5970 vs 280 SLI benchmarks

J

JCDenton

Gawd
Joined
Jan 6, 2008
Messages
701
Here are my initial tests comparing 2x GTX280 to a 5970. Ran them last night with the 280s then today on the same system with no changes. I7 9.20 at 3.8. The GTX280 numbers will be on the left, 5970 on the right. The GTX280 used 190.62 drivers, while the 5970 was on beta 5s.

Crysis - DX10, pretty much maxed with one of the modified configs floating around. 1920x1200, no AA, 16x AF
GPU: 39.22 vs. 36.22 (the 280s dominated here and it's worse than the numbers indicate. Major stuttering on the 5970)
CPU1: 40.26 vs. 45.38
CPU2: 26.78 vs 26.54 (I always use the second loop, but oddly the first loop is always faster on this test. Both cards have a spot where they slow down massively on the second loop, but it was way worse on the 5970, hitting single digit frames for an extended time)

Unigine- DX10, 4xaa, 16xaf, 1920x1200
Tropics: 55.3 vs 71.2 (nice!)
Heaven: 51.1 vs 60.5

Arma 2
allmark: 3954 vs. 4006(there may be a cpu limitation here. when turning off hyperthreading and pushing the i7 to 4.1, I got a near 1:1 boost when playing around with the 280s last night, resulting in 4283)


Initial conclusions: Only the non-game benchmarks showed the kind of performance I would expect of this card. It may not be performing to its full potential in crysis as the air being exhausted wasn't as hot as it should have been, and in fact it beat the 280s in the CPU1 test, which is most similar to gameplay. The card is fairly quiet. Haven't tried to overclock it yet because CCC doesn't have a fan control slider for some reason. My other major knock on it is that switching between my 1920x1200 desktop monitor and my 50inch 1920x1080 plasma is a real pain in the ass. It's like 2 clicks with nvidia and a long drawn out process with ATI's CCC. There's no way I'll keep the card if there's not a better way to do this. Overall it seems like this card is definitely not a "must-have" right now for those who already have a good setup. I would recommend seeing how much new drivers can help this card and what Nvidia has in store. If the Unigine performance were duplicated in games and then 15% added to that through overclocks, it could be a very solid card, but no guarantee of that happening.
 
JCDenton said:
Haven't tried to overclock it yet because CCC doesn't have a fan control slider for some reason
Click to expand...

Hmm, no fan slider control seems like a possible driver issue? Like an old driver that needs to be removed with DriverSweeper or something to that effect?

So your getting bad performance in Crysis at the same time as what appears to be a driver issue? bad performance + driver issue. I wonder if these two could possibly be linked? Just saying...

Anyways, I think you'll be more pleased with your purchase once you at least overclock the 5970 to stock 5870 speeds.
 
WBurchnall said:
Hmm, no fan slider control seems like a possible driver issue? Like an old driver that needs to be removed with DriverSweeper or something to that effect?

So your getting bad performance in Crysis at the same time as what appears to be a driver issue? bad performance + driver issue. I wonder if these two could possibly be linked? Just saying...

Anyways, I think you'll be more pleased with your purchase once you at least overclock the 5970 to stock 5870 speeds.
Click to expand...

There ISN'T an old driver to be removed since this is the first time I've ever put ATI drivers on this machine. I took the NVIDIA drivers off in safe mode with DriverCleaner pro. ATI just decided to put this card out before decent drivers were ready. Google and you'll see lots of people missing fan control. I went and played a bit of Crysis and it's definitely way worse on the 5970. It seems like 5970 is only better than 280 SLI when there isn't a lot of vegetation around...so almost never in other words.

Here's some more indication that they rushed things:



Far Cry 2 shows slightly higher frames on the 5970, but compare the lines. Although overall FPS shows basically the same peaks and valleys, the 5970 line is thick and jagged, showing some big-time microstutter.
 
WBurchnall said:
Hmm, no fan slider control seems like a possible driver issue? Like an old driver that needs to be removed with DriverSweeper or something to that effect?
Click to expand...

That's a bios issue that can be remedied by flashing to an updated Sapphire OC 5970 bios that was given to an overclock.net forum member. Of course that voids you warranty, but I won't tell if you don't!

There was a review posted somewhere yesterday that found much better SLI scaling with 285s enabled them to be extremely competitive with the stock 5970. This card could really benefit from slightly better drivers and water cooling. I've also noticed I'm trading blows with a cheaper 280 sli setup I bought a year ago.
 
your number for Crysis seems to be way off the chart.

my 5870 have the same numbers as what you got on there... it seems somehow your CF is disable while running that benchmark.
 
well hard already benched 5970 against gtx 295 which is similar to your setup there is noway gtx 280 sli can touch 5970
 
This should become an interesting thread pretty fast.



Romir: I would like to read that review you mentioned if you could post a link.

I have to admit I find the numbers more than just a bit odd.
 
poppachocks said:
This should become an interesting thread pretty fast.

Romir: I would like to read that review you mentioned if you could post a link.

I have to admit I find the numbers more than just a bit odd.
Click to expand...

I'll dig through my history and find it. They did use physx for a few examples but I ignored those results, because I personally hated my framerate dropping by half at spots even with two 285s.

Edit: This isn't what I was looking for but here's a 5970 review with 285 SLI thrown in. http://techreport.com/articles.x/17986/1.
 
Last edited:
Romir said:
I'll dig through my history and find it. They did use physx for a few examples but I ignored those results, because I personally hated my framerate dropping by half at spots even with two 285s.
Click to expand...

Thanks. I'm sure many people here would like to read that.
 
yea this should be an interesting one...

as i will also be getting a 5970 in these two days..( waiting for it to arrive)
while i am using gtx 280 sli

i kind of knew the performance different is there.. but may not be that huge..
and so i am also getting enough monitor so i can use the eyefinity...
( i am chep so )
i am going to use 2 22" 225bw samsung monitor for the side and 1 24" benq in the middle.

all will be running 1680*1050
 
eric66 said:
well hard already benched 5970 against gtx 295 which is similar to your setup there is noway gtx 280 sli can touch 5970
Click to expand...

No, 295 is inferior to 280 sli by a decent margin.

Testing out some more stuff. the 5970 won by 10% in the GTA IV benchmark, and SEEMS to be faster in Crysis Warhead but I didn't bench. The train level was extremely smooth with all Enthusiast 1920x1200 no AA, and I was even able to put on 4x AA and it still ran well.

I do not want to flash the bios on a $600 card I just bought and might be ebaying. Not a good idea.

Also, I don't think CF is disabled in vanilla crysis. If that were the case, 1x GPU of the 5970 would be getting slaughtered by the pair of 280s, which scale really well in that game.
 
Last edited:
poppachocks said:
Thanks. I'm sure many people here would like to read that.
Click to expand...

Here we go: http://www.xbitlabs.com/articles/video/display/radeon-hd5800-crossfirex_9.html

"The Radeon HD 5870 tandem delivers an average 53% performance growth in the low-quality mode and 59% in the FSAA+AF mode. For the Radeon HD 5850 CrossFireX these numbers are 57% and 64%. For the GeForce GTX 285 SLI, the performance growth is 70 and 77%, respectively. Why does the Radeon HD 5850 tandem show more efficiency under the same conditions than the Radeon HD 5870 tandem? I guess the more advanced configuration is just more limited by the CPU."

They had a relatively beefy 4.1ghz i7 HT OC as far as these reviews have gone. I've been trying to compare my new cards results with reviews to see how much a 4.5ghz HT or 4.6ghz non-HT oc helps.
 
probably drivers still need some work don't know usually 5970 kicked gtx 280 slis ass in reviews lol maybe in specific games cf isn't that great
 
eric66 said:
well hard already benched 5970 against gtx 295 which is similar to your setup there is noway gtx 280 sli can touch 5970
Click to expand...
The GTX295 is 2 GTX275's in sli, not 2 GTX280's.
I'm sure that once ATI drivers mature the results will drastically change.
 
03grandam said:
The GTX295 is 2 GTX275's in sli, not 2 GTX280's.
I'm sure that once ATI drivers mature the results will drastically change.
Click to expand...

Two 275s do a decent bit better actually because of the higher clocks. Stock 280 performance is awfully close to a stock 275 in most cases though.

Jon55 said:
Enough said.
Click to expand...

My initial impressions so far are much more positive of the current beta driver than the one the 4870x2 launched with!
 
there is no hardware that launched with perfect drivers it takes time sometimes a lot for drivers to mature 8800 gtx had tons of issues when it launched first so its pretty normal give ati some time i am sure with 9.12 things will get better
 
JCDenton said:
Here are my initial tests comparing 2x GTX280 to a 5970. Ran them last night with the 280s then today on the same system with no changes. I7 9.20 at 3.8. The GTX280 numbers will be on the left, 5970 on the right. The GTX280 used 190.62 drivers, while the 5970 was on beta 5s.

Crysis - DX10, pretty much maxed with one of the modified configs floating around. 1920x1200, no AA, 16x AF
GPU: 39.22 vs. 36.22 (the 280s dominated here and it's worse than the numbers indicate. Major stuttering on the 5970)
CPU1: 40.26 vs. 45.38
CPU2: 26.78 vs 26.54 (I always use the second loop, but oddly the first loop is always faster on this test. Both cards have a spot where they slow down massively on the second loop, but it was way worse on the 5970, hitting single digit frames for an extended time)

Unigine- DX10, 4xaa, 16xaf, 1920x1200
Tropics: 55.3 vs 71.2 (nice!)
Heaven: 51.1 vs 60.5

Arma 2
allmark: 3954 vs. 4006(there may be a cpu limitation here. when turning off hyperthreading and pushing the i7 to 4.1, I got a near 1:1 boost when playing around with the 280s last night, resulting in 4283)


Initial conclusions: Only the non-game benchmarks showed the kind of performance I would expect of this card. It may not be performing to its full potential in crysis as the air being exhausted wasn't as hot as it should have been, and in fact it beat the 280s in the CPU1 test, which is most similar to gameplay. The card is fairly quiet. Haven't tried to overclock it yet because CCC doesn't have a fan control slider for some reason. My other major knock on it is that switching between my 1920x1200 desktop monitor and my 50inch 1920x1080 plasma is a real pain in the ass. It's like 2 clicks with nvidia and a long drawn out process with ATI's CCC. There's no way I'll keep the card if there's not a better way to do this. Overall it seems like this card is definitely not a "must-have" right now for those who already have a good setup. I would recommend seeing how much new drivers can help this card and what Nvidia has in store. If the Unigine performance were duplicated in games and then 15% added to that through overclocks, it could be a very solid card, but no guarantee of that happening.
Click to expand...

Dude, I hate to tell you this, but you are PUNCH DRUNK if you think these numbers are accurate. I have a HIS 5970 and nearly the same setup and my numbers are markedly better that what you are getting. Its either a driver install issue or a bios issue or a card issue but these numbers are nothing like mine.
 
Zepharus said:
Dude, I hate to tell you this, but you are PUNCH DRUNK if you think these numbers are accurate. I have a HIS 5970 and nearly the same setup and my numbers are markedly better that what you are getting. Its either a driver install issue or a bios issue or a card issue but these numbers are nothing like mine.
Click to expand...

those numbers are what I have for a single 5870....that is the odd part...:confused:
 
Still waiting for my 5970 but I have watched some off camera videos of Crysis running on it and it seems smooth, void of any stuttering. Haven't seen any reviews complain about stuttering either, so that's odd.
 
Zepharus said:
Dude, I hate to tell you this, but you are PUNCH DRUNK if you think these numbers are accurate. I have a HIS 5970 and nearly the same setup and my numbers are markedly better that what you are getting. Its either a driver install issue or a bios issue or a card issue but these numbers are nothing like mine.
Click to expand...


Please post your numbers for either of the unigine tests at stock clocks,DX10, 1920x1200, 4xaa, 16xAF, everything maxed. These guys are running basically the exact same test I did and getting the same results:

http://benchmarkreviews.com/index.p...k=view&id=403&Itemid=47&limit=1&limitstart=12

x3sphere said:
Still waiting for my 5970 but I have watched some off camera videos of Crysis running on it and it seems smooth, void of any stuttering. Haven't seen any reviews complain about stuttering either, so that's odd.
Click to expand...

Warhead is looking great so maybe there is a driver problem with the original crysis only. I want to try DragonAge and SHIFT but neither will load. Both say PhysXLoader.dll was not found. I thought that was an Nvidia-only thing?
 
psyside said:
This is 5970 Crysis Warhead numbers

1920x1200 2xAA

Min 29

Avg 43

Max 53

Codepath ambush. Your numbers are way off i agree with everyone in the thread.

http://www.youtube.com/watch?v=8FP7AHw3g-c
Click to expand...

I just DLed and tried that same tool and I get on the best run (OCed to 850/1200, which is fine for crysis but throttles in Furmark):

Min: 26
Max: 61
Avg: 46

I don't want to swap cards again but I found this:

http://forums.nvidia.com/index.php?showtopic=137799

Post 15 shows GTX280 sli at STOCK speeds hitting 47.83 average, but at 1680x1050 and 0xaa. So adjust for the slightly lower res/aa and not having an overclock, and it looks like the 5970 would win but not by a lot. Sorry that the emperor has no clothes and that ATI's next gen dual GPU card doesn't beat two last-gen nvidia cards by much. Look at it struggling to beat the pair of 285s here:

http://www.anandtech.com/video/showdoc.aspx?i=3679&p=8
 
With the 9.12 beta 5, 5970 should rape GTX285 sli with a quite big difference.
Dunno about the tests tbh.
 
yeah I would have thought so aswell, otherwise it doesn't look good performance/value wise.
 
psyside said:
With the 9.12 beta 5, 5970 should rape GTX285 sli with a quite big difference.
Dunno about the tests tbh.
Click to expand...

Ok, show one review where this happens in the majority of games. I apologize that ATI has underdelivered and consequentially caused me to be unable to post benchmarks that would please ATI fanboys. I'm willing to chalk Crysis 1 up to ATI shipping the card before making proper drivers, but Crysis Warhead is working properly as shown by the similarity between my results and that youtube vid, and the results are underwhelming for a "next-gen" dual GPU solution.

If we make the relevant cross-gen comparisons, we have 5970 compared to two of Nvidia's last gen GTX280 cards, and the last-gen GTX295 compared to two of ATI's cards from the generation before that, which is 3870. The 5970 is just not the major jump people want to make it out to be. The only place it's impressive is Unigine, which sadly isn't playable.
 
JCDenton said:
Ok, show one review where this happens in the majority of games. I apologize that ATI has underdelivered and consequentially caused me to be unable to post benchmarks that would please ATI fanboys. I'm willing to chalk Crysis 1 up to ATI shipping the card before making proper drivers, but Crysis Warhead is working properly as shown by the similarity between my results and that youtube vid, and the results are underwhelming for a "next-gen" dual GPU solution.

If we make the relevant cross-gen comparisons, we have 5970 compared to two of Nvidia's last gen GTX280 cards, and the last-gen GTX295 compared to two of ATI's cards from the generation before that, which is 3870. The 5970 is just not the major jump people want to make it out to be. The only place it's impressive is Unigine, which sadly isn't playable.
Click to expand...

TBH, it just your test got some problem while others doesn't.....
 
Umm did you even look at the Tomshardware test? The 5970 is barely ahead of 2x GTX285 and only even gains much at all on it with 8x AA.That's pathetic. It's more like a refresh than a next-gen card. I'm glad mine was only $500. Looks like I'll be making a $300 profit.
 
JCDenton said:
Umm did you even look at the Tomshardware test? The 5970 is barely ahead of 2x GTX285 and only even gains much at all on it with 8x AA.That's pathetic. It's more like a refresh than a next-gen card. I'm glad mine was only $500. Looks like I'll be making a $300 profit.
Click to expand...


Then do it already. It's pathetic you coming back and arguing here when you have no plans to keep the card.

Where is that "attention whore" pic when you need one.
 
JCDenton said:
Umm did you even look at the Tomshardware test? The 5970 is barely ahead of 2x GTX285 and only even gains much at all on it with 8x AA.That's pathetic. It's more like a refresh than a next-gen card. I'm glad mine was only $500. Looks like I'll be making a $300 profit.
Click to expand...

You said find me one review i posted 3? who cares if its toms...


I'm out btw this is getting ridiculous.
 
JCDenton said:
Post 15 shows GTX280 sli at STOCK speeds hitting 47.83 average, but at 1680x1050 and 0xaa. So adjust for the slightly lower res/aa and not having an overclock, and it looks like the 5970 would win but not by a lot. Sorry that the emperor has no clothes and that ATI's next gen dual GPU card doesn't beat two last-gen nvidia cards by much.
Click to expand...

lol wut? 1680x1050 has ~76% the number of pixels as 1920x1200. So 47.83 * .76 ~= 36 fps. Take another 10-15% for 2xAA and you are down to ~30-32 fps (admittedly a very rough estimate) for the GTX 280 SLI vs. 46 FPS for the 5970, which works out to the 5970 being ~50% faster. Again, I know that that is just a very very rough and crude estimate, but you are the one that offhandedly implied that that 1680x1050 0xaa results vs. 1920x1200 2xaa are easily comparable and deduced that the 5970 would only be slightly faster, when the math just doesn't support such a conclusion.
 
kllrnohj said:
lol wut? 1680x1050 has ~76% the number of pixels as 1920x1200. So 47.83 * .76 ~= 36 fps.
Click to expand...

LOL, this calculation is full of fail. You should know better than that. Why not just rely on the actual numbers instead of your made up theories? Oh, that's because the actual numbers show things like:

http://www.tomshardware.com/reviews/radeon-hd-5970,2474-8.html
Crysis 1920x1200
GTX285 SLI: 46.9 fps , 5970: 49.4 fps

LOL @ a <10% gain for "next-gen" dual gpu.
 
did the 280gtx run dx11?
oh, they cant...
Nvidia all current hardware, dx11 0fps, yes you did read correctly, zero fps in dx11.
now, then a 5970 hell even my 5850 beats the crap out of anything Nvidia can do with any combination of cards.
Old code paths as dx9 and dx10, and even to some degree dx10.1 older cards are better optimized for.
That has been consistent with every new generation cards.

If I would run a hardware site, I do a dx11 only test, I guess anyone would know what card to recomend then.
Its a no contest what cards to buy currently.
 
Flopper said:
did the 280gtx run dx11?
oh, they cant...
Nvidia all current hardware, dx11 0fps, yes you did read correctly, zero fps in dx11.
now, then a 5970 hell even my 5850 beats the crap out of anything Nvidia can do with any combination of cards.
Old code paths as dx9 and dx10, and even to some degree dx10.1 older cards are better optimized for.
That has been consistent with every new generation cards.

If I would run a hardware site, I do a dx11 only test, I guess anyone would know what card to recomend then.
Its a no contest what cards to buy currently.
Click to expand...

I wish nvidia had patented this kind of idiocy, I'm not sure I can stand this coming from the ati side as well.
 
best thread ever. The poor guy is simply stating that he was expecting a little more of the results and everyone jumps on him telling him that he is wrong.
It's stil a great card and better than any other single card offerings.
However the OP was comming from a dual card setup comparable to a GTX295 and expected a bigger jump in the the most graphic intensive game Crysis and didn't get it. Reviews back his up...end of story.
The HD5970 IS top dog, so everyone can relax...he just expected more of a leap in performance for the huge outlay that's all.
 
You must log in or register to reply here.
Back
Top