Santa Rosa: Spec Specifics, And Graphics, Video and Communications Plans
Continued from 'Santa Rosa: Intel's Livin' La Vida Loca'….
Eden claimed a further 0.87x SPECint boost in migrating from the T2400-based system to the T7700-derived follow-on, yet in this case a number of evolutionary factors are at work:
- Pentium M to Core microarchitecture
- 667 MHz FSB to 800 MHz FSB
- 1.83 GHz core clock to 2.4 GHz core clock
- 2 MBytes shared L2 cache to 4 MBytes shared L2
- (Presumably, albeit unspecified) faster-speed DDR2 main memory in the T7700 case
- etc.
Given the diversity of variables involved, it's unclear from Intel's incomplete information how other cherry-picked variants of Yonah and Merom would stack up against each other.
During the Q&A, I asked Eden about DirectX 10 support in the 965GM. It's planned, he was quick to respond, and will be fully implemented in hardware (prior-generation Intel graphics cores, in contrast, implemented pixel shaders in hardware but emulated vertex shaders via software running on the system CPU, for example). But it's not yet ready. From his vague response, it wasn't clear to me if the DX10 drivers just weren't yet mature, or if DX10 capability would require a future silicon spin. For now, though, Eden claimed that the GMA X3100, running at 500 MHz and capable of supporting Vista's full Aero interface, would deliver over 2x the performance on the DirectX v9-based Futuremark 3DMark '06 benchmark versus its predecessor graphics core (presumably referring to the GMA 950).
I also asked Eden whether or not the 965GM would hardware-accelerate portions of the H.264 and VC-1 video decode pipelines, thereby unburdening the CPU when (for example) playing back Blu-ray and HD DVD movies. Surprisingly, neither he nor anyone else from Intel in the room was able to answer the question, although subsequent Google research on my part unsurprisingly suggests that the answer's 'yes' for both codecs. How much of each codec's decoding is done on the graphics core versus in software remains unknown, however.
Speaking of video, it provided the basis for the one significant snafu of the morning. Eden and an assistant attempted to stream a (claimed) 50 Mbps high-def MPEG-2 video payload to two laptops, one with a 802.11a (presumably to dodge any existent 2.4 GHz interference) transceiver inside and the other containing Intel's Draft 802.11n module (again, presumably running in the 5 GHz band, since the 4965AGN supports both it and the more common 2.4 GHz Draft 802.11n flavour). The 802.11n-delivered stream should have been smoother than its 802.11a peer, but wasn't. After two attempts, the presenters gave up and moved on.
When I asked Eden how many OEM partners would be unveiling first-wave Santa Rosa systems containing Intel Turbo Memory, and in what density (or densities). He wasn't capacity-specific in his response but claimed than 10 OEMs, comprising more than 30 total systems, would be "time-to-market with us." Touted advantages of Intel Turbo Memory included an up-to-2x faster application load latency, up to 20% faster system boot times and (unqualified) power savings from reduced HDD spin-up.
Another journalist asked Eden about integrated 3G cellular data support, and the Intel VP regretfully admitted that the last-year touted partnership with Nokia to co-develop a HSDPA transceiver for Santa Rosa had been shelved because Intel "eventually decided that the return on development didn't justify the program." Instead, Intel is throwing its full weight behind WiMAX and, when I pressed him for details, he first pointed out Sprint and Clearwire's strong support and aggressive implementation plans for the technology. He then suggested that Intel would have a proof-of-concept card out by year end (whether this would take the form of an embedded mini-PCI Express module, an ExpressCard add-in, or both wasn't specified) with full integration in the next mobile chipset generation, scheduled to ramp into production in the 2nd half of next year.
