HD 4870 X2 (R700) review thread

One thing Walt, if you plan to get two R700 plan to get a beefier PSU. I'm running mine on a Corsair 620w just fine, but I wouldn't want to try two with it!

 

Tech Report lists two R700's in full load at 752 watts under load, which is the highest I've ever seen. Which means there is finally a use for 1000W units. Though a single R700 is only 448W under full load and a 620W unit is fine.

 

Anandtech managed to get that high of a load with three 8800 GTXs SLIed.

 

TR's numbers are at the wall.

 

Your point is? Everyones numbers are at the wall. Even at 80% efficiency you're going to need more than a 620W PSU for two R700s and you're going to want the system to fall into the 50% load range for best efficiency and not make the PSU's fan go crazy loud.

 

My calculator disagrees

50% load for best efficiency? I'm not sure where you've received this figure, but most quality PSUs deliver the best results about 75% load, some even working optimally all the way up to 100%. PC Power & Cooling comes to mind.

 

I would suggest you do a bit more research. Nearly every PSU on the market is most efficient at around 50% load. You're also very wrong about PC Power & Cooling as all of their 80 Plus certified PSU's are most efficient at 50% load. You can check that at the 80 Plus website and many others across the web, and if you look at the other 80 Plus certified PSU's you'll see they almost all follow the same exact trend.

 

Assuming the measurement was taken at the wall and the power supply was working with 80% efficiency at the time of the measurement, the unit was providing 725x80/100= 580W of usable power to the system. If you want to avoid exceeding 60% of the full potential of your power supply, the unit you require should have at least 580x100/60=996W maximum power.

 

Why avoid exceeding 60% utilization?

This seems like a huge expense for no gain. You'd be far better off buying a quality 700W range unit (that can handle the load) then spending more money on a PSU that you will never utilize.

I don't get the push to only use 50-60% of the PSU. Sure, it is slightly more efficient, but IMO, if you buy quality parts, you are just wasting your money.

 

There are two marks I consider, full load and idle load. I would personally say a 850W unit would be the perfect match for a HD 4870X2 Crossfire system (4 GPUs). It's going to be a bit below the highest of maximum load and it's not going to be well below the lowest of maximum. Power supplies have their lowest efficiencies below 25% and above 75%, if your system can stay within those ranges then you're going to do very well. Going by Tech Report's numbers and assuming 80% efficiency then you're going to stay perfectly within that range (it's very close in the idle end). My remark about 1000W unit wasn't to suggest it was the proper choice, but more candid because I laugh at people who buy so much more PSU than they need. Even this system doesn't fully need such a system, but it's closer than what the majority of people who have 1000W units have.

 

This card is a bit of a letdown imho. There was so much talk about R700 being something more than "crossfire on a stick" yet that's exactly what it is. I really wonder about that disabled sideport though... It's totally unsurprising that for traditional AFR methods (and non-unified memory for textures) you don't really need more bandwidth than what the bridge provides. So why put it there (on a pad-limited chip no less!) in the first place? Somehow I think maybe something else than "crossfire on a stick" might have been planned at some point...
Also, power consumption (especially idle) is sky high (that's of course still true for 4870 single too). There was some hope newer drivers would fix this, but I guess it looks like clock gating is FUBAR and won't get worked around in drivers (clock gating is pretty much always broken so that's not a big surprise but still it's disappointing it's not driver fixable).

 

When do we get to see the demo's?

Anyone?

Well these high res shots look MENTAL !!

http://www.flickr.com/photos/amd_unprocessed/2585849022/sizes/o/

http://www.flickr.com/photos/amd_unprocessed/2585016023/sizes/o/

 

Hehe, those are some incredible screen shots!

 

I didnt realise how photoreal the scorpio looked till I saw the high res image of it.

Its insane, Check out the small lumps and bumps in the claws!

Im still having a hard time believing this is real time.

 

There's one of the long demo movies where they are zooming in and rotating around the scorpion to prove it's being rendered in real time, and they've modelled tiny sensory hairs and everything.

I wonder if they put an actual scorpion into one of those 3D modelling rigs they were showing that turns visual info into 3D models?

 

Indeed,

I just did a bit of digging around and found this I hadnt seen before which I think explains all this photorealism we keep seeing.

http://uk.youtube.com/watch?v=aCgeVDLpV6k&fmt=18

What I think is important here is that although its all very well we can get photo realistic datasets from this, but they produce massive amounts of data that would be very difficult to manipulate so they are effectively pre rendered voxel sets that you can view from any angle and change the surface lighting information of (via the normal map information held at the location of the voxel in 3D space).

Ive just realised having seen this new vid that the city from the ruby demo is also using this voxel render technology, thing is, all the lighting info is prebaked, the city wont have destructability and not really have that much flexability apart from you can lterally just fly around it, Then I believe what they would do is composite polygonal elements into it (like the giant spider robot), and use the z depth info from the voxel scene to correctly relate the resulting renders.

EDIT: Even if the cars in the scene are voxel derived they can still use the normal and voxel info combined to raytrace secondary reflection effects, probably using a real 3d lower polygon version of the original voxel set, to ascertain the required reflection colour information and then combine that with the voxel for a given pixel.

So this in effect is like the Rage tech from ID, Its awesome for static scenes but limited for truly interactive gaming environments. Effectively its like 3D film, and that is why they are calling it Cinema 2.0.

The penny has dropped.

 

IIRC they were talking about compression ratios of 1000-1 or some other bonkers numbers.

Again, I think I remember from one of the presentations they denied it was prebaked, and did all sorts of realtime things with lighting and various other effects to prove it.

Still there's a lot of stuff you can do in games with this kind of level of detail. It's a serious leap as long as it's something that actually gets used, as opposed to one of those demos that look great but the techniques never find any place in game development.

 

Well, what it is, that global illumination soft shadowing you can see under the stairs on the outside of the building is a pre rendered layer of colour information that can either be added or subtracted from the resulting voxel scene. Thats why when he is doing the demo he talks about layers of colour info.

So, what they would then do is have some nice real 3d shadows being generated for the polygonal robot spider as a unique render pass, that is then composited into the frame. ie the robot shadow is 'added' as composite layer on to the (voxel) road below it.

 

It's not so simple. CPU limitation varies from game to game.

 

http://www.tomshardware.com/reviews/4870-x2-amd,1992-9.html

I wonder if it bothers Nvidia that they lost to ATI in performance crown!