Big earthquake hits old Taipei

IT IS September, and, aside from scary earthquakes - I just got jolted from sleep by a 6.6 Richter one - Taipei is still hot and humid despite the recent rains - which somehow subsided the moment I reached the place.

The weather still makes one hot under the collar, especially if that one is trying to get the hold of long-awaited next-generation Nvidia PC chipsets.

I had a round of closed-door chats with several of the local biggies, and yeah, the feeling shared across the board is that of upsetness and slight irritation with now substantially delayed C72 and C73 chipsets - entries which, according to at least one of the vendors, should have been at least shown at the past Computex, three months ago!

C72 is basically a CMOS shrunk, drop-in (supposedly pin-compatible) replacement for the current Nforce 680i. It brings PCI-E 2.0 support to its plethora of I/O lanes (so you could do a more efficient triple SLI now, as even that third x8 slot is at double data rate), on top of slightly improved memory controller and a bit more FSB headroom - 2 GHz quad-core FSB on Yorkfield C72 system should now be a no brainer, as long as the North Bridge is well cooled. According to the vendors, the "replacement" is still almost as hot as its "dim sum steamer" predecessor, the first chipset North Bridge to be regularly hotter than the quad-core CPU in all temperature monitors.

The "drop in replacement" point was moot with at least two vendors - the comment is that, while it is theoretically nice to try to refresh a board by a chipset drop in change, it would be impossible to keep the rest of the board as it is. The things have simply changed too much over the past one year since the Nforce 680i intro - power subsystems for CPU and memory alone need enough update to require a substantial board redesign. But yes, it would have still been doable if the chipset appeared in June during the most Computex.

The vendors were in general more excited about the C73, expected to be the first to combine ultrafast DDR3 memory controller and up to quad SLI. C73, expected around 2 months after C72, is supposed to have more new stuff and, of course, new pinout, not limited by the drop-in requirements for the C72. If Nvidia delivers the performance like it did with the 680i, C73 might beat the X38 for Intel desktop platform performance title.

Talking about the X38, I saw quite a few new boards, a substantial refresh of the prototypes seen at the Computex. The boards work now, and can even be benchmarked. Before I come out with any results - expect that in another story soon - let's just say that the memory scores in both DDR2 and DDR3 board flavours of X38 are slightly higher than the equivalent P35 designs. Only slightly - as most vendors like Asus, MSI and Gigabyte anyway burned the midnight oil fine-tuning their P35 entries to the hilt over the past six months or so. Also, watch out for a couple of interesting X38-based workstation offerings: Asus has a nice offering with both multiple PCI-E and PCI-X slots, and top-notch heat pipe cooling & overclocking features.

More than one X38 board will also come with factory-built North Bridge waterblocks - two of Asus Maximus series offerings, for instance. Gigabyte is also working on a brand new X38 cooling system.

As for the AMD fans, there is at least one RD790-based QuadFX Phenom/Barcelona boards around that I saw - not fully productised yet, so no pics, but it seems this time the boards are really ready way ahead of the CPUs. Unlike Intel's dual socket Skulltrail, where the 45 nm CPUs are ready for quite a while already, but we need to wait a little longer for the board...

The vendors weren't exactly happy with the Barcelona B1 stepping results, the wait is now on for the B3 stepping, expected, oh well, anytime. I am expecting it too, it would be fun to compare Barcelona/Phenom QuadFX performance on RD790 CrossFire board vs the Nforce 680a SLI one (hint: 680a still seems a bit ahead... I didn't tell you this). µ