Nvidia's Ion: A Hands-On Try-On

Remember Nvidia’s 9400M single-chip CPU companion? You first heard about it from me in mid-October, as perhaps the biggest-news aspect of Apple’s next-generation MacBook, MacBook Air and MacBook Pro (where it was paired with Intel’s ‘Penryn’ 45nm CPU). In mid-December, I informed you that Nvidia was aspiring to pair that very same IC with Intel’s Atom processor. And in early January, I suggested that if successful, this pairing might fundamentally restructure the parameters of the PC architecture. By doing so, it would redefine the balance of power between Intel and other participants in the PC ecosystem.

When Nvidia briefed me in early December, company representatives expressed an aspiration to get a reference system in my hands by the end of January. And in fact they did…barely A Fedex shipment showed up on my front door last Thursday morning, and I immediately dove into the benchmarking task, as an adjunct to my already-published data on various Intel- and Via-based ‘value’ designs. As reminder, here are a few shots of the diminutive Nvidia system that I put under the microscope:

And here are some key system parameters:

Note that whereas the hardware I previously benchmarked was based on Windows XP SP3, the system Nvidia sent me came with Windows Vista Enterprise pre-installed. Also, whereas SiSoftware’s Sandra is now at version 2009 SP2, I used the then-latest-and-greatest 2009 SP1 variant when I did my earlier testing in mid-November 2008. Fortunately, I still had a copy of the earlier Sandra Lite version archived, and chief developer Adrian Silasi assured me that the Vista-vs-XP differential between this and my prior tests would have little to no impact on the end numbers (in no small part because Sandra does such an effective job of isolating its focus to a particular subsystem and minimizing the software overhead between itself and that particular subsystem hardware).

I admit that I’m more than a little uncomfortable with testing a vendor-preconfigured system, because I’ve got no assurance (for example) that the software and driver versions I’m evaluating are comparable to those available to you now or soon. However, schedule limitations forced me to go the pre-configured route in this particular situation; to be clear, I’ve got no evidence that Nvidia is cheating…this time, at least…Nvidia’s Ion reference platform contains the same 1.6 GHz Atom CPU (therefore with the same single-core HyperThreading-capable architecture, 533 Mhz front-side bus speed, on-board L1 and L2 cache sizes, etc) found in the Intel D945GCLF mini-ITX board included in my earlier testing. Note that whereas the prior systems employed SSDs, the Ion reference design leveraged a Seagate 7200 RPM SATA HDD with a 16 MByte buffer. Also, this particular design incorporated 2 GBytes’ worth of DDR3-1066 SDRAM, whereas the earlier Atom-based designs’ Intel chipset limited their memory to DDR2-533 speeds.

Below, please find updated (to those found in my early-January cover story) Sandra data-sourced graphics, incorporating the Nvidia Ion results.

ALU Performance

Cache and Memory

Cryptography

Java Arithmetic

Java Multi-Media

Memory Bandwidth

Memory Latency (Linear)

Memory Latency (Random)

.NET Arithmetic

.NET Multi-Media

Processor Arithmetic

Processor Multi-Media

I’ve added Sandra’s benchmark report files on Nvidia’s Ion to the consolidated ZIP archive that I’d earlier provided you; you can find the augmented version of the file here. Similarly, the updated Excel spreadsheet consolidation of the Sandra-derived data can be found here. Below are the idle and peak power consumption statistics I measured using my Kill A Watt hardware on Ion, which are inclusive of all subsystems save the LCD:

Perhaps not surprisingly, those benchmarks which most heavily exploited available DRAM subsystem performance potential without simultaneously bottlenecking the Atom CPU’s front-side bus showed the greatest gains in comparing the Nvidia Ion-derived data to earlier Atom 230- and 270-based hardware results. Other tests were comparable (to within normal benchmark run-to-run variances), if not slightly slower in the Ion case…perhaps a reflection of Windows XP-vs-Vista variance, versus any fundamental hardware discrepancy. However, Sandra wasn’t able to encompass all aspects of perceived system performance, thereby encouraging me to conduct additional studies which I’ll discuss next.

Continue reading with part two, ‘Nvidia’s Ion: A Video Evaluation‘…