facebook rss twitter

Review: MSI vs. ZOTAC: shootout at the GeForce 8800 GTS 512 Corral

by Tarinder Sandhu on 5 February 2008, 08:22

Tags: MSI

Quick Link: HEXUS.net/qaljq

Add to My Vault: x

GeForce 8800 GTS 512, and ZOTAC's card



NVIDIA has kept the GeForce 8800 GTS moniker intact in the move from the 320MiB- and 640MiB-equipped cards that were released around a year ago, but the GeForce 8800 GTS 512 adds in more than just a different-sized frame-buffer.

Built on a 65nm manufacturing process; endowed with 128 stream processors; a 256-bit memory interface; VP2 video-processing engine; PCIe 2.0 connectivity; and double-G80 bilinear filtering speed - its brute power will see it faster at rendering than a GeForce 8800 GTX in many cases.

The basic statistics can be though of as a GeForce 8800 GT - also recently introduced - but with performance cherries on top.

Ensuring that it remains on reasonably level footing with the G80s that it competes against, the GTS 512 is hamstrung via a 256-bit memory interface, although the near-2GHz GDDR3 clocking makes up for bandwidth deficits.

The extra oomph over and above a GeForce 8800 GT necessitates a dual-slot-taking cooler, though.

Take a look at the various stats in table-mode.

Graphics cards NVIDIA GeForce 8800 GTS 512 NVIDIA GeForce 8800 GT 512 NVIDIA GeForce 8800 GTS 320 NVIDIA GeForce 8800 GTX 640 NVIDIA GeForce 8800 GTX 768
PCIe PCIe 2.0 PCIe 1.x
GPU clock 650MHz 600MHz 500MHz 500MHz 576MHz
Shader clock 1625MHz 1500MHz 1188MHz 1188MHz 1350MHz
Memory clock (effective) 1940MHz 1800MHz 1576MHz 1576MHz 1800MHz
Memory interface, size, and implementation 256-bit, 512MiB, GDDR3 256-bit, 512MiB, GDDR3 320-bit, 320MiB, GDDR3 320-bit, 640MiB, GDDR3 384-bit, 768MiB, GDDR3
Memory bandwidth 62.08GB/sec 57.60GB/sec 63.04GB/sec 63.04GB/sec 86.40GB/sec
Manufacturing process TSMC, 65nm TSMC, 65nm TSMC, 90nm TSMC, 90nm TSMC, 90nm
Vertex, fragment, geometry shading (shared) 128 FP32 scalar 112 FP32 scalar 96 FP32 scalar 96 FP32 scalar 128 FP32 scalar
Peak GFLOP/s 624 504 342.1 342.1 518.4
ROPs 16 16 20 20 24
Peak GTexel/s (bilinear) 41.6 33.6 12 12 18.4
Hardware-assisted video-decoding engine NVIDIA's VP2 - full VC-1 decode and partial H.264 decode run via GPU's shaders
Reference cooler dual-slot single-slot dual-slot dual-slot dual-slot
Retail price (default-clocked model) £199 £160 £175 £199 £235





The ZOTAC card, being based entirely on the reference design, ships with a dual-slot cooler that's barely noticeable when the card's running at full chat.

The cooler's a little different to the GeForce 8800 GTS 320/640's, and whilst the GTS 512 is higher-clocked than the G80 GTS duo, it consumes less power, thanks to a more-efficient manufacturing process.

Now, the default clocks for a GTS 512 are 650/1625/1940 for the core, shader, and memory, respectively. ZOTAC has pushed these up to 678/1728/1944 for its AMP! Edition, pushing performance up a notch or two.


Two GTS 512s can be coupled together to form an SLI team, promising near-double performance. The 512MiB of GDDR3 RAM is stitched on the upper side of the PCB.

Note that there's only a single connector for inter-GPU communication. Three-way SLI remains the domain of G80 - GTX and Ultra - cards.


NVIDIA probably could have opted for a single-slot cooler for the GTS 512, but having a dual-slot model opens up the way for some serious overclocking. A single six-pin PCIe connector provides more than enough auxillary power.



Being a modern GPU released at the end of 2007, the '512 incorporates NVIDIA's VP2 video-processing engine that takes most of the computational load from decoding VC-1 (albeit not complete) and H.264-encoded media - HD DVD and Blu-ray, for example.

Both DVI ports offer dual-link compliance, meaning you can run a couple of 2,560x1,600 panels concurrently.


Summary

ZOTAC gets into the GTS 512 game by releasing a slightly-overclocked model with respect to core and shader speeds. We know that the core goes much higher, but NVIDIA's current policy is that partners cannot clock their core's above 678MHz. Overclocking above this figure is left to the customer.