Curing ATI's headaches: veni, (N)vidi(a), vici?

THESE DAYS, the Nvidia-ATI battle is getting more interesting than the eternal Intel-AMD showdown - the performance and features on the GPU front are surely advancing far faster than on the CPUs, yet the benefits to the majority crowd are, at best, unclear. How many PC users really even need a Shader 3 in their everyday use, frankly? At least, a faster CPU makes the M$ Office XXX run, well, less slow.

But in any case, the success (or lack of it) at the high end of the battle affects the vendor's image across the board, so the fight must continue. It is, in a way, like the supercomputing show-offs by big iron vendors - no money made, but a bombastic kind of (expensive) advertising.

There are interesting parallels between Intel-AMD and Nvidia-ATI competition here. Intel's P4 has faster clock, but does less per clock than AMD Athlon64. Of course, Intel is (still) perceived as the big bully, while AMD is the underdog loved by the 'righteous ones'.

What if looking at the 7800 series vs X1800 series? Nvidia is more like AMD here, with more pipes (for more work per clock) but lower clock, while ATI has (just like P4) less pipes, but higher clock. On the other hand, Nvidia could right now be the (perceived) bully, and ATI the underdog?

Performance Race
For the past few months, Nvidia had the nominal performance advantage with the 7800GTX vs the old X850XT. After the X1800XT was announced, Nvidia was expected to reach with a 7800 Ultra - it came in the shape of 7800GTX512. Nvidia might be keeping the Ultra moniker for yet another G70 spin before the 90 nm version is out.

In the meantime, as CrossFire was launched, Nvidia also updated their SLI with boards that support 2 x PCI-E X16 SLI, but with a clumsy combo of two bridges connected by HyperTransport, so one of the PCI-E X16 ports, the one on the south bridge, is then a hop farther away from the CPU compared to the other one.

Coming back to the GPUs, right now, 7800GTX512 at 550 MHz GPU clock and 1700 MHz memory clock seems to look better in quite a few benchmarks than the (cheaper, though) X1800XT at 625 MHz GPU clock and 1500 MHz memory clock. Of course, we all hear that X1800XT scales very well in clock speed and, after all, GDDR3 memory speeds are upgraded constantly.

So, what can ATI possibly do then at the high end right now, while readying the R580 family?

Well, a X1800XT PE comes to mind first - the excellent clock scaling of the current X1800XT, combined with further tweaking along the way, and better memory, should give us a 700 MHz GPU / 1700 MHz GDDR3 X1800XT PE card at least, if not a 750 / 1750 MHz one (nice numbers, easy to remember). The first, worst-case, solution, should bring in an extra 12% in performance across the board, while the other one would come to around 20% speedup - enough to at least match, if not exceed, the 7800GTX512 in many benchmarks - after all, we'd be talking about over 10,000 3DMark05 default score on that 3.5 GHz Pentium 4 XE that I used last week for HIS IceQ X1800 series tests.

And, if that is not enough? Well, companies like HIS, Sapphire, Leadtek and others have been playing around with various types of advanced cooling for awhile - so what applies to Nvidia, (usually) applies to ATI too. That bit of extra cooling could bring the X1800XT PE beyond 800 MHz / 1800 MHz in "manufacturer approved" configuration. Now, the question is - will ATI come out with the PE on time for X'mas?

The supposedly record-breaking R580 GPU, as well as the RD580 north bridge and its Intel-equivalent chipsets are beyond that date, but they should offer some interesting capabilities beyond the Nvidia SLI32. For instance, the RD580 grouping of both PCI-E X16 ports on one chip, with direct communication between them without an external bottlenecklink, would in theory make this a faster and more elegant solution then SLI32, and produce better scaling from 2 X 8 to 2 X 16 compared to the current Nvidia approach. Not to mention, the abundance of PCI-E lanes and expected very high HT link speed makes the RD580 a very good candidate for high-end Opteron workstation boards, too.

Back to the future: AGP
The recend Diamond X1800 AGP PR fiasco actually shows a clear opportunity for ATI - with so many AGP-based systems around, high-end ones included (who won't change the whole system overnight just to fix in a PCI-E graphics card), it would be quite an opportunity to address this real and existing market.

Even the current X1800XT and X1800XL with an added Rialto PCIE-AGP bridge to fit into an AGP slot, would probably take in all the current AGP-based high-end GPU demand, as there is NO competition from Nvidia there. If you want to play H.264 HD video on an AGP system, this would be the only solution, full stop. So, while fighting the high-end battle, ATI (or any of the specialist ATI card vendors like HIS or Sapphire) could take this market.

Last but not least...
...is the price, of course. As mentioned before, there is also a limit on how much excessive performance does anyone (including gamers) really need. If the X1800XT is at the same price at 7800GTX, or no more than US$ 50 above it (roughly the 256 MB to 512 MB memory cost difference), it wins. If the X1800XT PE when (if) it comes out, is US$ 50 cheaper than the similar-or-slower 7800GTX512, it wins. But, ATI has to deliver pronto this time, both performance and price-wise. That would give it some breather while launching the true counterattack - the R580. µ