ATI Mobility Radeon 9700
Sneaky naming but not too wide of the mark
Think about the technical side of the desktop Radeon 9700, especially the Pro, and your geeky heart starts to flutter at the prospect of the full R300 core in a focussed mobile chip.
256-bit memory bus, 8 pixel pipelines, a clock speed befitting a mobile part and all the rendering goodness we've experienced on the desktop since its launch. When the PR came through about MR9700, I had visions of ATI perfecting the move of the R300 core to 130nm and reducing its power consumption enough to make it viable as a new mobile part.
A second thought about that makes you realise why it would never happen. ATI have a set Mobility Radeon lineage now, since the birth of MR9600. Any new mobile part was 99% going to be a relation of that core technology, not a direct mobile adaptation of R300.
So what's MR9700 if it's not a mobile R300? Nothing more than MR9600 using the Low-K process enhancements they rolled out in the recent XT range of desktop GPUs. It allows them broadly similar power and thermal characteristics as MR9600 with a slight boost in clock speed. At up to 450MHz, MR9700 is more like MR9600XT than anything else, but maybe the marketing inference is nearly correct in some respects.
At core speeds that high, the MR9700's pixel pushing power approaches that of a regular Radeon 9700 (non-Pro). 450MHz core ([H]ardOCP report 446MHz on a sample unit) and 4 pixel pipelines, each with a texturing unit, means 1800Mpixels/sec of pixel fillrate and 1800Mtexel/sec of multitexturing fillrate. 275MHz and 8 pixel pipes on the R300 powered 9700 gives 2200Mpixels/sec and 2200Mtexels/sec. There's 22% more raw power on the discrete 9700 but we can forgive ATI the comparison, on a mobile GPU the comparison is probably close enough.
Memory bandwidth wise, the MR9700 is most likely to be paired with DRAMs running at ~DDR530, close to the DDR540 on a regular Radeon 9700. But with half the memory bus width and only 64MB of memory in most mobile configurations, it falls far short of its desktop counterpart.
So the obvious comparison is with MR9600 rather than any desktop part. With an MR9600 sample in-house, clocked at 350/DDR400, that does very well in most of our benchmark tests (just shy of 10,000 marks in 3DMark2001SE at default driver settings, powered by Athlon 64), expect MR9700 to be that bit quicker with 28% more core speed and 32% more memory bandwidth.
Plenty of oomph in DX8 tests at decent resolution and AA and AF applied, it should do great at DX9 titles too with that core speed boost, since they tend to be shader intensive, stressing core clock.
So bar the clock change over the high-end MR9600 parts, has anything else really changed, bar the Low-K process switch? Not really, which explains the low key launch. Parts were shipping to vendors before most people heard it existed, logical drop in upgrades for the high-end desktop replacement models that most will sell you. ATI are keen to get the LAN party crowd interested in MR9700, the first part they are happy to push at those most demanding of gamers.
Makes you wonder what NVIDIA have to counter it, if anything. Anyway, just a little PR-driven opinion piece to complement David's MR9700 interview with Chris Hook. I'm impressed with my current MR9600 sample in most respects. In the places I have misgivings, maybe MR9700 can reassure me. We should hopefully have a sample soon.
Gotta love the screenshot of the Radeon 9800 Pro control panel in the MR9700 press .pdf too, nice touch guys.
There's plenty more to MR9700 than clock speeds, Low-K and comparisons with MR9600, they just require a physical sample to outline, PR fluff only takes you so far. Keep your eye out for more coverage of mobile parts at HEXUS in the near future, it's definitely something that's inside our broad enthusiast remit now and ATI's MR9600 and MR9700 go a long way to achieving that.
