A Three Year Age Difference
We already described the characteristics of the GeForce 9400/9300 chipset when we first tested it—see Move Over G45: Nvidia's GeForce 9300 Arrives. The GeForce 9400M is an almost identical product, featuring the same GeForce 9-series core with its 16 stream processors. The biggest difference we can find between the desktop 9400 and its mobile version is simply the clock frequencies of the core and shaders: 580 MHz /1,400 MHz for the 9400, and 450 MHz/1,100 MHz for the 9400M.
Aside from the slower clock speed, which is necessary in order to keep power consumption low, the features we liked on the 9400 are still there: DDR2-800 or DDR3-1333 memory compatibility, support for all Intel processors with 1066 MHz FSB maximum (so far there are no mobile Intel CPUs with a faster FSB), support for dual-link DVI, HDMI, and DisplayPort interfaces, Gigabit Ethernet, an HD Azalia audio codec, a maximum of 12 USB 2.0 ports, and 3 Gbps SATA ports. As for peripheral interfaces, the mobile GeForce manages one PCI Express x16 slot, four PCIe x1 slots, and five PCI slots. The GeForce uses a 65 nm fabrication process, and has a total die surface of 1225 mm2 (35 x 35 mm). Nvidia rates its TDP at 12 W.
The 945GC chipset is part of the large “Lakeport” family of Intel chipsets, which has no fewer than seven members. If you count their “mobile” cousins from the Calistoga family, there are a total of 13 memebers. That’s a lot of core logic, but the differences between each model are subtle. All you need to know is that the 945GC is the chipset Intel sells coupled with its Atom desktop processors (the 230 and 330). For its netbook mobile Atom processor (N270), on the other hand, Intel requires that its 945GSE chipset be used.
GeForce: Smaller, With Better Performance
These chipsets are far from being brand new. The first representatives of the Lakeport family hit the market in Spring 2005 (early 2006 for the Calistogas). One of the consequences of their advanced age is that the 945GC uses a 90 nm fabrication process. The 945GC is also not a single-chip product like the GeForce 9400M. The 82945GC northbridge has to be attached to a 82801GB southbridge, also known as ICH7.
The consequences of this are twofold. First, the 945GC + ICH7 combination eats up a lot of real estate: the northbridge measures 34 x 34 mm, and the southbridge 31 x 31 mm. Add a few square millimeters of space between the two ICs and you end up with twice the surface area taken up on the PCB when compared with the GeForce 9400.
The second impact of the older design is that total power consumption is high. According to Intel, the 945GC’s TDP is 22.2 W, and the Atom 230 + 945GC + ICH7 platform has a TDP of 29.5 W. The GeForce 9400M claims a TDP of only 12 W. The Atom itself consumes 4 W, putting the Ion at 16 W total—just above half the TDP of the Intel platform.
Finally, let’s look at performance. The 945GC’s integrated graphics core is the well known (and not necessarily for good reasons) GMA950. It’s DirectX 9- and Shader Model 2.0-compatible, clocked at 400 MHz, and armed with four pixel shaders, one vertex shader, and four ROPs. In other words, it’s from the pre-DirectX 10 unified shader era. It uses part of the main memory as video memory. On the D945GCLF, the inclusion of a single DIMM slot reduces bandwidth to approximately 5 GB/s. By comparison, the GeForce 9400M, with its DDR3-1066 slot, has 8.3 GB/s of bandwidth.
Theoretically, out of the starting gate, the GeForce 9400M has an enormous advantage over its only current competitor. But will that theoretical lead hold up in practice? Isn’t so much graphics power for a CPU as weak as the Atom casting pearls before swine? And is the Nvidia chipset good at anything besides graphics?