ATi 4870/4850 Review Thread

[Janooo]

Originally posted by: Sylvanas

Originally posted by: chewietobbacca
FYI 4870 review at guru3d finally up

http://www.guru3d.com/article/...-hd-4870-review--asus/

This is an interesting bit of info:

GPU Computing -- Much like NVIDIA just announced with the help of CUDA, ATI (AMD) recently announced cooperation with Intel's HAVOC engine. Though currently far less substantial, PhysX calculations over the GPU are in the work. As it works right now (example debris/cloth) physics calculations are computed over the CPU with games that support the HAVOK API. AMD is working on moving these functions to the GPU. Thus have the stream processors (shader engine) compute these functions.

It's work in progress and during a recent press-briefing we asked when we can expect driver support for GPU HAVOK physics. The answer was unfortunately a bit cold. It could be a matter of two months, yet also easily be the end of the year. Fact remains though that the Series 4000 do support the feature and AMD's driver team is working on it.

I'm still waiting for PhysX to actually do something instead of improve a 3Dmark score but these developments might be good later on down the track, Havok vs PhysX.

PhysX on 3850 here.

 

[chizow]

Originally posted by: BFG10K
No it isn?t. If the game is limiting frames then the driver can?t make them go faster by adding multiple GPUs into the equation.

I asked you for such examples and you?ve failed to deliver. Furthermore I provided two UT3 engine games that demonstrated they showed no such thing.

The driver could easily be disabling frame capping in SLI/CF if there was any internal capping/Vsync enabled by default.

And about the UE3 engine games, you never responded if UT3 had frame capping removed by default via patch or if it requires an .ini change. I'm pretty sure Bioshock does not have frame rate smoothing and never said otherwise. But its also not the only game that has some kind of capping going on, as both Assassin's Creed and Witcher exhibit similar behavior (in AT's review for sure).

http://www.anandtech.com/video/showdoc.aspx?i=3341&p=16

We can see AT?s results are capped at 60 FPS or thereabouts and neither CF/SLI is removing that cap (how on earth could it?).

If you have evidence that demonstrates multi-GPU/SLI is removing the cap - say all single cards are limited to ~60 FPS while multi-GPU is far higher - you need to post it up or your claims are yet again unfounded.

Witcher @ 1920 and 1680
Its clearly obvious there is some kind of frame capping going on, but the 4870CF manages to average 72 FPS. Derek also posted a comment that verified this saying

it looks like the witcher hits an artificial 72fps barrier ... not sure why as we are running 60hz displays, but that's our best guess. vsync is disabled, so it is likely a software issue.

The results at 1920 are convincing, but the results at 1680 confirm it with all the SLI configurations managing to average higher than 60FPS, which shouldn't be possible if there's some 60FPS software cap or hidden Vsync function. Again, I think it simply comes down to timing differences/sync issues with FRAPs or a driver override allowing slightly faster, but superficial frame rates.

Right, right, but nobody is discussing those situations. We all accept and understand what external limitations look like and the ramifications of such.

Its still valid when someone points to a review and says "Oh look, the 4870 ties the GTX 280, and in some cases, even outperforms it" and the results are say, 58.9 FPS for the 280 and 60.1 FPS for the HD4870. Again, this is not to say the 4870 is a bad part, its not, its a fantastic part, but its clearly not a good indication of relative performance if both parts are hitting a CPU/frame cap.

No. it?s not clear at all. The only thing clear would be the 4870 CF being bottlenecked because it?s not much faster than the 4850 CF.

Hothardware, multiple GPU, single GPU 1920 and 2560
This should be easier to visualize then. This review does an excellent job because it compares just about every relevant single card and multi-GPU configuration as well, but also puts 1920 and 2560 results side by side. Now, its a lot to digest, and I'm not going to go through everything to try and explain how I get to my conclusion, but its pretty clear there is CPU bottlenecking occurring even at 2560 once you move into the multi-GPU parts. Some of it may be due to poor scaling, but you'll see that many of the faster single-card solutions (GTX 280 and 4870)are near cap at 1920 and don't scale in CF/SLI. But at 2560, they drop below cap, but multi-GPU scaling brings them back to a CPU capped frame rate. If you look at the other games they are similar and often more pronounced.

That I agree with, but again that was never under contention.

Well again, if you look at something like AT's review and it shows <10% difference between the GTX 280 and 4870 and just about every other recent card then I think it is relevant. Its also extremely relevant for anyone considering 4870CF or 4870X2.

I don?t think you understand what micro-stutter is especially since you?re claiming it can somehow remove game framerate caps. Micro-stutter is simply the uneven distribution of frames; it doesn?t impact a game?s frame cap.

I never said micro-stutter removed game framerate caps, micro-stutter is the result, not the cause. I already showed one example where CF/SLI is breaking 60FPS when all other cards are capped with Witcher, Oblivion and Assassin's Creed in AT's review. I don't know the cause, but annihilat0r's post provides insight on the timing that would result in a higher average FPS with multi-GPU.

In 10 ms a single card might get 3 frames while a multi-card might get 5 frames. The difference - and this is what causes the micro-stutter - is the single card?s frames are distributed more evenly than the multi-card?s.

But this in no way invalidates the fact that it?s objectively provable the multi-GPU has a higher framerate because it?s rendered more frames in the same amount of time, something that can?t happen if a framerate or CPU limitation is in place.

Sure you can, you can increase the number of pre-rendered frames which means more input lag, but higher frame rates for the multi-GPU solution. Again, this can easily be regulated by the driver or profile even in frame capped or CPU limited situations.

Well that?s definitely a serious oversight of AT not to mention this. Still, they can?t possible be doing it in all games:

http://www.anandtech.com/video/showdoc.aspx?i=3341&p=14

Even an LCD set 75 Hz would top out at 75 FPS, something that clearly isn?t shown on that graph.

And once again, I never said it was doing it in all games but when there's serious frame capping and CPU bottlenecking in 4 of the 7 games (AC, Oblivion, Crysis, Witcher) in your test suite and people are drawing conclusions from that, I think its pretty clear we're not seeing accurate performance differences between the parts as a whole.

Pardon? Let?s quote what you said about UT3 based games:

I then produced UT3 and Bioshock benchmarks that debunked your claims. If you?re now saying you don?t think the framerate cap is in effect for those games you need to retract that claim, but don?t go around pretending you never mentioned such examples.

I mentioned UT3 as proof there was frame capping occurring in games and you reacted as if it was a problem that didn't exist. I proved evidence to the contrary and asked if it was patched out in UT3 or if you needed to still manually disable it. You still haven't replied. If it needs to be manually disabled then I'm absolutely correct in saying you have to assume there may be similar instances occurring in other games that may be unknown or can't be disabled, especially when I've already shown 2 others in AT's review that seem to have similar FPS caps (Witcher and AC). Its also a verifiable fact that frame rate smoothing is enabled in GoW and Mass Effect by default. I never said Bioshock was frame capped and specifically said I was pretty sure it wasn't.

After all this what are you trying to say exactly? That graphs that are close together demonstrate an external limitation? Well no sh!t given everyone here agrees with that.

So you think the 4870 is as fast as a GTX 280 even in cases of frame capping/cpu limitations?

People are taking issue with your original comment:

No I think people are taking issue because they don't full understand the issue and see it as an attack on the 4870, but its certainly relevant if they were considering 4870CF or 4870X2, even moreso than just a single GTX 280.

This posted in the context of a 4xxx thread makes it sound you?re claiming when the 4870 is competitive it?s somehow irrelevant because of some external limitation, most of which are fictional.

No, I never maligned the 4870's performance, I'm simply pointing out that direct comparisons between a GTX 280 and 4870 may vary well be skewed due to CPU limitations or frame caps.

Between that and you misquoting Anandtech above along with your comments in other threads ?the 4850 has been reduced to mediocrity because of the 9800 GTX++? and ?when CF is faster than the GTX280 both cards are fast enough so it doesn?t matter? really speak volumes.

Misquoting of Anandtech? Really? Where? Oh, you took the bait of some guy highlighting the first 3 words instead of the next 3 words as a misquote. Good job.

I never said the 4850 was mediocre because of the 9800GTX+, I said it was mediocre compared to 7-8 month parts that brought far more in terms of performance and cut far deeper into existing prices relative to performance. Are you going to argue that's not the case? Do you think 4850 was a more impressive part in terms of price/performance than the 8800GT? Honestly, people who even attempt to make that comparison have no sense of perspective.

My reference to CF vs GTX 280 was to show there may not be much point in upgrading to CF at all or expecting vast performance gains with the 4870X2 due to bottlenecking.

 

[chizow]

Originally posted by: BFG10K
Re: Assassin's Creed.

http://www.anandtech.com/video/showdoc.aspx?i=3320&p=9

Both single and multi-GPU are scoring well over 60 FPS and it's from AT's own analysis of the game.

Based on those figures I don't think the game is capped at 60 FPS at all but rather they were probably running vsync on a 60 Hz LCD.

Well again, I specifically asked Derek if they had Vsync on in the 4850 review (and GTX 280, 9800GTX+ etc since they're all archived results) and he said they do not force it off. I remember very clearly why they don't anymore, as it resulted in worst performance in Crysis. From the previous review it does look like AC isn't always capped, but its also clear different reviewers can have different results based on obscure settings that you may or may not be able to change.

From the 4870 Review comments:

CPU bottlenecking, Vsync, Frame Caps and Frame Rate Smoothing? by chizow, 2 days ago

It looks like there seems to be a lot of this going on in the high-end, with GT200, multi-GPU and even RV770 chips hitting FPS caps. In some titles, are you guys using Vsync? I saw Assassin's Creed was frame capped, is there a way to remove the cap like there is with UE3.0 games? It just seems like a lot of the results are very flat as you move across resolutions, even at higher resolutions like 16x10 and 19x12.

Another thing I noticed was that multi-GPU seems to avoid some of this frame capping but the single-GPUs all still hit a wall around the same FPS.

Anyways, 4870 looks to be a great part, wondering if there will be a 1GB variant and if it will have any impact on performance.

Reply
RE: CPU bottlenecking, Vsync, Frame Caps and Frame Rate Smoothing? by DerekWilson, 2 days ago

the only test i know where the multi-gpu cards get past a frame limit is oblivion.

we always run with vsync disabled in games.

we tend not to try forcing it off in the driver as interestingly that decrease performance in situations where it isn't needed.

we do force off where we can, but assassins creed is limiting the frame rate in absentia of vsync.

not sure about higher memory variants ... gddr5 is still pretty new, and density might not be high enough to hit that. The 4870 does have 16 memory chips on it for its 256-bit memory bus, so space might be an issue too ...

Reply
RE: CPU bottlenecking, Vsync, Frame Caps and Frame Rate Smoothing? by JarredWalton, 1 days ago

Um, Derek... I think you're CPU/platform limited in Assassin's Creed. You'll certainly need something faster than 3.2GHz to get much above 63FPS in my experience. Try overclocking to 4.0GHz and see what happens.

Looks like Jarred caught onto it as well. Maybe one of the other mods can pass it on to Anand/Derek, as it seems to be a non-issue for some.

 

[anindrew]

Any ideas about when the 4870x2 is coming out and it's rumored price?

I'm planning out a new system, so of course the 4870 looks great to me. But if the 4870x2 isn't unreasonably more, I may go for it instead.

 

[Aberforth]

Originally posted by: anindrew
Any ideas about when the 4870x2 is coming out and it's rumored price?

I'm planning out a new system, so of course the 4870 looks great to me. But if the 4870x2 isn't unreasonably more, I may go for it instead.

It's coming in 4 to 6 weeks, it will probably be priced around $499 to $550 - can't say for sure. To be on the safer side you might want to buy a 1kw PSU....

 

[bryanW1995]

<---ignoring chizow. hmmm, I might make that into my sig. Seriously, how many of us other than bfg are even trying to reason with chizow any more? A lot of people wanted 3870 to beat 8800gt but they went away as more and more benchies proved the g92 superiority. Guess what? A lot of people wanted gt200 to beat r(v)7xx, but the value is so much greater with ati this time around that only the blind fanboys are even debating the issue at this point. Nvidia will be back, they'll probably stomp the crap out of ati next round or at least be much more competitive, but for now ati is the card to own. period.

 

[zod96]

I bit the bullet and got a 4870 from fxvideocards for $299 It was between that and the Gtx260. But after reading the reviews on both cards, theirs no doubt in my mind that the 4870 is a better all around card for $100 less...

 

[bryanW1995]

not only that but you do have an x48 mobo, yet another advantage for ati atm.

btw, 4850's are now as low as $170 AR shipped at newegg. most of the brands have mir's on them. is this a push to get sales very high from the beginning, a response to the 9800gtx price drop, or a response to the 4870's strong price/performance???

 

[ShadowOfMyself]

Originally posted by: bryanW1995
not only that but you do have an x48 mobo, yet another advantage for ati atm.

btw, 4850's are now as low as $170 AR shipped at newegg. most of the brands have mir's on them. is this a push to get sales very high from the beginning, a response to the 9800gtx price drop, or a response to the 4870's strong price/performance???

Id say 9800 GTX, considering at 300$ the 4870 is 50% more expensive than the 4850, yet, its not 50% faster

So the 4850 really has the best value/performance... The thing is the 4870 really stands out for having ZERO competitors on its price range, and of course, besting a card 100$ more expensive

 

[bryanW1995]

it could also be that amd wants to kill off 8800gt once and for all, too. with 4850 at $199 it doesn't do that, at $169 it makes it very tough to recommend an 8800gt to somebody now, while also making it nearly impossible for 9800gtx(+) to compete. figure that the 9800gtx and 4850 will be about the same price AR, that gives 4850 a bunch of wiggle room to have oc/aftermarket coolers and still be under $200, leaving 9800gtx+ in no man's land...

 

[KeithTalent]

Probably a stupid question, but is it safe to say one 4870 will cream my 2900XTs in Crossfire? I've been reading a tonne of the reviews, and that's what I've gathered, but just looking for confirmation. I guess it may depend on the game as well and how well the Crossfire scales, but just wondering in general if this holds true.

Thanks!

KT

 

[epphllps]

Just got my HIS HD 4870 up and running and did some quick benchmarks before I head out of town.

3DMark Vantage

3DMark06

 

[chewietobbacca]

Yeah it will beat 2900XT in CF pretty well. The 4850 alone matches the 3870X2 which is probably faster than the 2900XT's in CF so a 4870 should beat it soundly single card.

 

[KeithTalent]

Originally posted by: chewietobbacca
Yeah it will beat 2900XT in CF pretty well. The 4850 alone matches the 3870X2 which is probably faster than the 2900XT's in CF so a 4870 should beat it soundly single card.

As I figured, thanks. Time to sell these puppies and upgrade.

KT

 

[pcgamer321]

How much would a 2900xt sell for?

 

[JPB]

Originally posted by: pcgamer321
How much would a 2900xt sell for?

Well, in my opinion. Right now since a HD 3870 is around $155 on average. I would say since it would be used....about $120.

Of course, Id be willing to give you that for it :thumbsup:

 

[flexy]

why would you even need TWO 4870s or TWO GTX 260/280....come one people, JUSt for playing crysis w/ relatively good FPS?
Otherwise a *single* 4870 (or GTX if you must) would be suitable for anything you throw at it, at least under common resolution (say 1600)...WITH all eye candy on.

The only legitimate reason to get any of those cards in CF or SLI would be monster resolutions and monster-AA...or just doing it since with a 4850 it's "so cheap"

I wouldnt say this but even my overclocked 8800GTS handles as good as everything i throw at her....so temptation for 4870 is big..but "real life" benefit would be rather "meh".....but TWO?

Add: 4870 is EU203 *shipped* right now...i am literally hovering over the "buy now" button.

The ONLY thing holding me back is the remote chance that NV *might* still release a stereo driver for vista supporting my 50" samsung 3D plasma and LCD shutter glasses....i would seriously **** my pants getting a 4870 now and then seeing NV releasing a driver for those 3D modi...thats the ONLY thing holding me back right now.

 

[BFG10K]

The driver could easily be disabling frame capping in SLI/CF if there was any internal capping/Vsync enabled by default.

Sorry, no, it can?t. You keep repeating this but repeating it doesn?t make it right. The driver can?t remove a game framerate cap. If the game is limiting frames every 5 ms the driver can?t go in there and adjust the game?s tick loop.

About the only thing the driver could do is ignore vsync in which case the driver is broken; but then reviewers shouldn?t be running with vsync in the first place as it makes their tests broken.

And about the UE3 engine games, you never responded if UT3 had frame capping removed by default via patch or if it requires an .ini change.

I provided benchmarks that demonstrated it didn?t affect any of the current round of benchmarks and I?m still waiting for your evidence otherwise. Until you provide such evidence please do not bring up this issue again or I will assume you are trolling.

I'm pretty sure Bioshock does not have frame rate smoothing and never said otherwise.

LMAO.

Witcher @ 1920 and 1680
Its clearly obvious there is some kind of frame capping going on, but the 4870CF manages to average 72 FPS. Derek also posted a comment that verified this saying

it looks like the witcher hits an artificial 72fps barrier ... not sure why as we are running 60hz displays, but that's our best guess. vsync is disabled, so it is likely a software issue.

The results at 1920 are convincing, but the results at 1680 confirm it with all the SLI configurations managing to average higher than 60FPS, which shouldn't be possible if there's some 60FPS software cap or hidden Vsync function. Again, I think it simply comes down to timing differences/sync issues with FRAPs or a driver override allowing slightly faster, but superficial frame rates.

So what if it averages 72 FPS? That's the game's cap according to Derek.

60 FPS cap? What the hell are you talking about? You just quoted Derek saying there is a 72 FPS cap!

You're saying what exactly? That CF is breaking your fictional 60 FPS cap by getting 72 FPS (which just happens to be the actual cap Derek described), thereby proving multi-GPU can "work around" game caps?

LMFAO.

Its still valid when someone points to a review and says "Oh look, the 4870 ties the GTX 280, and in some cases, even outperforms it" and the results are say, 58.9 FPS for the 280 and 60.1 FPS for the HD4870. Again, this is not to say the 4870 is a bad part, its not, its a fantastic part, but its clearly not a good indication of relative performance if both parts are hitting a CPU/frame cap.

But there are plenty of examples were they aren?t hitting a CPU cap but you?re simply making sweeping generalizations. Take the CoD 4 results: the 4870 provides 90% of the GTX280?s performance at less than half the cost and half the VRAM. I don?t know how anyone can claim that is a CPU limitation.

Now, its a lot to digest, and I'm not going to go through everything to try and explain how I get to my conclusion, but its pretty clear there is CPU bottlenecking occurring even at 2560 once you move into the multi-GPU parts.

I agree, but that was never under contention.

I already showed one example where CF/SLI is breaking 60FPS when all other cards are capped with Witcher, Oblivion and Assassin's Creed in AT's review.

Where? Your link http://www.anandtech.com/video/showdoc.aspx?i=3341&p=21 shows no such thing. Furthermore since when is Oblivion capped at 60 FPS?

Sure you can, you can increase the number of pre-rendered frames which means more input lag, but higher frame rates for the multi-GPU solution. Again, this can easily be regulated by the driver or profile even in frame capped or CPU limited situations.

Based on your response to what you quoted I don?t think you understand what you quoted. I also don?t think you understand what micro-stutter is or even how multi-GPU scales.

Pre-rendered frames is purely a function of the driver. If the game tick is limiting frames to begin with pre-rendering won?t cause the tick to be lowered.

And once again, I never said it was doing it in all games but when there's serious frame capping and CPU bottlenecking in 4 of the 7 games (AC, Oblivion, Crysis, Witcher) in your test suite and people are drawing conclusions from that, I think its pretty clear we're not seeing accurate performance differences between the parts as a whole.

You?ve made one sweeping generalization and each time we have this discussion you include more and more games. CPU bottlenecking? Frame capping? In Oblivion at 2560x1600? Please tell me you?re joking.

Also I just thought you were telling us Oblivion has a 60 FPS cap but now it?s CPU bottlenecking? And in the next post no doubt you?ll deny you ever mentioned Oblivion and claim I?m making up scenarios. :roll:

I mentioned UT3 as proof there was frame capping occurring in games and you reacted as if it was a problem that didn't exist. I proved evidence to the contrary and asked if it was patched out in UT3 or if you needed to still manually disable it. You still haven't replied.

I provided several benchmarks the debunked your claim and until you provide relevant benchmarks to the contrary do not bring up this topic again.

So you think the 4870 is as fast as a GTX 280 even in cases of frame capping/cpu limitations?

No, I think your appraisal of the situation is overblown. I also think you constantly shift the goal-posts to vsync to capping to CPU limitations to multi-GPU whenever it suits your agenda without ever actually providing any evidence to back your claims.

No, I never maligned the 4870's performance, I'm simply pointing out that direct comparisons between a GTX 280 and 4870 may vary well be skewed due to CPU limitations or frame caps.

It might be, but then the situations you describe aren?t the ones people are drawing inferences from.

Misquoting of Anandtech? Really? Where? Oh, you took the bait of some guy highlighting the first 3 words instead of the next 3 words as a misquote. Good job.

Yep, absolutely. This is what you said:

Well, I'd say its a bit premature to say GT200 is a flop, if you look at this latest round of reviews I think you'll see that there's quite a bit of CPU bottlenecking and frame capping going on, even at higher resolutions like 16x12 and 19x12. That's not to say 4870 isn't a great part, it is, but clearly a large part of the reason its so close to GTX 280 is because of CPU bottlenecking.

For example, quoted from the AT article:

See the highlight, your claim of CPU bottlenecking? Now let?s see what you quoted from Anandtech to ?back? that claim:

Performance of the Radeon HD 4870 continues to be strong, but because of the frame rate cap we're not able to see if the GTX 280 could stretch its legs further and eventually outperform the 4870. In actual gameplay, the 4870 and GTX 280 appear to be equals.

Nowhere in that quote does it mention CPU bottlenecking.

Now let?s look at the missing section of the quote, the one you conveniently left off when using it as ?evidence? for you claims of CPU bottlenecking:

Assassin's Creed is capped at near 60 fps, which is why we see most cards reaching but not significantly exceeding that marker.

This is why people don?t take you seriously. You chop and change whenever it suits your agenda, mis-quote and then claim you never made such claims when called out.

I never said the 4850 was mediocre because of the 9800GTX+,

Here?s the quote from you ?Case in point is the GTX+ that needed a clock speed boost to push 4850 back into mediocrity."

You?re at the stage now of denying things that you said in the past.

Looks like Jarred caught onto it as well. Maybe one of the other mods can pass it on to Anand/Derek, as it seems to be a non-issue for some.

Actually it looks like half of those quotes refute what you?ve been claiming. You were claiming game caps in Assasin?s Creed when Jared points out CPU limitations. You were claiming Anandtech don?t force off vsync when in actual fact they don?t force it off in the driver (but do so in the game, again selective quoting on your part). I could go on but honestly it?s a waste of time with you.

 

[BLaber]

Lol chizow stop it , PLss... , your comments (...arguments) don't provide any use full information that may be help full in making the next upgrade decision , so S#F* ..

 

[srp49ers]

I upgraded from my ati 3780 to the 4850 yesterday. And when I go play Hd videos the cpu utilization is very high like 70-80% and the video is not as smooth as with my 3870.
the cpu usage with the 3870 was around 40%.

Anyone else having this problem. could it just be drivers. Im using the hotfix version.

 

[zod96]

Anyone know if a artic cooling s1 will fit on the 4870?

 

[Elcs]

Cancelled my order for my HD4870. A store issue, I was guarunteed the card would be in on Friday and shipped that day and on that basis I placed my order. Company didnt get them in, refused to admit to any blame or offer me any help so I cancelled. I could have got one that day for the equivilent of $38 more but when you are paying the equivilent of $380 already, not worth it.

 

[Hunt3rj2]

The people who are fanboys of great performance to price (like me) are definitely thinking ATI has won this match. Hoping for both companies are making up some even better GPU designs for the next cycle.


I am happy that ATI has won for once though, if the competition always loses then eventually there will be none. And then you'll have a monopoly situation where graphic card progress slows to a standstill.

 

[jzodda]

Do you think performance in crossfire would suffer if the second PCI Express slot was not full speed? I have an Abit AB9 QuadGT and would like to get more life out of it if I could. It has:

One PCI Express x16, and one PCI Express x4 (x16 physical)

Will that degrade performance in the 2nd slot to the point that makes it unwise to get 2 cards?

 

[Sylvanas]

Originally posted by: jzodda
Do you think performance in crossfire would suffer if the second PCI Express slot was not full speed? I have an Abit AB9 QuadGT and would like to get more life out of it if I could. It has:

One PCI Express x16, and one PCI Express x4 (x16 physical)

Will that degrade performance in the 2nd slot to the point that makes it unwise to get 2 cards?

Not worth Crossfire on a 16x4 PCI-E 1.1 board- Source and that was done 16x16 and 8x8 2.0. It would be best to stick to a single 4870, and if you absolutely must go Crossfire then best to upgrade the motherboard aswell to 16x16 2.0.