HD 4870 review thread.

What makes you think 4870 is better at DX10 ? With respect to Bioshock I think AMD gets high utilization out of the vec5 ALU's and that could explain their success there.

 

X-bit have published their detailed performance review.

http://www.xbitlabs.com/articles/video/display/ati-radeon-hd4800-games_12.html

 

A few months ago BS DX10 performance sucked (on 600 and 670). DX9 was good.

The main reason is no hack for MSAA under DX10, but BS is also less texture heavy as its other UE3 compatriots.


Heck, we should try benchmarking Roboblitz.

 

Well that goes against just about every other review Ive read.

 

How so?
4850/4870 soundly beat 9800GTX, sounds about right.

No GT260/280 or 9800GX2 to outshine it in this review (no particular reason given but presumably price related?)

 

Or they havent benched those cards yet. Hopefully they will have the GTX series review out soon for comparison.

 

UT3 performance was significantly improved in HD48xx hotfix / series 5 (just FYI)

 

Seriously?

XBIT


Techreport (IMO the best review site at present)


XBIT : 4870 = 70.3 | 9800GTX = 74.5
TECH : 4870 = 79.8 | 9800GTX = 56.1

We're not talking percentage points difference here, we're talking about one of those sites completely LYING. The 9800GTX is roughly the same power as a 8800GTX, yet XBIT have it scoring like its in SLI. Bullsh....

 

Lying or just using different motherboards & drivers

Individual results aside, overall result is the same: 4850 beats 9800GTX & 4870 even more so. Both at good price.

 

That part, at least, is truth. If you match up scores on Guru3d, 4870CF goes head to head with GTX280SLI. Am looking forward to my 4870CF setup replacing my 8800GTXSLI rig on Monday.

 

XbitLabs video test are conducted with Alpha-test AA enabled by the drivers, for both NV and ATi parts. I don't understand this attitude of benching - weird, but it's a fact.

 

Dooby, here's the test setup for each article: http://www.techreport.com/articles.x/14990/8

http://www.xbitlabs.com/articles/video/display/ati-radeon-hd4800-games_2.html

I think that explains the differences.

 

It clearly does NOT explain the undeserved delta between the 9800GTX and 8800GT, which according to computerbase.de, is about 85% of the 98GTX.

Xbit is either recycling scores, or making stuff up. Looking at the delta of the 9800GTX and 4850, I'd say the latter.

 

Agreed, no way in hell 9800GTX beats the 8800GT by 40%, something is going on here. That is the only review where I have seen the 9800GTX beat the 4870 at anything, and I have read almost all of them.

Is something going on with driver quality settings? I noticed Xbit is using high quality for both Catalyst and Forceware, anything in there that could reduce 4870 performance by this much? I noticed "Enable Adaptive Anti-Aliasing: On/Quality" for the ATI card, does that force supersampling transparency AA in the 4800s? "Antialiasing - Transparency: Multisampling" is pretty much free on Nvidia hardware, but it doesn't look nearly as good and it wouldn't be fair to compare like this.

 

Their 9800GTX numbers are definitely out of whack at Xbit on HL2, perhaps AA isn't working or something.

Also here is a 4800 series review on Linux

 

If Xbit wanna go towards High IQ on both, 9800GTX and HD 48x0, they should have enabled 8x MSAA, IMO.

Is Alpha test that causes this drop in performance on ATI hardware? If so, does it increases IQ that much to justify the performance penalty?

Or is the enabled Transparency AA that makes this difference? If so, NVIDIA and ATI produces comparable IQ at the settings used? If the answer for the second question is no, the results are meaningless, again IMO.

To much ifs for me. Anybody wanna please help? I don´t wanna think that xbit hand picked settings that would make ATI look worse against Nvidia cards, especially if these settings could, in the not too far away review from the same site, show GTX 280 and 260 in a better illusionary shape.

Thanks in advance!

 

It's not really that the ATI performance is so bad (look at the tech report numbers posted back 8 or 10 posts) as much as the 9800 performance is too good. The ATi hardware is about 10% worse in the xbit review, but the 9800GTX performs about 30% better in the xbit review than it does in the tech report review.

 

Now that you pointed it out, I look at it and can´t stop think how weird the results are.
To much difference between the 8800GT and the 9800GTX, without a reasonable explanation.
Maybe a driver bug?

 

The only thing I can think that would cause that big of a disparity is that AA wasn't being applied on the 9800 in the xbit review. Without confirmation from the author about what's happening we can only assume the 9800 result is an anomaly. That or he ran a gx2.

 

Xbit has always been controversial with Radeons (at least since the 2k) though, so it's not particularly surprising per se.

Thankfully they got the cards so late that I don't think any purchaser is taking any notice either way.