AMD and Nvidia do the double-green shuffle

TODAY WAS the time for warm and humid Singapore - a countrywide sauna so well liked by the Finnish - to have its AMD Spider platform talk on all its componentry

A couple of hours before, Nvidia had a briefing on ESA, 780i/a and triple SLI, with a few more press attending in this case. Why? Well, there was a free lunch by Nvidia, I recall now.

After checking out the new stuff demos - our hacks have covered the lurid spec details many times already - I had some thoughts come to mind.

Both companies showed off stuff which either won't be shipping yet (Nforce 780i and ESA, Phenom 9700) or is just trickling out now (790FX, Triple SLI, HD3800 GPUs). While the GPU "competitive focus" against each other was clear, culminating in AMD's DX10.1 and VC-1 HD movie claims, and Nvidia's 8800GT performance talk, it was otherwise on the other fronts.

On the chipset side, there is a bit of competition between AMD 790FX and Nvidia 780a, but at the end, your choice there will depend on whether you want SLI or CrossFire graphics - or, frankly, none, since I still don't see that much value in having multiple GPUs just for gaming fps.

The real target in both talks was - Intel. Nvidia's chipset business, having badly delayed both the 780i and 790i for a variety of reasons -including PCI-E v2 bridge add-on and late 45 nm Yorkfield support - is threatened by the X38 and X48 chipsets on the high end. Not to mention the vastly improved integrated graphics in the G40 series chipsets this coming spring.

Furthermore, once Nehalem enters the desktop, most of their profitable high-end North Bridge business will be gone too - fat returns on the US$ 100+ Nforce 680i pricing will then be just a distant past glory for the Green Goblin. Too bad they missed the most promising period this year, before Intel got its new high-end chippery read.

AMD doesn't have such chipset problem - 790FX is ahead of Nforce 780a, and supports Crossfire X with all of 4 GPUs at once, compared to Triple SLI of Nvidia's chipset. But it does have a problem with the CPUs - a huge one.

Since my rant last week on that, things went further downhill when the 2.4 Phenom 9700 was further delayed due to the TLB buffer "errata". The AMD demo machine today with that famed unlocked "AMD engineering sample" nameless CPU was able to get up today via AMD OverDrive to 2.7 GHz on all cores without upping the voltage - but no benchmark were ran on it by any hacks present. Still, not a bad feat for a quick run - on an, I assume carefully hand-picked CPU though. The $300 (Phenom average selling price) question: now many initial shipping Phenoms will be able to achieve that feat?

Its year-old direct competitor in the same price segment and the target of AMD's benchmark slides, Intel Core 2 Quad Q6600, still seems to perform a tad faster clock for clock according to most sites - and, well, all of these seem to overclock to 3 GHz and beyond too. Luckily for AMD, Intel still did not release the Yorkfield-based Core 2 Quad Q9450, a 2.66 GHz FSB1333 part in the same US$ 300+ price bracket, which would totally obliterate the future Phenom 2.6 GHz, too, even without considering its 4+ GHz overclocking potential. Naturally, Intel wants to make the most of its early 45nm ramp-up right now by charging as much as possible, above $1K, per each chip since the yields seem to be so good that all of them can be labelled as high end parts - why not, when the likes of Apple and HP are buying the whole lot right now?

I really hope to see the Phenom - and Barcelona - B3 stepping real soon now, hopefully breaching the critical 3GHz limit. We need AMD healthy and back on track, or else Intel will be able to push up the high-end PC CPU prices beyond $1500 soon: the QX9770 is near there already.

As for the GPUs, my feel is that AMD has a bit of upper hand over Nvidia, despite slight performance loss of HD3870 vs 8800GT. Their new cards do have a more complete feature set - DX10.1 is far less critical to me there than the complete HD media codec support. Also, they create far less noise and, especially, heat. AMD is also very open to 1 slot-thin watercooled HD3870 cards by the partner vendors or cooling experts like Vadim or Aquacomputer - the two CrossFired samples in the demo systems passed all the tests at the 800 MHz GPU / 1200 MHz memory setting.

Keep in mind their CrossFire X, while not having a sophisticated "all to all comms" bridge like the new Nvidia Triple SLI one, allows that quad-GPU operation on all new 3800 series cards - cheap 3850 or less cheap 3870 alike. Of course, teaming two 3870 X2's is a far more practical CrossFire X solution to me than linking four 3870's - I'm sure there will be matching slim waterblocks for the X2's too. I'm particularly pissed at Nvidia keeping the Triple SLI only to 8800GTX and Ultra cards - both overpriced, bulky and technically obsolete, as everyone sane would await their faster yet lower-power G9x successors.

Yes, I also love the AMD Overdrive utility - Intel or Nvidia have nothing like that, full stop. Nvidia Ntune is a total disaster, just using Auto Tune on it often crashes the system without doing any benefit. Intel, on the other hand, overclocks madly anyway - I don't even want to think what kind of results could be achieved if, for instance, I could have such a utility there and overclock one of the two dies within Yorkfield faster (dual thread game front end) and let the other one stay cooler to handle background thread. 5 GHz + 3 GHz water cooled, anyone?

AMD had a better GPU demo too - check out the new John Woo's Stranglehold game, where you are hyper realistically rendered Chow Yun Fatt (a famous Hong Kong actor, btw) fighting his way through triad-infested Hong Kong streets at a rate of 30 gpm (gangsters per minute) killed in pure blood-spurting gore. Oh boy, was it realistic - if you're in favour of this kind of stuff, check out the new title.

In summary, both green brands seem to have plenty of work cut out for them. AMD "Spider" platform play makes sense with a good chipset and even better cards, but weak, slow and delayed CPU component, the "Phenom is no Venom", cuts it out of the high end - until the next stepping at least. They need that 3 GHz part now, and even then the balance may not be restored: 'deep throat' says Intel could be just a stepping away now from official 4 GHz FSB1600 Yorkfields, if they decide to launch those.

Nvidia, on the other hand, has no CPU leverage in this platform battle, and the chipset delay did hurt them. The G92-based 8800GTS parts should bring some reprieve, and I look forward to the Nforce 790i chipset with DDR3 memory and brand new feature set to fight Intel's X48 for the best Intel Yorkfield desktop platform early next year - the Nvidia demo man, a nice Taiwanese chap called Jeffrey Yen, promised "they will have an entry nothing short of stunning in the overclocking department real soon now". Well, I'd like Nvidia to have more friendly faces like him, for a start - it will be of help to them in the tougher times ahead.

Talking about Yorkfield, most of the regional press I chat with already do have QX9770 from Intel - so, it is very much a real part that Intel wants to, and surely will, sell for a very real US$ 1,400 or so, too. Yes, if you leave it on default settings, it is hotter than the QX9650 at the same 3.2 GHz FSB1600 settings. The trick that seems to work on my Asus Maximus Extreme - beta BIOS 701: lower the voltage for the QX9770 CPU and its PLL manually in BIOS, rather than leave it on Auto settings. It works fine even at 1.22 volts, noticeably cooler - more on that tomorrow in our test! µ