I'd be willing to lob in in the next half an hour or so and re-run the DMC4 benchmark to see if makes an obvious difference, not entirely sure if they're copper, but what the hey.

Took a few pics of the machine with the two heatsink assemblies taken off if anyone needs to see them

 

DMC 4 maxed out no VSync




I'm afraid to play with temps like this..

sad part is it was even worse before applying as5..

 

post everything in project freeze.

i would like you to take some pictures of your heatsinks and it on your notebook so i can see that size so you know that best place where to place everything for best cooling.

also rip off the cloth next to your fans on the cpu and gpu. and you will have to prop up the back because there's no air flow at the bottom of the notebook. i would use bottle caps.
@max
dmc4 is a hot benchmark because it really stresses out your shaders on your gpu.

but how did you get all that info on your gpu. when i used everest, i never got so much detail

 

Max, don't worry, the desktop ones run at similar temps under load; and the use dual slot heatsinks ATI cards are known for running pretty hot!