AMD's Radeon HD 2900 XT Reviewed

J. Dzhugashvili writes "The folks at The Tech Report have whipped up a detailed expose of the new AMD Radeon HD 2900 XT graphics card's architecture and features, with plenty of benchmarks. While the card dazzles with 320 stream processors, a 512-bit memory bus, and oodles of memory bandwidth, its performance and power consumption seem disappointing in the face of Nvidia's six-month-old GeForce 8800 graphics cards."

Comments

Let's hope

[xx01dk]

this drives down prices. I still want an 8800GTX. :)

Re:Let's hope

[Psiven]

You can do pretty well estimating your performance if you have a general understanding of how your components work together.

First measure FPS in your favorite app at the lowest resolution. That's the measure of your CPU bottleneck. No matter how nice of a GPU you buy, you'll never get higher FPS than that.

Memory is one of those things you can never have enough. Just don't worry about the bandwidth too much. Your only going to squeeze out just a few frames per second with top of the line RAM. Just watch to see if your comp is hitting the hard drive much and consider more if it is.

Most new games are still GPU limited and this is where you want to focus your attention. Look for benchmarks at resolutions you play at. This is a good baseline of what to expect. Anything over 60fps avg I tend to be happy with, but you may want consider the minimum too. Right now the only benchmarks I've really been interested in are of Rainbow Six: Vegas. It uses the Unreal 3 Engine, and a lot of games are coming out that are going to be using it too. Other benchmarks might be important to you as well, but they tend to rank in the hundreds and so you know performance won't be an issue.

Re:Let's hope

[19thNervousBreakdown]

This comes up every single time, so forgive me if I'm not as polite about it as I could be.

The human eye sees ~25-30fps, true, but it does not sample the same way your monitor outputs it. The human eye refreshes at that rate, which means anything that's seen for less than that amount of time leaves a partial imprint. Thus, the motion blur you see when something, even in real life, goes by really fast. Since the monitor is outputting static frames, you don't get that partial imprint, and it looks choppy. Television, on the other hand, does pick up the motion blur, because of the way the cameras work. There are a number of studies showing that we benefit from higher FPS, up to and over 100 sometimes.

Also, there's really no such thing as 32-bit color. I suppose you could put a different number of bits for RGB, and many schemes do, but the 32-bit you're thinking about is RGBA. 24-bit is the exact same thing, without the alpha channel, and we also benefit from far more colors than current hardware outputs, because the current 16.7m colors that are output don't account for much luminosity, and for plenty of other reasons I don't care to look up right now.

When you try to stick your hand through the monitor and pick up a Coke, or catch a running kitten, that's when you can say we've got enough. Until then, please try to at least understand the subject you're discussing, and not try to come off as authoritative when you don't even know what 32-bit color means.

Strangely it's Nivida with sucky drivers right now

[Anonymous Coward]

At least on Windows. I got a free el-cheepo x1300 which I ended up replacing my GF6600 with. Sure the latter scored better in 3D Mark whatever, but at the cost of jerky frame rates in non-mainstay games. Such as Outrun 2006 (Which is a bit odd since I heard Sega use the GF6600 in their arcade machine).

Anyway, while these x2900 do not seem to be great performers I suspect their Vista drivers are better. As a Vista user the GF8800 is right now out of the question, less the driver situation have changed recently.

Whomever gets good DX10 drivers out first, got a sale from me.

What has AMD done with ATI

[MountainMan101]

AMD/ATI losing out to nVidia in the extreme power cards.
AMD/ATI losing out to Intel with the onboard graphics.

nVidia has a better closed source linux driver than ATI.

At the moment the only appeal of ATI is there mediocre graphics cards have open source 2D+3D drivers on Linux with R200(helped by ATI) or R300(no help from ATI/AMD) drivers.

At the moment AMD's best strategy is to build some fantastic onboard graphics chips for their AMD processors and try and beat nVidia by basically making and AMD chip + on board graphics as brilliant combination (ie no need to add an aftermarket card).

Re:What has AMD done with ATI

[InsertCleverUsername]

There are problems at AMD/ATI in addition to falling behind the competition. I have a recurring problem ticket I re-opened recently at ATI Support where I got a little bitchy and suggested I'd be going back to NVIDIA if they couldn't get their act together. (I must admit my ticket was mostly a complaint about sloppy work, since I already hacked my system registry and fixed their issue.) Judging from their response to the ticket, I'd say there might be an attitude problem developing there as well.

We respect your decision to follow your prerogatives regarding future product selection preferences. Thank you.

AMD Customer Care

Gallows humor from people who don't see a long future for their jobs perhaps?

Bah

[drsquare]

Graphics cards are all too expensive anyway. You shouldn't have to pay more than the actual processor just to draw pictures on the screen.

Re:Bah

[maxwell demon]

You don't even need a processor to draw pictures on the screen. A simple permanent marker suffices.

Re:Bah

[Anonymous Coward]

Oh you can. You only need to make a processor-to-VGA converter cable. This should be trivial for someone with your intelligence level.

Re:Bah

[dsginter]

You don't even need a processor to draw pictures on the screen.

A real hacker doesn't even need a screen - they just stick their tongue on the HD15* cable and imagine what the screen looks like from the electrical pulses.

(*) Get off of my LAWN!

This is why we need open source 3D drivers...

[kcbrown]

While the card dazzles with 320 stream processors, a 512-bit memory bus, and oodles of memory bandwidth, its performance and power consumption seem disappointing in the face of Nvidia's six-month-old GeForce 8800 graphics cards.

The hardware probably screams. But ATI has a reputation for really shitty drivers. Without solid, fast, high-quality drivers, fast hardware doesn't matter as much.

NVidia has typically produced fast drivers. They're not open-source, but they're at least good.

If ATI can't get its shit together and write some decent drivers, the only reasonable option for them would be to open-source their 3D drivers so that the community can fix them properly. And I expect the community would do just that, because a lot of developers are also avid PC gamers, so they have a personal stake in it.

It'll be interesting to see where this heads, given the statements made by ATI about open-sourcing their drivers, but I'm not going to hold my breath over it. For now, it's NVidia for my gaming rigs. That'll change as soon as ATI actually open-sources their full 3D drivers.

Re:This is why we need open source 3D drivers...

[dave420]

You *do* realise those are the bits of the drivers that interface with the hardware? The same bits that cost millions of dollars to produce, due to the sheer amount of raw performance needed to be squeezed out of them. I doubt the open-source community, regardless how talented (and I know there's some insane talent out there) could replicate those in a timely fashion. Remember - they're playing catch-up with AMD. AMD will keep bringing out new cards, and these guys will have to keep re-engineering these bits of missing code. It's not going to result in open-source drivers comparable to their closed-source counterparts.

A number more reviews

[bad_fx]

As usual Anandtech is extremely thorough: http://www.anandtech.com/video/showdoc.aspx?i=2988 &p=26

[H]ardocp's take: http://enthusiast.hardocp.com/article.html?art=MTM 0MSwxLCxoZW50aHVzaWFzdA==

techPowerUp (Warning, streaming video at the start >.>): http://www.techpowerup.com/reviews/ATI/HD_2900_XT/

The Inquirers expected vapid coverage: http://www.theinquirer.net/default.aspx?article=39 580

I think I'll wait for more ATI drivers and some DX10 games before calling this one... Looks a little underwhelming at the moment though. I'm not regreting my 8800GTX purchase yet. ;)

idle & load power ratings are scary

[Adult film producer]

Idle
----
Radeon 2900XT - 183
GeForce 8800 Ultra - 192
GeForce 8800 GTX SLI - 296
Radeon 2900XT Crossfire - 317


Full Load
---------
Radeon 2900XT - 312
GeForce 8800 Ultra - 315
GeForce 8800 GTX SLI - 443
Radeon 2900XT Crossfire - 490


This could get very expensive for people that leave their computers running 24/7.

Re:idle & load power ratings are scary

[bad_fx]

Just a note, that is for the entire system, rather than just the graphics card. Still high compared to older generations. Just thought I'd point it out, since it may not be clear.

Re:idle & load power ratings are scary

[odoketa]

Given most people's expected load during non-gaming periods, it makes a lot more sense to have a second computer for your 24/7 machine.

I use a mac G4/dual 500 (i.e. an OLD old machine) as my 24/7 box - cost about $200 bucks, and does just fine quietly humming away in the corner drawing 75 watts.

If your idle numbers are right, you'd better have a good friend at the power company if you plan on leaving that machine running 24/7.

Re:idle & load power ratings are scary

[Abeydoun]

Here's a quote from TFA that I also found quite unnerving... "Also, we found that our 700W power supply wasn't up to the task of powering a Radeon HD 2900 XT CrossFire rig. In order to achieve stability, we had to switch to a new Thermaltake 1kW PSU with a pair of eight-pin connectors that AMD supplied."

Now don't get me wrong, I love to see these types of improvements in real time graphics rendering, but you know there's something wrong with the industry if they can ask PC Enthusiasts with a straight face to use power supplies powerful enough for Air Conditioning Units (albeit small ones) in their computers. That being said, I still commend the improvements made and I look forward to the lower end, passively cooled, versions becoming available for my next HTPC.

Re:idle & load power ratings are scary

[tknd]

Power supply manufacturers typically pick a number close to the maximum possible consumption the unit can provide utilizing the maximum across all the different voltages it provides. So when you look at the sticker on the PSU, it will show you the maximum amps per each voltage. You take all of these numbers, multiply the voltage by the amps to get the watts (watts = volts * amps). Then add all of those numbers together to get the total maximum power the PSU can provide. That number should be fairly close to what the manufacturer advertises their PSU as.

The problem with the components in the computer is that they utilize different voltages. So what eventually happens is components require more current on a certain voltage than the PSU can offer at that voltage, however, the other voltages aren't maxed out. That's why people typically have problems with PSUs because they didn't read the sticker and compare the maximum current per each voltage compared to their actual needs. That's also why if you're smart, you can actually get by on a PSU rated at a lower total maximum wattage than a higher rated one (that may just load up on the lower voltages but leave the +12V line crappy to inflate the total maximum power number).

The final bit of information is that many reviewers are reviewing systems by total system power draw at the plug. The biggest factor that is altering this number is the PSU's efficiency rating. There are some really cheap PSUs that advertise high numbers but come with really bad efficiencies (in the 60% to 50% efficiency range). The efficiency of the PSU typically isn't given by many manufacturers. So what all this means is while the total system power consumption is 430watts at the plug, the components may only be eating 300 watts and the PSU is only 70% efficient. If you were to buy a PSU that was 80% efficient while providing 300watts, then your total system power at the plug would be about 376watts. But since many manufacturers don't provide the efficiency rating, you have to rely on data from the internet and more knowledgeable reviewers.

Re:Ob

[wild_berry]

I'd answer "not yet" -- I'm sure that there's a memory management unit on the chip, so don't be surprised if someone does a port...

Re:Worth the wait?

[Psiven]

The 2900XT offers great value though. It sucks power like 12 pound new born, sure, but in modern game engines it clocks just a hair under the 8800GTX. The difference is literally just a few frames. Not bad for a part thats up to $200 cheaper.

At least wait for a june refresh if you're going to buy nvidia.

AMD's big future problem

[Morgaine]

At the present time, the problems that AMD inherited when it bought ATI don't really matter greatly (except as a perception), because only enthusiasts buy graphics cards that cost as much as a basic PC. It's not the volume market.

However, unless AMD sorts all this out over the next couple of years, they are in for a huge amount of very costly trouble, and it may be terminal to their future in the desktop market. The problems ahead lie in the area of CPU-GPU integration.

We are told that AMD purchased ATI because they needed graphics expertise for a projected future in which scalar and vector processing is merged in an extremely parallel multi-core processor architecture. It's easy to see the reasoning here, as tight integration would decrease communication latencies and power consumption simultaneously. The benefits of tight integration are likely to be collosal, and AMD knows this from their success with hypertransport.

Unfortunately, such tight integration also means that ATI's remarkable incompetence at producing even half-decent drivers will bring AMD down badly, unless something is done about it. And short of firing the whole ex-ATI driver team, it's hard to see how to resolve this issue. You can't resolve it by trying to educate bad software engineers, that's for sure.

AMD have quite a problem on their hands.

Re:AMD's big future problem

[Gordo_1]

You had me for a moment there, but then you went and used the old "ATI makes bad drivers" shtick again. I understand that their Linux support has been more or less non-existent, but believe it or not, ATI drivers have been quite solid on the Windows side of the house for a while now.

We all know ATI had really poor driver development in the 90s. However, for at least the past five years or so -- since the introduction of the Radeon 9x00 DirectX9 (R300) generation hardware, their drivers have been at least as good as nVidia's, and in many ways better. They continue to get a bad rap here mostly because of their lack of Linux support and the difficult to defeat consensus that was built up over many years of bad drivers.

Now, onto R600. They really blew this generation. *But* it's not the driver team's fault per se. ATI simply took some gambles in the design phase and those gambles didn't pan out. It probably doesn't help that they had the XBox 360 and the AMD acquisition as distractions during development. On the plus side, it turns out that the 2900XT has enormous headroom -- it's going to overclock like crazy... but guess what? Thermally and power-wise, it's a disaster. Though the silicon may well clock 50-100% higher and blow away nVidia's 8800GTX, it turns out that it eats 600W PSUs for breakfast -- that's the real reason AMD couldn't release a high-end part: Few except for hardcore overclockers have 700W power supplies ready to feed this thing. That, and no one in their right mind wants a computer that uses 400+ watts idling on Microsoft Word. So AMD had to settle for the mid-market, with mediocre performance that's within an acceptable (albeit still very high) power envelope.

At far as driver quality goes, sure they've theoretically had a long time to get things right this time around, but there are bound to be issues with any newly designed generation of video hardware as complex as this. nVidia had many problems with the 8800 series when it first came out 6 months ago, but consensus is that it's settled down nicely with the recent Forceware 1.58 release. Anyway, I just wanted to point out that I don't believe it's the AMD driver team that really deserves to take the hit this time around. If AMD can deliver on the long-term promise of integrated CPU+GPUs, I think this R600 fiasco will go down as a temporary deviation in the grand scheme of things.

Not Direct Competitor to 8800gtx

[gyranthir]

AMD has not released and probably will not release for some time a direct competitor to the 8800gtx or the 8800ultra.
The 2900XT is a competitor to the 2 8800GTS models.
They are avoiding the top end market because more often then not the risk of that market does not meet the reward.
They are playing little ball to compare to base ball, trying to manufacture base hits and runs not home runs.
Offering 3 Cards starting at less than $100 and going to $400ish is a good strategy for the main stream market.
The HDMI dongle innovation (carries video and audio on the video card because all of the new cards have an audio processor on them) is a boon for them as well, helping carry the image of media center capable video cards, for a newer computer user age.
These will help push down prices on all of the cards within that price range. And possibly help push innovation in the marketplace.