ASUS EAH5870 Review @ [H]

FrgMstr

FrgMstr

Just Plain Mean
Staff member
2FA
Joined
May 18, 1997
Messages
58,534
ASUS EAH5870 Review - Today we look at ASUS’s brand new, voltage tweakable and highly overclockable EAH5870 based on AMD’s new ATI Radeon HD 5870. With the ability to increase the core voltage we cannot wait to see how high it will go, but first we must see how it compares to its rivals in some of the latest and most demanding games.

ASUS has delivered a fantastic ATI Radeon HD 5870 based video card with the option to manualy tweak GPU voltages. SmartDoctor worked easily, and with voltage modification allowed us to achieve above 1GHz GPU frequencies improving performance.
Click to expand...
 
Good review. I'm a little shocked how little improvement there is over the 285 in some games. Then you come to Arma II and Crysis and you see the kind of improvement you expect.

I wonder how much of this is directly caused by TWIMTBP and how much is immature drivers (is there a real difference or is that splitting hairs?).

Scary overclocking though, a little surprised it doesn't result in more preformance. I guess it's being limited by the speed of the memory? (Seeing how the memory jumped by 8.3% while the shadders had a 19% increase) Did ya'll test any of the other games with the overclocked or just those two?

Ninja edit: The difference on those crysis screen shots is insane!
 
I got the Asus 5850 and don't really see that much performance increase when overclocking the memory. On stock clock 725 core/1000mem, I get 12777 GPU on Vantage. Overclock to 725/1000mem, I get 13500. Overclock to 980 core/1000mem, I get 15777 GPU. Overclock to 980/1200 mem, and I get 16200 GPU score. THis is all with a 6850 @ 3.83ghz. THe CPU Score is always around 6000 is all tests. Point is I don't think the 5800's are really that memory limited otherwise would see bigger performance increases when OC'ing the mem. Maybe Kyle can confirm if he notices this too. Makes me think if/when the 5890 comes out, if they get crazy core clock on it, like 1150 to 1200 core, you'd see rediculous performance jump.
 
vengence said:
Good review. I'm a little shocked how little improvement there is over the 285 in some games. Then you come to Arma II and Crysis and you see the kind of improvement you expect.

I wonder how much of this is directly caused by TWIMTBP and how much is immature drivers (is there a real difference or is that splitting hairs?).
Click to expand...

I think you are hitting a nail on the head in regards to drivers. We made the statement in the evaluation that the GTX 280/285 has had over a year for its drivers to mature. The 5870 is only a month old with its first WHQL driver just going live recently.

AMD has also rather had some bad luck with recent games launched lately:

Batman - NV slant is obvious
NFS Shfit - AMD claims not getting the performance they think they should be getting
RE5 - Current bug on 5000 series possibly causing loss of performance

Crysis being heavily shader limited shows the potential, and Arma 2 I guess was more shader limited than we thought!

Scary overclocking though, a little surprised it doesn't result in more preformance. I guess it's being limited by the speed of the memory? (Seeing how the memory jumped by 8.3% while the shadders had a 19% increase) Did ya'll test any of the other games with the overclocked or just those two?

Ninja edit: The difference on those crysis screen shots is insane!
Click to expand...

Quite possibly being memory bandwidth limited, hard to tell though without the ability to really crank up the memory frequencies. The 5870's aren't overclocking as high as we thought they would in the memory department yet. Perhaps a card will come out supporting memory voltage tweaking, then we can see. Perhaps also things like DX11's tessellation will benefit more with overclocking, that will be interesting to see. In this evaluation we did not testing other games for overclocking performance, but in future evaluations we will be expanding on that.
 
SirCumsized said:
I got the Asus 5850 and don't really see that much performance increase when overclocking the memory. On stock clock 725 core/1000mem, I get 12777 GPU on Vantage. Overclock to 725/1000mem, I get 13500. Overclock to 980 core/1000mem, I get 15777 GPU. Overclock to 980/1200 mem, and I get 16200 GPU score. THis is all with a 6850 @ 3.83ghz. THe CPU Score is always around 6000 is all tests. Point is I don't think the 5800's are really that memory limited otherwise would see bigger performance increases when OC'ing the mem. Maybe Kyle can confirm if he notices this too. Makes me think if/when the 5890 comes out, if they get crazy core clock on it, like 1150 to 1200 core, you'd see rediculous performance jump.
Click to expand...

I would not trust 3DMark's GPU score as a metric for video card performance advantages via memory overclocking.
 
SirCumsized said:
I got the Asus 5850 and don't really see that much performance increase when overclocking the memory. On stock clock 725 core/1000mem, I get 12777 GPU on Vantage. Overclock to 725/1000mem, I get 13500. Overclock to 980 core/1000mem, I get 15777 GPU. Overclock to 980/1200 mem, and I get 16200 GPU score. THis is all with a 6850 @ 3.83ghz. THe CPU Score is always around 6000 is all tests. Point is I don't think the 5800's are really that memory limited otherwise would see bigger performance increases when OC'ing the mem. Maybe Kyle can confirm if he notices this too. Makes me think if/when the 5890 comes out, if they get crazy core clock on it, like 1150 to 1200 core, you'd see rediculous performance jump.
Click to expand...

AMD has been quick to tell everyone that the 5870 is actually compute limited, and not memory starved. If you look on Google for "5870 Oc Scaling", you'll see several articles that show the 5870 responding a lot better to core overclocking than memory. Goes to show that AMD knows a hell of alot more about their cards than we do.

Link to XtremeSystems Forum where some testing was done. http://www.xtremesystems.org/forums/showthread.php?t=235181

Nice review [H]. Can't wait for your Eyefinity article(s).
 
Last edited:
I just want to re-iterate, when reading the evaluation, and we made this point in it, that the advantages over the Radeon HD 4890 that the 5870 is replacing are great. Coming from last gen, the 5870 really kicks up performance from that.
 
SicKlown42012 said:
AMD has been quick to tell everyone that the 5870 is actually compute limited, and not memory starved. If you look on Google for "5870 Oc Scaling", you'll see several articles that show the 5870 responding a lot better to core overclocking than memory.
Click to expand...

I have also been told this by AMD, a point I made in thee 5770/5750 evaluation.
 
Great review. Just wondering why you didn't use 9.10 drivers. Although I had some problems with Batman and the 9.10 drivers with Win 7 and right now I am using the 9.11 Betas and Batman works and just installed HawX and that works also with Win 7.
 
fentzinator said:
Great review. Just wondering why you didn't use 9.10 drivers. Although I had some problems with Batman and the 9.10 drivers with Win 7 and right now I am using the 9.11 Betas and Batman works and just installed HawX and that works also with Win 7.
Click to expand...

The Betas we used are Cat 9.10, just not WHQL'd. This evaluation was performed before the WHQL'd version was released. Our drivers, however, are no different than the WHQL's, they contain the same fixes, performance and features.
 
Brent_Justice said:
I just want to re-iterate, when reading the evaluation, and we made this point in it, that the advantages over the Radeon HD 4890 that the 5870 is replacing are great. Coming from last gen, the 5870 really kicks up performance from that.
Click to expand...

QFT
I made that exact upgrade myself, and I've noticed a huge improvement. I'm getting better performance at 1920X1200 8XAA with the 5870 than I was getting at 1680X105 8XAA with the 4890. Add in Windows 7 on a SSD and this thing is soaring.
 
Just put two of these in my rig and i have to say they are really nice. I can't see the need for overclocking them as of yet tho. Every game i've thrown at them at highest settings hasn't phased them what so ever.
 
fentzinator said:
Great review. Just wondering why you didn't use 9.10 drivers. Although I had some problems with Batman and the 9.10 drivers with Win 7 and right now I am using the 9.11 Betas and Batman works and just installed HawX and that works also with Win 7.
Click to expand...

Hmm, I played though Batman this weekend with the 9.10 drivers and didn't have a problem at all.
 
Nice review Brent. I have the same card myself and think it's a beauty! I've only done some mild overclocking at stock volts and hit 930 core but only 1240 on the memory. The performance when going to 1250-1260 on the memory does start to degrade, but after that it definitely starts to artifact and when I hit over 1300 sometimes it can even lock up. I thought the EDC was supposed to eliminate that? Not complaining because the core is where it's at!
 
mrrogers747 said:
Nice review Brent. I have the same card myself and think it's a beauty! I've only done some mild overclocking at stock volts and hit 930 core but only 1240 on the memory. The performance when going to 1250-1260 on the memory does start to degrade, but after that it definitely starts to artifact and when I hit over 1300 sometimes it can even lock up. I thought the EDC was supposed to eliminate that? Not complaining because the core is where it's at!
Click to expand...

At some point, EDC can't keep up with the errors. When I put the memory speeds above 1325, I get a flashing green screen that last exactly 1 frame, and it occurs every 10 seconds or so. If I push it to 1335, I get instant hardlocks with any 3D program.
 
mrrogers747 said:
Nice review Brent. I have the same card myself and think it's a beauty! I've only done some mild overclocking at stock volts and hit 930 core but only 1240 on the memory. The performance when going to 1250-1260 on the memory does start to degrade, but after that it definitely starts to artifact and when I hit over 1300 sometimes it can even lock up. I thought the EDC was supposed to eliminate that? Not complaining because the core is where it's at!
Click to expand...

I can hit stable and with fps increase on 1300..

on core frequency, there is some odd issue, where I get performance downgrade at 1050Mhz with, but at 1060 I have performance increase again from 1045...its kind of odd there..

not quite sure what exactly is going on with the chip, but thats how my test goes..

and I find out that my other 5870 is 1300 Memory stable in Furmark for an hour, but in actual application/software where it have hardlock within first 10-20 min gameplay.

from that point, I think furmark is a bad stability test tool for memory OC.
 
Brent_Justice said:
AMD has also rather had some bad luck with recent games launched lately:

Batman - NV slant is obvious
NFS Shfit - AMD claims not getting the performance they think they should be getting
RE5 - Current bug on 5000 series possibly causing loss of performance
Click to expand...

You can't blame "luck". If ati can't get the drivers to work properly and don't spend enough time working with developers that's their own fault. No more excuses (blaming nvidia/fate/whatever) - they need to be much more pro-active with game developers. If they get a programme as wide reaching as TWIMTBP I bet that "bad luck" will dissappear.
 
mrrogers747 said:
Nice review Brent. I have the same card myself and think it's a beauty! I've only done some mild overclocking at stock volts and hit 930 core but only 1240 on the memory. The performance when going to 1250-1260 on the memory does start to degrade, but after that it definitely starts to artifact and when I hit over 1300 sometimes it can even lock up. I thought the EDC was supposed to eliminate that? Not complaining because the core is where it's at!
Click to expand...

Once the EDC starts correcting errors the preformance goes down. Push it too far and you eventually you have more errors than you can over come.

It's like doing a worksheet of 100 simple math problems and being given 10 mins to do it. Then being told what answers are wrong. At 10 mins you probably aren't going to get any wrong. Get up to 5 mins and you are going to get a couple of them wrong but are going to have to spend an extra min fixing your errors which is ok because you are now only spending 6 mins. Give you 2 mins and you might only get 20 right and 80 wrong. Then you are going to take another 10 mins to finish all the answers (because again, youa re goin at the 2 min pace). Which is how preformance degrades again. Then we'll give you 10 seconds and you aren't going to get any right and you'll never get anywhere.

The real "problem" with EDC is it's not as easy to overclock with it. You need to run something kind of synthetic benchmark to know when to stop. By the time errors start showing up you are WAY to far.
 
sbuckler said:
You can't blame "luck". If ati can't get the drivers to work properly and don't spend enough time working with developers that's their own fault. No more excuses (blaming nvidia/fate/whatever) - they need to be much more pro-active with game developers. If they get a programme as wide reaching as TWIMTBP I bet that "bad luck" will dissappear.
Click to expand...

its only the matter of time for them to fix the driver problem.

overall, I doubt ATI have more cash than nVidia to do TWIMTBP..

btw, in games like Crysis, ATI did help with their development, it just they don't have their logo on it...it doesn't mean they didn't work with them...
 
sbuckler said:
You can't blame "luck". If ati can't get the drivers to work properly and don't spend enough time working with developers that's their own fault. No more excuses (blaming nvidia/fate/whatever) - they need to be much more pro-active with game developers. If they get a programme as wide reaching as TWIMTBP I bet that "bad luck" will dissappear.
Click to expand...

I agree. But it seems that they're trying to get more involved with game development, at least with DX11. Since they have the only usable DX11 hardware, the game makers are having to use them as the reference point, and AMD personal have been at their disposal. I can also see them pushing OpenCL/ComputeShader Bullet physics to anyone willing to listen. Hopefully they keep up with this increased presence.
 
SicKlown42012 said:
I agree. But it seems that they're trying to get more involved with game development, at least with DX11. Since they have the only usable DX11 hardware, the game makers are having to use them as the reference point, and AMD personal have been at their disposal. I can also see them pushing OpenCL/ComputeShader Bullet physics to anyone willing to listen. Hopefully they keep up with this increased presence.
Click to expand...

I don't know about that though. Anyone who is farther than 3 months out in release knows Nvidia's hardware will be to market by the time the game is released. Oh yeah and there is that giant pile of cash they keep handing out for the TWIMTBP badge.
 
Wow, great review as always. GTX285 really showed a lot of muscle in this review in batman, re5, etc. With this review, it basically shows the gtx285 and 5870 being equal performers. In benchmarks I ran the 5850 far surpassed that of the 4890, so I was surprised to see that the 5870 didn't wipe the floor with the gtx285.
 
For some reason I thought I remembered at launch the 5870 was expected to be a competitor to the GTX 295 not GTX 285, but that is defiantly not happening here. Maybe I am just in need of more coffee.
 
Tom128 said:
For some reason I thought I remembered at launch the 5870 was expected to be a competitor to the GTX 295 not GTX 285, but that is defiantly not happening here. Maybe I am just in need of more coffee.
Click to expand...

Well not necessarily. There are plenty of games where a 5870 is considerably faster than a gtx285. Fallout 3 is a good example where it's considerably faster than a gtx285, and left 4 dead is another one. Now we know that gtx285 drivers are matured, and those are relatively old games, so this to me shows that ATI has the hardware...just not the drivers...

but give ati some slack, these cards just barely launched.
 
Only way i can see it.

Dont buy a 285 since 5870 is futureproof, cost the same, packs more power.
 
/me puts on the graph nazi hat

please cease and desist the use of any graphs not having a ZERO point (as used in the 'Overclocking' page) me HATES it :rolleyes:

seriously, i was like 'WHOA look at the overclocked....uh...NEGLIDGIBLE difference"
 
PS3 said:
Well not necessarily. There are plenty of games where a 5870 is considerably faster than a gtx285. Fallout 3 is a good example where it's considerably faster than a gtx285, and left 4 dead is another one. Now we know that gtx285 drivers are matured, and those are relatively old games, so this to me shows that ATI has the hardware...just not the drivers...

but give ati some slack, these cards just barely launched.
Click to expand...

You have a good point with Fallout 3, but saying it's faster in LFD is saying that a lambo is faster than a civic in a school zone, it doesn't mean much. :p
 
The fact that a 4890 is just as "fast" as a 285 in crysis. which was VERY much "best at nvidia" is no longer true, in fact, it runs GREAT on ati hardware now!
 
AtkinS said:
The fact that a 4890 is just as "fast" as a 285 in crysis. which was VERY much "best at nvidia" is no longer true, in fact, it runs GREAT on ati hardware now!
Click to expand...

its been like that since 8.12 driver..
 
I don't know if HardOCP testing rigs typically run Everest or other sensor tools in the background, but I have noticed an issue with my EAH5870 using Everest and either Asus SmartDoctor or MSI's Afterburner. This issue has been discussed over in the Guru3D forums pertaining to several different 5870 manufacturers, but apparently there is an issue with the Volterra VRM and multiple programs trying to poll the chip for monitoring data. I typically run Everest so I can output a custom set of stats to my G15 LCD screen and noticed that if I used this in conjunction with either the Asus SmartDoctor or MSI Afterburner the GPU Vcore will randomly spike to 1.69V or even up to 2V. This is definitely not measurement bug because the core temp rapidly increases during these events. The issue is easier to reproduce using MSI Afterburner since it maintains a history for a window of time and maintains the absolute min / max values encountered while in operation.

I am not sure about the Asus SmartDoctor, but the MSI Afterburner is based upon RivaTuner and Unwinder was more than a little defensive when users blamed his tool. He did however add the following comment: "I repeat for 1000th time that Afterburner is NOT writing data in runtime to VRM. To have false voltage set data must be WRITTEN there. The only periodic WRITES to VRM, which can potentially having a place in the system, are performed by VRM temperature monitoring tools (VT1165 architecture requires slave chip selection, which is performed by write to VRM registers). VRM access collision and erroneous data write can be performed at that side only."

I can confirm that my issue went away once I uninstalled everything but Everest and the base ATI driver (leaving the OC mode locked). If I understand Unwinder's reply, it sounds like there is a potentially dangerous situation that can arise if more than a single VRM monitoring tool is being used as the two tools can potentially collide and write erroneous data causing GPU Vcore increases. In my case the random spikes occurred mainly in 2d, but occasionally in 3d mode. The main reason I even bothered with a tool like MSI Afterburner was for the custom fan profile to allow for a bit more aggressive fan curves and I made NO attempt to OC or increase the voltage through these tools. I would be interested if someone at HardOCP could confirm this or if anyone here in the forums would be willing to test, baring in mind that the voltage spikes jump to dangerous 1.69V and possibly even up to 2V. Several over at Guru3D thought they had defective cards, but are seeing the issue return on their replacement cards if the same conditions are repeated (i.e. Win7, MSI Afterburner, Everest).

The thread I was referencing at Guru3D can be found here: http://forums.guru3d.com/showthread.php?t=305640&page=5 ,
 
Last edited:
Thanks for the detailed write up Mailichite. That goes a long way to explaining the issues I was having.
 
Brent_Justice said:
NFS Shfit - AMD claims not getting the performance they think they should be getting
Click to expand...

I know your test systems kick the hell out of my old-ass processor/mobo/RAM combo, but using lesser hardware, I hope that claim from AMD is still the truth. I'm kind of speculating it's not directed at 4890 owners.

Nurburgring is a track where cars disappear quickly around the many sharp corners and aside from some close corners involving 2-3 cars, the field is usually much more spread out than on many of the other circuits in the game. Not many fans hanging out, either. The 4890's performance is greatest on this course; dropping under 30fps @ 1920x1080 w/4xAA enabled is a rarity on my rig. That's awesome and in my opinion, Shift delivers one of the best looking 'Ring runs available on any console or PC.

However, with your 'tested at' settings, I would've loved to have seen [H]'s results using 10 minutes worth of say, the Tokyo circuit. With a bunch of cars in the lead and a complete mob in the grandstands, along with a great number of roadside fans, the framerate tanks for me. I've got to back off on AA and some detail settings to keep the game playable at over 30fps on many circuits. Unfortunately, that means I'm playing with decreased 'stuff that doesn't really matter' settings on the courses my card can easily handle as well, since I don't want to be switching detail settings between races. That first turn is important, especially when the race is close...unless you're in first, of course! :D Just to know where I'm at, we'll call ~27fps minimum my real lowest acceptable threshold for Shift.

I was hoping to discover for certain that I'm just a whiner with a slow processor incable of efficiently calculating the CPU-based physics. Someone who should go i7 and a buy a 5870 or STFU and just play at the lower settings. I've gone with the latter solution, for the most part. :p

I noticed you mentioned hardware PhysX earlier in the article, mostly in a 'detrimental to performance' kind of way and not at all in a 'multicard setup' kind of way. Does the [H] notice any difference when adding a dedicated PhysX processor into the mix with a 5870 and NFS: Shift? Or in combo with a 285GTX, for that matter. And speaking unofficially, of course. ;) Since it's not mentioned, I'm assuming not but I have to ask. I haven't tried an 8800GT for PhysX at all with my 4890, yet, and I've been hoping for someone to suggest a game which I play that would benefit from me swapping my Corsair 520 PSU out for my 620. It just looks like alot of playing with wires for no real gain, to me.

Good review, as always.
 
Ok, I'll substitute far cry 2 in place of L4D, sheesh, :p . There really is a lot of instances where the 5850 is faster than a gtx285..in older games. And since the GT200 series drivers should be well matured in these games...the only logical reason why a gtx285 is competing with a 5870 in a few games is that AMD drivers are not up to par.

HD5850-49.jpg
 
snaggletooth said:
I noticed you mentioned hardware PhysX earlier in the article, mostly in a 'detrimental to performance' kind of way and not at all in a 'multicard setup' kind of way. Does the [H] notice any difference when adding a dedicated PhysX processor into the mix with a 5870 and NFS: Shift? Or in combo with a 285GTX, for that matter. And speaking unofficially, of course. ;) Since it's not mentioned, I'm assuming not but I have to ask. I haven't tried an 8800GT for PhysX at all with my 4890, yet, and I've been hoping for someone to suggest a game which I play that would benefit from me swapping my Corsair 520 PSU out for my 620. It just looks like alot of playing with wires for no real gain, to me.

Good review, as always.
Click to expand...

NFS shift does not support GPU HW ACCEL PHYSX.

Runs only on cpu due to how the devs made the game.
 
PS3 said:
Ok, I'll substitute far cry 2 in place of L4D, sheesh, :p . There really is a lot of instances where the 5850 is faster than a gtx285..in older games. And since the GT200 series drivers should be well matured in these games...the only logical reason why a gtx285 is competing with a 5870 in a few games is that AMD drivers are not up to par.

<IMG>
Click to expand...
hehe :D

Yeah, the 5870 is the better card, no question.
 
Nice review, but one thing puzzles me.
As an avid ARMA2 player, I cannot by the death of me understand your settings for that game.
Visibility: 1600 meters?
Visibility: 1000 meters?

No one in my squad (or that I know who plays ARMA2, uses such a low setting...IMHO it's unplayble...3000 meters is a minimum, then you adjust the other settings accordingly.



Resolution : 1600x1200
Visibility : 3500
Texture detail : Normal
Video Memory : Very High
Anti-aliasing : High
AF: Very High
Terrain Detail : Normal
Object Detail : High
Shadow Detail : High
Post Process Effects : High


Other game odities:
Setting shadow detail to High (or Very High) can give better performance than normal or low settings.
 
Nice review.
It would have been nice to also throw in either the 295 or 4870x2 as a reference .
 
You must log in or register to reply here.
Back
Top