Introduction

Just two weeks ago, NVIDIA launched their Geforce 6800 Go mobile solution. Today, ATI is announcing their competing Mobility Radeon X800 and more wallet friendly Mobility Radeon X300. Formerly codenamed M28 and M22, respectively, the introduction of these GPUs mark ATI's push towards top to bottom mobile PCI Express graphics.

We were able to benchmark the Mobility Radeon X800 earlier this month with the launch of the 6800 Go. We weren't sure of the shipping clocks of the then M28, but we can confirm that the part we tested is the part that ATI is announcing today. Running at a 400/400 core and memory clock speeds, the MR X800 bests the Geforce 6800 Go which can be found shipping right now in notebooks at the 300/300 clocks we tested.

NVIDIA has stated that a higher clocked version of the 6800 Go will be paired with GDDR 3 memory at 450/600 clocks, in which case the performance battle could get much more interesting. Of course, the availability of a 450/600 6800 Go notebook is exactly the same as the MR X800 right now. We will try our best to get our hands on some parts for testing as soon as possible, but getting a hold of a shipping notebook will take some time.

We will be counting in weeks, but ATI is saying we will definitely see the MR X800 before Christmas, with the MR X300 following in early January. While the history of mobile graphics has gotten us quite used to waiting months between a GPU launch and it's inclusion in a shipping notebook, the advent of AXIOM, MXM, and custom modular components has really helped to decrease the time it takes to get silicon into a laptop. Weeks down from months for the MR X800 should not go unnoted, but the achievement is quite overshadowed by the fact that NVIDIA pulled off immediate availability of its 6800 Go two weeks before this announcement. We would definitely like to see ATI match this in the future. The blessing and curse for ATI is that they ship a much higher volume of mobile parts than NVIDIA. This may make the logistics of a launch harder, but if anything it only increases the responsibility ATI has to maintain a leadership position in minimizing the time between an announcement and high availability of shipping product. To top that off, NVIDIA has a good jump on the holiday/spring semester season by getting product out in the market place before Thanksgiving.

Just to clarify, this isn't the first time we've seen X800 based tech go mobile. The MR 9800 was based on the desktop R420 chip, but configured with 4 vertex and 8 pixel pipelines (more like a desktop 9800). There are a few differences here: PCI Express, GDDR3, higher clocks, and ... well, let's wrap it all up in a nice package.

ATI Mobile Technology

OK, so all the technology might not be brand new. We've covered the basic R420 architecture, delved into the first mobile AGP R420 solution, and we've even already benchmarked the MR X800 against the 6800 Go. And now it's time to fill in the gaps and tie everything together.

The following table shows the differences between the desktop X800 part, the mobility radeon 9800, and the new improved MR X800, and the lower end MR X300.

Radeon GPU Comparison
	Radeon X800 Pro	Mobility Radeon 9800	Mobility Radeon X800	Mobility Radeon X300
Fab Process	130nm low-k	130nm low-k	130nm low-k	130nm low-k
Vertex Pipelines	6	4	6	4
Pixel Pipelines	12	8	12	4
Core Clock	475	350	400	350
Memory Clock	450	300	400	300
GDDR3 Mem	YES	NO	YES	NO
Max Mem	256MB	256MB	256MB	64MB
Bus Type	AGP/PCIe	AGP	PCIe	PCIe

DLCS

Even though similarities with previous products abound, ATI won't be without new and interesting technologies to include in their mobile PCI Express launch. Included in their POWERPLAY 5.0 power management system is a very interesting new technology called Dynamic Lane Count Switching (DLCS). DLCS is designed to save GPU and system power when less graphics intensive applications are being run. The idea is that if I'm running OpenOffice.org on my laptop and typing up my next article, I don't really need 16 lanes and 75 Watts of raw graphics power being poured through the PCI Express bus. ATI can dynamically switch between x16 and x1 PCI Express operation. Dropping down to one lane cuts off connections on the GPU, lowers the incoming bandwidth it needs to hand, and physically turns off the rest of the x16 power connections to the part. To enter this mode, the GPU must enter its underclocked power saving state in order to handle the drop in power, and the drop in bandwidth at this underclocked state should have no further impact on system performance.

This functionality to control the PCI Express bus in built into the GPU itself, and the feature can be disabled by the user if so desired. ATI has said that they are exploring the ability to switch between x1, x4, x8, and x16 PCI Express operation depending on the power savings and graphics horsepower needed, but the current limitations are on the Intel system bus side. That may be a very interesting feature for power savings on games where graphics needs are low. In any event, this looks to be the most promising power saving technology of ATI's PCI Express generation of hardware, and we can't wait to get our hands on a notebook for testing of real world impact. ATI has indicated that power savings are up to 30% over "typical usage scenarios," whatever that means.

Business users who might also like to do a little gaming on the side could well benefit from this technology. But again, we'll have to get it into our labs and see what happens when we crank up winstone, sysmark, pcmark, worldbench, or the flavor of the week business benchmark in power saving mode on a shipping notebook to really talk about how useful this feature is.

Continued in this generation are ATI's POWER-ON-DEMAND feature (which allows dynamic voltage and clock frequency control based on power needs), and clock gating. Anand covered clock gating in detail in our 6800 Go article two weeks ago, and in case you missed it in our Mobility Radeon 9800 coverage, the actual block layout of an R4xx based GPU with clock gating follows this basic pattern (adjusted for number of vertex and pixel pipelines of course):



Hypermemory

On the low end, ATI is pushing a discrete, 32MB, and 64MB X300 part. At the same time, in order to make up for the lack of framebuffer, ATI is pushing the high bandwidth advantage of PCI Express.

When a game or application requests more memory than the graphics card has available, ATI will allocate enough system memory to fulfill the task (up to a quarter of system RAM). This is basically just giving a name to the idea of virtualizing memory and allowing the GPU to transparently access RAM off the graphics subsystem. Of course, in the past, using system memory as part of the framebuffer was suicide for framerates, but from the demo we saw, 3dmark 03 doesn't do a bad job of running with part of its framebuffer in system RAM. Hopefully PCI Express will push developers to use the CPU, main memory, and entire computer in ways only game consoles have been exploited before. Of course, we won't hold our breath.


The Future Video on ATI Hardware

Last week, ATI once again reminded us just how important it is to have high quality dedicated video hardware embedded in a GPU. They ingrained in us the fact that all around video quality on current computer systems is sub-par compared to much less expensive consumer electronics devices. Yes, ATI does have blocks dedicated to video functions (iDCT, adaptive de-interlace, etc.), but no has a truly high quality hardware video playback solution integrated into their GPU as of yet. Yes, NVIDIA has hardware, but we haven't been able to get our hands on shipping drivers to enable its functionality as of yet, and even then we'll have to run our own quality tests on it.

Anyway, the point is this: ATI brought up the fact that they have a division dedicated to making hardware specifically for taking a DTV signal, decoding it, and displaying it on a television. They indicated that they plan on leveraging the experience they have in developing the Xillion technology to bring consumer electronics quality to ATI GPUs.

The long and short of that conversation is: why not before now; we're tired of waiting. We know you've got silicon in Sony televisions. We know the technology in the Radeon (and Geforce for that matter) parts falls unbelievably short of it. It's time to fix the problem and show us the solution.

The Test and Halflife 2 Performance

We used a test bed configuration similar to our previous M28 review: 3.4GHz P4 Prescott 1GB RAM. Our 6800 Go system this time around was the same but with a speed boost to 3.4GHz as well. We were running the same system clocks as last time as well: 400/400 for the MR X800 and 300/300 for the 6800 Go.

We had to do our testing on site under very rushed conditions, and we were only able to get a couple of Halflife 2 runs in to update our benchmarks from last time. This time around we'll also include Anand's desktop numbers. Unfortunately, he was benching with an Athlon 64 4000+, so the numbers will not be directly comparable. Of course, the only way we could really get directly comparable numbers would be to get PCI Express mounted versions of the mobile GPUs and compare them in a desktop system against our desktop cards. Or if we were to stick to mobile, the modular design of some laptops would allow us to swap out ATI and NVIDIA solutions. But here we will make due. First, we'll take a look at what happens when we run without AA and AF turned on.



As we would expect from our previous look at the two GPUs, the M28 based chip outperforms the NVIDIA solution at both resolutions we tested. Not only that, but the ATI mobile part puts in a decent showing, only losing to the desktop cards. Again, keep in mind that we are looking at two different platfoms, and that Anand hasn't run down all the caveats of Intel vs. AMD performance numbers under HL2. Even so, these numbers are impressive. Next we take a look at just the mobile parts with 4xAA and 8xAF enabled. Not a setting mobile gamers are used to running.



Under Halflife 2 with 4xAA and 8xAF, the MR X800 means the difference between playability and choppy framerates when compared to the 6800 Go. That's a very nice option to have for the gamer on the go, and one that will be hard to turn down for the hard core fan of the notebook DTR.

Again, if you're interested in seeing more extensive benchmarking of this card, check out our previous article on the subject. The long and short of it is that the MR X800 seems the more powerful card at initial shipping clock speeds. We will be waiting to get our hands on faster parts to see how these GPUs scale, both in performace, and in power.

Final Words

Even with the benchmark wins, the bottom line is that NVIDIA is still ahead until ATI can get product out the door. It doesn't matter how fast a part is until it can get into the hands of the users. We commend NVIDIA for synchronizing their launch with availability, and urge ATI to catch up.

It is understandable that carrying a higher percentage of the mobile market can be a burden for ATI, but we see it as an opportunity for them to set the standards rather than follow the lead of others. Market share is an important metric by which to measure success, but with that success comes even more responsibility.

ATI has an excellent part on their hands. The R4xx generation scales down well to the mobile segment, and the 130nm low-k process seems to be taking ATI a very long way. We are excited to see technologies like DLCS, Hypermemory, and PCI Express in general make it's way to mobile. This is a very exciting launch for ATI, but we just wish we could have seen shipping notebooks today.

Again, we love talk of high quality video incorporated top to bottom in a company's GPU lineup. But that's going to have to be something we see to believe.

The MR X800 will be here sometime between a couple weeks from now and Christmas. Until then, NVIDIA is getting a very good head start on the holiday season.