Although the GeForce FX 5700 Ultra is already familiar, NVIDIA's new spin of it provides an early look at GDDR3's potential for graphics implementations. Read on to see what GDDR3 does for the 5700 Ultra, and what the new memory type could do for next-gen graphics products.
Despite using swanky new memory chips, NVIDIA's GeForce FX 5700 Ultra with GDDR3 is almost indistinguishable from the old DDR2 card.
NVIDIA tapped Samsung to provide the memory chips for this 5700 Ultra card. The K4J55323QF-GC20 chips are rated for operation at clock speeds up to 500MHz and run on only 1.9V. For comparison, the original FX 5700 Ultra's memory chips were only rated to run at 450MHz and required 2.5V.
Higher clock speeds and lower power consumption highlight the main benefits of GDDR3 memory over DDR2, but the FX 5700 Ultra only scratches the surface of GDDR3's capabilities. Samsung currently makes GDDR3 chips rated for operation at speeds up to 800MHz—300MHz faster than its best DDR2 chips. And even at 800MHz, Samsung's GDDR3 chips don't require more than 1.9V.
Although NVIDIA isn't using exotic 800MHz GDDR3 chips with the FX 5700 Ultra, GGDR3 does give the new GeForce FX 5700 Ultra little more juice. The new card's memory clock speed has been increased to 475MHz, or effectively 950MHz once we take DDR memory's clock-doubling scheme into account, giving it a slight memory bandwidth edge over its predecessor.
|Core clock (MHz)||Pixel pipelines||Peak fill rate (Mpixels/s)||Texture units per pixel pipeline||Peak fill rate (Mtexels/s)||Memory clock (MHz)||Memory bus width (bits)||Peak memory bandwidth (GB/s)|
|GeForce FX 5700 Ultra||475||4||1900||1||1900||906||128||14.4|
|GeForce FX 5700 Ultra GDDR3||475||4||1900||1||1900||950||128||15.2|
The GDDR3-equipped GeForce FX 5700 Ultra still has a 128-bit memory bus, so the boost in memory bandwidth only works out to about 800MB/sec. Of course, it's possible this slight increase in peak theoretical bandwidth could be offset, at least in part, by higher memory access latencies with GDDR3. (We don't have an exact means of measuring memory latencies on a graphics card, unfortunately.) Also, NVIDIA isn't fiddling with core clock speeds for the new 5700 Ultra, so its theoretical pixel and texel fill rate peaks are unchanged.
|MSI bringing one of everything to Computex||12|
|First GeForce GTX 1080 driver out with new VRWorks features in tow||9|
|AOC set to release quantum-dot-flavored monitor||14|
|Thermaltake's Level 10 M Advanced mouse offers 16000-DPI sensor||15|
|Customer frustration leads to Windows 10 upgrade dialog changes||45|
|HP joins the black and red clan with Omen gaming PCs||15|
|Crytek releases Cryengine source code on Github||30|
|Zotac beefs up lineup of mini-PCs for Computex||23|
|Toshiba releases 8TB X300 HDD||23|