ATI Radeon X850 XT Platinum Edition Review

[Dave Baumann]

http://www.beyond3d.com/siteimages/b3dsmall.gif (http://www.beyond3d.com/reviews/ati/r480/)After a little longer than 6 months preceding the Radeon X800 launch ATI are preparing to announce the series “mid-life kicker” with not one, but two new chips: R480 (http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=75) and R430 (http://www.beyond3d.com/misc/chipcomp/?view=chipdetails&id=76). With R480 designed to address the very high end enthusiast space, and R430 aimed more towards the high-end mainstream segment, ATI are hoping to fill the entire price range from $100 - $550 with the “X” series.

In this article we are taking a closer look at ATI’s top enthusiast product, the R480 based Radeon X850 XT Platinum Edition. With fairly minor clock speed increases from the X800 XT Platinum Edition, dual slot cooling and an estimate $550 price tag we take a look at the performances offered by this latest high-end product. Read the full review here (http://www.beyond3d.com/reviews/ati/r480/).

[DerekBaker]

A typo here:

we had heard that initial X800 boards suffered yield some issues at the board level

[Tim]

Typo "Video Drivers Forceware 66.93 WHQL"

[Dave Baumann]

Perhaps thats why the performance gains were so limited! ;)

[wireframe]

Reading while picking out typographical errors.
Page 5, bottom[/b]]One element that does stick out, though, is that after one texture player, where the performance difference between the new and old XT Platinum Editions is roughly equivalent to their theoretical fill-rates differences...
My bets are that this should read "layer".

Page 6, middle [/b]]As such, the performance all the X800 class boards, with their 6 vertex shading units each...
Looks like there is missing an "of" between "performance" and "all" to make it read "the performance of all the X800 ..." and we all know the rest. It's just a theory...

Page 8, middle[/b]]In fact, is we look at the low resolution performance we can sere that there are large variances between the boards...
Although we all like seeing the inclusion of random words from Romantic languages in otherwise numbingly technical analyses, I strongly suspect this was in fact intended to be "see".

Bah, I need to stop looking for errors now and actually read this. :D

Oh, and allow me to curse a bit as I was secretly hoping the X850 series would be more than the speculated speed bump. I was desperately wanting to read something like "much improved efficiency," but, alas, that was not to be. Dag nab it!

[CJ]

It seems that in the VPU charts, the core and memspeeds of the X850XT and X850Pro have been switched.

It says the X850XT and X850Pro have core speeds of 540Mhz and memspeeds of 520Mhz. Shouldn't this be the other way around? :?

[Tim]

It says the X850XT and X850Pro have core speeds of 540Mhz and memspeeds of 520Mhz. Shouldn't this be the other way around? :?

Yes the conclusion says the vanilla XT and the Pro runs 520MHz core, but the spec chart says 540MHz core.

The 850 might not be a big step forward but it is nice that Ati has mad overdrive usefull now.

[Anonymous]

Nice review, Dave, considering how thoroughly unappalling this new product is - to me, it's hardly more than a dual slot cooler upgrade for an extra $50.

What I wonder though, and maybe I missed it in your article, but did ATi say anything about an expected availability date for the new cards?
(not that I'd put to much hope on such an information, given this past year's release history)

Btw, I like the dual-DVI outs on the reference board, but will they transpire to the retail boards? I've seen a HIS X850XT PE using VGA+DVI outs on HKEPC.com this morning, and guess that's what we're going to see from other ATI partners, too.

cu

incurable

[mczak]

talk about a really boring refresh...

However, the r430 cards indeed look interesting. The X800XL (as well as the X850Pro but who is interested in that at this price?) is potentially a decent 6800GT competitor, even at a probably lower price, where the X800Pro couldn't quite match it. And, despite not using low-k, it looks like the r430-based cards have reasonable power consumption (no additional molex connector, so below 75W).

And I think it's a bit silly to have a different asic just for the "bragging rights" market niche (all r480-based cards), when intel does that with the EE edition they can at least use the same asic as for the server cpus.

I'm wondering though, will those cards actually compete against 6800Ultra/GT/normal/LE, what happened to the NV4x refresh (specifically NV45 refresh, NV48)?

mczak

[DegustatoR]

I'm wondering though, will those cards actually compete against 6800Ultra/GT/normal/LE, what happened to the NV4x refresh (specifically NV45 refresh, NV48)?
NV45 is NV40+BR2 (GF6800GT/U PCIE). NV48 and NV44 are to be launched 1Q05 afaik. NV48 being the new hi-end part and NV44 -- the new low-end part. So in middle it most probably will stay the same for NV till NV5x arrives...

[Nathan]

Dave, the heat sink is extruded/stamped, not machined.

Actually, to be fair, there is a small amount of machining involved.

[Anonymous]

Nice details DB, I would still like to see some benching with "Super High Resolutions" no sites are doing this....

[Dave Baumann]

Typo's and charts updated, thanks guys.

The 850 might not be a big step forward but it is nice that Ati has mad overdrive usefull now.

Somewhat behind the curve from CoolBits, but I do like the fact that it (unlike other apps) can work with the temperature measuring / clock throttling and VPU recover is good. In our tests we never had a situation where throttling came into play (probably not hot enough) but VPU recover did catch a few hangs and reset the device. We’ve got to see what other products this is enabled on though.

There is also an option in ACE to turn on the scene rendering as it does the detection to see what is happening.

What I wonder though, and maybe I missed it in your article, but did ATi say anything about an expected availability date for the new cards?
(not that I'd put to much hope on such an information, given this past year's release history)

Of the things I did ask, that was not one! Skimming a few other reviews I see that a couple are stating “e-tail” online availability today. I’ll see if ATI can confirm.

And I think it's a bit silly to have a different asic just for the "bragging rights" market niche (all r480-based cards), when intel does that with the EE edition they can at least use the same asic as for the server cpus.

I’ve actually put a mild update to the conclusion in the section about yield. The two ASIC’s approach is again about availability. One of the problems ATI faced with the previous generation is that they had no next gen product in the $250-$400 range and they ended up crippling themselves by promising a lot of X800 SE’s to OEM’s that filled that price bracket – the problem for ATI was that this both ate a lot of die allocation (cheaper boards, so more volume) but they weren’t actually yielding a lot of chips with two quad failures so they basically gave away 12p or 16p chips as 8p chips, impacting the availability of 12p and 16p boards. With the R430 ASIC they now have a part that can straddle this price bracket and they won’t have to end up selling more crippled R420/R424/R480’s making the availability of the faster high end R480’s greater.

Dave, the heat sink is extruded/stamped, not machined.

Ta. The main point was to point out that this appears to be a single bit of copper, rather than a base with the fans soldered on. This should have better thermal conducting properties.

Nice details DB, I would still like to see some benching with "Super High Resolutions" no sites are doing this....

This was mentioning in the “Which benchmark to Drop” thread. Sadly my test monitor doesn’t have capabilities to exceed 1600x1200.

[Dave Baumann]

NV45 is NV40+BR2 (GF6800GT/U PCIE). NV48 and NV44 are to be launched 1Q05 afaik. NV48 being the new hi-end part and NV44 -- the new low-end part. So in middle it most probably will stay the same for NV till NV5x arrives...

If thats the case NV44 sounds to be a bit late, since their initial projections stated they would have the top to bottom by the end of the calender year. Sure, 6200 is here with disabled NV43's, but I can't believe they projected that when they initially made those statements.

Curious to see what NV48 is as well - just a boosted NV40 on PCIe, or the rumoured 24 pipe part. If this is Q1 then that's moving into a 9 month refresh, inidcating that NV50 will still come out about year end '05.

[Xenus]

Wouldn't fully fleshing out the NV40's second ALU so it could do 2 shader Op's per pipeline per clock be a better use than adding 8 more pixel pipelines?

[aaronspink]

Ta. The main point was to point out that this appears to be a single bit of copper, rather than a base with the fans soldered on. This should have better thermal conducting properties.

The term you are looking for Dave is Skivved Fin. This is a process of effectively shaving fins out of a single block of material.

[mczak]

Nice details DB, I would still like to see some benching with "Super High Resolutions" no sites are doing this....
The firingsquad review http://www.firingsquad.com/hardware/ati_radeon_x850_xt_prev/ has some 2048x1536 numbers (not for all games though - maybe the others didn't support it?), even with some interesting numbers imho (GF 6800 cards seem to lose quite a bit more performance than X8X0 cards for whatever reason).

[Anonymous]

Great review as always. I can't wait for a review of the x800xl that card looks like a winner to me. I still wish ATI would drop the idea of selling crippled card for 399.

[Dave Baumann]

Wouldn't fully fleshing out the NV40's second ALU so it could do 2 shader Op's per pipeline per clock be a better use than adding 8 more pixel pipelines?

It can do (more than) two shader ops per clock already. However, you’re saying increase the instructions-per-pipe-per-clock by increasing the ALU capabilities, the question is why would they? Both (adding more pipes or adding more ALU capabilities per pipe) achieve the same thing (increasing the overall IPC), but increasing the ALU capabilities has some immediate drawbacks. For instance, to increase the ALU capabilities per pipe you would need a new software, and potentially, hardware compiler / instruction scheduler, meaning more work, whereas the current compiler and instruction scheduler would just work and the current quad dispatcher can just distribute to more quads fairly easily. The design block for the pipelines as they are at the moment is also already done, so adding more blocks is a much easier task than altering the blocks.

The term you are looking for Dave is Skivved Fin. This is a process of effectively shaving fins out of a single block of material.

Ahhh, thanks – that’s the idea I had, bit not the term.

The firingsquad review http://www.firingsquad.com/hardware/ati_radeon_x850_xt_prev/ has some 2048x1536 numbers (not for all games though - maybe the others didn't support it?), even with some interesting numbers imho (GF 6800 cards seem to lose quite a bit more performance than X8X0 cards for whatever reason).

Might be related to on-chip ZCULL / HeirZ capabilities. As they take up silicon they have to have a maximum size of pixels they can cater for – NVIDIA may still be optimising theirs for 1600x1200 (in fact, that wouldn’t surprise me as they would want to save transistors where they can with NV40’s size) whilst ATI’s can reach still fully apply to larger target resolutions.

[Xenus]

So your saying while it may possibly be cheaper to do from a transistor route. It would require more work in the drivers than it was worth in saved transistors and die space.

[Dave Baumann]

So your saying while it may possibly be cheaper to do from a transistor route. It would require more work in the drivers than it was worth in saved transistors and die space.

1 - Its no guarantee that its cheaper at the silicon level (entirely depends on what you target / the implementation)
2 - Would require a fairly major rework of the pixel shader architecture that would take time to design, build, replicate and layout; why bother whne you already have a decent design that you can Ctrl-C, Ctrl-V (and relayout the chip) to achieve a similar net result.
3 - Would require fairly major software update.

All-in-all, not worth the effort given that increasing the number of fragment quads will achieve the same thing.

[mczak]

The firingsquad review http://www.firingsquad.com/hardware/ati_radeon_x850_xt_prev/ has some 2048x1536 numbers (not for all games though - maybe the others didn't support it?), even with some interesting numbers imho (GF 6800 cards seem to lose quite a bit more performance than X8X0 cards for whatever reason).

Might be related to on-chip ZCULL / HeirZ capabilities. As they take up silicon they have to have a maximum size of pixels they can cater for – NVIDIA may still be optimising theirs for 1600x1200 (in fact, that wouldn’t surprise me as they would want to save transistors where they can with NV40’s size) whilst ATI’s can reach still fully apply to larger target resolutions.
That's what I thought too. I found it suprising though that the X800Pro didn't seem to lose more performance (comparatively to X800XT), since iirc the biggest change in "HD" of hyperz III was that each pixel quad now has its own hierarchical z tile cache (so hier-z didn't have to be disabled on the "crippled" parts) - if 3 quads are enough to cover 2048x1536, then the 4 quads of the X800XT should be good enough for something like 2368x1776 which is one heck of a resolution...
Though I have a pretty good idea of how hyperz/hierarchical-z works on the radeons, I'm lost when it comes to the same features on Geforces. There doesn't seem to be too much information available :(. It sounds reasonable that it would need some sort of (size-limited) on-die cache though.

[{Sniping}Waste]

Its a little OT but looks like you can pre order a X850 XT PE at ATI now. That was fast.

[Dave Glue]

Its a little OT but looks like you can pre order a X850 XT PE at ATI now. That was fast.
Considering they only need to stock about 14 for the potential purchasers, it's not that impressive. :)

[Anonymous]

If I was going to purchase a 6800gt then would it be better to wait for the new versions of the ATI cards when they release the AGP versions?

[digitalwanderer]

If I was going to purchase a 6800gt then would it be better to wait for the new versions of the ATI cards when they release the AGP versions?
You could be in for a looong wait there, I've been hearing that ATi won't be making any AGP ones until they fill their OEM orders for PCIe ones and they gots themselves a LOT of OEM orders for them PCIe ones.

You may want to hold off and see if any of the current 16-pipe X800 cards drop in price when the new ones come out, or the 6800 GT drops in price a bit.

That's my prediction, and I'm thinking that the X800/6800 are gonna be the last of the high-end AGP cards. I could easily be wrong on that and some new ones come out, but they'll be damned hard to find I'm betting and I think they'll actually cost a bit more than their PCIe counterparts since they'll be rare.

(Either that or I'm trying really hard to convince meself that buying this AGP X800 pro VIVO when it came out wasn't me panicking so much as thinking ahead... :oops: )

[Mendel]

there's a huge install base of agp slots...

[digitalwanderer]

there's a huge install base of agp slots...
True, but are there a lot of 939 AGP boards out there? (I really don't know, I only follow mobos/CPUs when I'm thinking of upgrading. :oops: )

The industry is definately pushing at retiring AGP is what I've been sensing. :?