Single page Print

Nvidia's nForce 750a SLI chipset

HybridPower comes to the mid range
— 3:42 PM on July 21, 2008

Nvidia's flagship nForce 780a SLI chipset for Socket AM2+ processors debuted a couple of months ago, bringing all-new silicon loaded with a Phenom-ready HyperTransport 3 link, integrated graphics with HD video decoding and HybridPower capabilities, and support for three-way SLI. While it's a technical marvel, this high-end chipset didn't really make sense to me—it's too much like wrapping the guts of a Ferrari around the engine from a Golf GTI. There's nothing wrong with the Golf, of course, but you're only working with so much horsepower, which makes the Ferrari bits overkill. What you need for a high-end exotic is an engine with a little more grunt.

If you're building a chipset for Socket AM2+ processors, there's only so much grunt available. The Phenom just can't keep up with Intel's fastest Core 2 Quads, forcing AMD to price itself out of the high end of the processor market. The most expensive CPU AMD currently makes is the Phenom X4 9950 Black Edition, which, at just $235, costs nearly as much as the cheapest nForce 780a SLI motherboards on the market. Hardly a match made in heaven.

What you really want to pair with a Phenom is a mid-range motherboard around the $150 mark. Something like XFX's $149.99 MD-A72P-7509, perhaps. This board is based on the 780a's little brother, the nForce 750a SLI, and the two are very much alike. Keep reading to see how the 750a stacks up against not only its big brother, but also AMD's 790FX chipset in a battle for Socket AM2+ supremacy.

All but an nForce 780a
Those not already familiar with the nForce 780a SLI would do well to read our initial coverage of the chipset. The 750a is very similar, and it doesn't bring any new features and capabilities. Mostly, it just takes some away.

The nForce 780a SLI is a two-chip, er, chipset made up of an nForce 780a SLI MCP that contains all the basic core-logic functionality and an nForce 200 SLI chip loaded with PCI Express lanes. The 750a ditches the nForce 200 completely, and despite a change in name, its MCP component is essentially identical to that of the 780a. This 65nm chip hosts not only core logic elements, but also an integrated graphics core very similar to what you'll find in the GeForce 8300.

Dubbed a motherboard GPU (mGPU), this graphics core is fully DirectX 10-compliant with 16 stream processors and support for Shader Model 4.0. More importantly, it's capable of full Blu-ray decode acceleration across AVC, VC-1, and MPEG2 formats, with HDMI output thrown in for good measure. HD video decoding is a neat trick, but it's not nearly impressive as the mGPU's HybridPower implementation. HybridPower allows the mGPU to take over for compatible discrete graphics cards at idle, completely shutting them down for substantial power savings.

The energy efficiency proposition for HybridPower becomes even more compelling with SLI configurations, where watts saved are multiplied. However, without an nForce 200 SLI chip riding shotgun, the 750a is limited to only 19 lanes of PCI Express 2.0 connectivity—not nearly enough for three-way SLI. 16 of those lanes are reserved for graphics and can be split evenly between a pair of cards for two-way SLI. Thanks to the higher signaling rate of PCI Express' second coming, this dual-x8 link provides just as much bandwidth as a dual-x16 link with gen-one PCIe.

Of course, Nvidia has long been adamant that the nForce 200 is more than just a collection of PCI Express lanes. The chip includes broadcast and posted-write shortcut functions that can reduce the bandwidth used by SLI configurations. One might expect these functions to be absent in the 750a, but Nvidia tells us they've been built right into the MCP. We'll see how the 750a's SLI performance compares with that of the 780a a little later in this review.

In addition to differences on the SLI front, the only 780a feature missing from the nForce 750a is support for Enhanced Performance Profile (EPP) memory. Yes, you'll have to set memory timings manually, just like in the olden days when we walked to school through knee-deep snow, up-hill both ways. EPP memory is certainly a welcome convenience, but I'd wager that many mid-range buyers would balk at paying the premium associated with EPP DIMMs, anyway.

AMD 790FX Nvidia nForce 780a SLI Nvidia nForce 750a SLI
Processor interface 16-bit/2GHz HyperTransport 16-bit/2GHz HyperTransport 16-bit/2GHz HyperTransport
PCI Express 2.0 lanes 38 35 19
Multi-GPU support CrossFire 2, 3-way SLI 2-way SLI
Chipset interconnect PCIe 1.1 x4 PCIe 2.0 x16 NA
Interconnect bandwidth 2GB/s 16GB/s NA
Serial ATA ports 4 6 6
Native Command Queuing Y Y Y
RAID 0/1 Y Y Y
RAID 0+1/10 Y Y Y
ATA channels 2 1 1
Max audio channels 8 8 8
Audio standard AC'97/HDA HDA HDA
Ethernet N 10/100/1000 10/100/1000
USB ports 10 12 12

Since they're based on the same MCP chip, the nForce 750a has the same Phenom-ready HyperTransport 3 link as its high-end brethren, the same six 300MB/s Serial ATA RAID ports (that work in AHCI mode, I might add), the same dozen USB ports, and the same Gigabit Ethernet controller. Reusing MCP chips in different products to hit a range of price points makes good business sense for Nvidia. This approach also pays dividends for enthusiasts, who benefit from high-end features trickling down to more affordable parts.

Before moving on, I should note that you won't find a chipset interconnect or interconnect bandwidth listed in the table above because the 750a is a single-chip solution, with need for a chip-to-chip interconnect. It's an elegant approach that eliminates a potential bottleneck.