HD 4850 review thread.

The place I used to consistently see a big difference between 4X and 6X MSAA is on Half-Life 2 and CS:S - particularly the bridge section, but all those wires as well.

I suspect a lot depends on the dot pitch of the display, so it would be 22" 1680x1050 panels that would probably be the worst offenders at this point in time for PC monitors, with laptop panels needing least AA - and, of course, any HDTV is going to have pixels like boulders compared to a PC monitor. Anyone playing PC games sat ten feet from a 50" 720p plasma? You'd want all the AA you can get there, I'd have thought. Even on a 40" 1080p LCD (likely one of the HDTV sweet spots) the pixels are nearly 4x the area of any PC monitor's.

So you can perhaps you can make an argument for testing 8X more at HDTV resolutions for that reason...

 

I personally find that higher MSAA rates than 4x yield marginally smaller returns in terms of subjective increase in IQ (e.g. going from 0x to 2x and then to 4x deliver far more noticable difference than further doublings), and that the most noticeable aliasing arises from either alpha-tests, shader undersampling, or shadow aliasing.

In fact, I'd say shadow aliasing effects stands out more than anything else these days. The stair-step shadows in Metal Gear Solid 4, for example, detract from what is otherwise, one of the least aliased PS3 titles.

Higher frame rates would actually be welcome. First of all, for PC titles, anything less than a rock solid 60fps IMHO is a travesty. Temporal multisampling to overcome poor refresh rates would be further welcomed, although to be really effective, it would probably require insanely high rates.

I don't like Crysis being used as a standard bearer benchmark in so many reviews. That game is bottlenecked in so many ways, and despite it's beauty, it is not like GTA4 or WoW, or some other triple-A title that is a "must own" Yes, it taxes GPUs, but it also taxes the system in a way that inhibits showing how GPU architectures scale. I mean, am I going to choose to buy a GPU because one runs Crysis at 21fps, and another manages to pull 23 fps!? At those speeds, I wouldn't even buy the game in the first place.

 

It's funny you say that, because I originally wrote "image" and replaced it with "3D scene" because I wanted to imply motion. I agree with you, but I still find >4xAA of marginal use.

 

Thanks a lot. I had a feeling this was the case when OpenGL guy first pointed it out, but wasn't sure.

This is a great feature. We can avoid the dither patterns of alpha-to-coverage, get better AA, and still avoid the speed and/or sorting overhead of other solutions.

 

I get that for competitive reasons there's going to be a lot of toing and froing on >4x modes. And that's fine --progress must be served. For me tho, the real takeaway here is we as a industry/community have driven "no need to ever think about NOT running at least 4xaa" to the $200 price point across the entire industry, and good on all of us for hitting that milestone.

 

X-bit Labs have their architectural review up with their usual synthetic tests, if anyone is interested:
http://www.xbitlabs.com/articles/video/display/ati-radeon-hd4850.html

 

IMO we've been at that point for years now on every DX9/DX10 GPU except NV3x.

It doesn't matter whether you're choosing between 640x480 w/4xAA and 800x600 w/o AA or between 1920x1200 w/4xAA and 2560x1600 w/o AA. The AA option nearly always performed and looked better. The same goes for AF, except even more so.

 

Well, I'm installing a pair of HD 4850's tonight, so ve shall see vat ve shall see... Heh... Been awhile since I've had this much fun--uh, with 3d, I mean.

 

So I bit the bullet and tried the updated BIOS for my 4850. After setting my card back to defaults in the CCC, it started to idle properly.



Instead of idling at 500MHz Core and 750MHz Memory, it's now idling at 160MHz Core and 500MHz Memory.

I'm going to run some benchmarks to see if there's any performance loss from this BIOS.
Okay, I've ran a few benchmarks and it performs identically and feels identical in gameplay too.

 

Nice, and 43c at idle is much more comforting. Now if one of the reviewers will retest with the new BIOS.

Regards,
SB

 

No, I have a 3rd party cooler so my idle is already super low. I need someone with the stock cooler to test this to see what the drop really is.

 

I tried that bios on my stock 4850, there is a small difference. Stock bios idles at 80 to 85c. That bios is more like 71 to 75c.

Chiphell released another revised bios that dialed the fan up a little more. Idle on that bios is 60c and load is 66c.

 

You can make your own bios with the latest Radeon Bios Editor, which is reccomended. Looking at how exactly the fan is setup by default will also make it clear why idle clocks are where they are.

 

Could anyone tell me if the HD4850 card is longuer or smaller than a X1900 XT ?

 

I've been disappointed this gen that lengths seemed generally left out of reviews. I always look for that.

 

4870 is only 9.5"? I'm suddenly somewhat interested. . . . I was assuming another 10.5" monster.

 

The length monster is reserved for the HD4870X2.

 

No, its olny about 11mm longer than an HD 3850/4850.

its ~5mm longer.

 

Hmm, I think I'll let you and Sireric fight that out. Somewhere I think I can dig up a post of his from the R3/4 era that says, paraphrasing, "go for res first, then AA". Possibly at R3D rather than here.