8800GTS Likenesses & differences



The cards we reviewed are all based on NVIDIA's reference design. This means they are all build on the A01 board revision, and they all use the dual slot active cooled heat piped aluminum heatsink. The G80 GPU is fed with 1,3V, while the memory on all cards is being fed with 1,91V. This makes sense since all our cards use Hynix HY5RS 1,1ns GDDR3 memory. This type of SDRAM is made for 1,8V supply voltage and should do its work up to 900MHz (1800MHz effective). With the good overclockable GPU and overvolted GDDR3 RAM NVIDIA left quite some performance in their 8800GTS cards. And that's exactly where the differences are to be found, here is a line-up of all our tested cards: (note: highest value in yellow)

GPU CORE CLOCK

Manufacturer	Box label clock	Realtime clock
Sparkle	500	513
Zotac	570	567
XFX	580	576
Gainward	550	540
PNY	500	513
Twintech	550	540
ASUS	500	513
Albatron	500	513
Leadtek	500	513
Calibre	580	576

During our review we noticed that the real time core clock Rivatuner showed us did not always match the numbers we saw on the boxes. Therefore it is most important to include those in our review. With the memory clock we noticed the same thing:

MEMORY CLOCK

Manufacturer	Box label clock	Realtime clock
Sparkle	800	792
Zotac	900	900
XFX	900	900
Gainward	880	891
PNY	800	792
Twintech	900	900
ASUS	800	792
Albatron	800	792
Leadtek	800	792
Calibre	890	891

Finally we like to mention the shader clock too. In our previous 8800 GTS 320 vs 640Mb review we learned that the 8800GTS cards come with 96 unified shader processors being clocked at 1200MHz. At least, that's how NVIDIA clocked them in their reference design. During our overclocking tests we saw that the shader clock is bound to the core clock in some way. The perfect example of what we noticed is how NVIDIA clocked their GTX. The GTX is clocked at 575MHz, right? Now notice how the shaders are also clocked at a higher speed compared to the 8800GTS: 1350MHz. With all those different core clocks we are being faced with in this review, it is good to have a close up on how all those manufacturers reacted on the above statement. Here is a line-up:

GPU SHADER CLOCK

Manufacturer	Box label clock	Realtime clock
Sparkle	1200	1188
Zotac	1200	1188
XFX	1500	1512
Gainward	1200	1188
PNY	1200	1188
Twintech	none	1296
ASUS	1200	1188
Albatron	1200	1188
Leadtek	1200	1188
Calibre	none	1350

It seems like most manufacturers stick with the reference clock speed (1200MHz), even those with overclocked GPU's. At first I thought that the XFX card was creating an error in the readout somehow, though XFX's Europe marketing manager informed us that it is complete normal that the shaders are clocked that high. Later on more manufacturers confirmed that the shaders indeed can be increased and that it can be used to extract extra performance. We will see the impact of those high clocked shaders in our game tests.

Cooling

9 out of the 10 samples we tested use an identical heatsink design. For those cards only different GPU speeds will determine different GPU temperatures, but even those were extremely closely to each other; for reference I’m including what you can expect cooling/performance wise with the standard heatsink design. Calibre's 8800GTS sample included the MACS M-Sorceress II MA-8280-2 aftermarket heatsink which we heared would offer great performance. We logged temperatures over a period of time with Rivatuner while running Futuremark 3DMark05 in loop to stress the GPU. The ambient room temperature is logged with a GTH 175/Pt Digital Thermometer

To insure that the heatsink is capable of keeping the 8800GTS at a decent temperature, we tried to measure how it performs inside our test housing:

The Antec Nine Hundred series, read our review here.

Here are the results:



Amazing how much heat these cards generate, you can actually feel warm air leaving the dual slot heatsink at the back of your pc enclosure. The heatsink/fan combination keeps the temperature well within specs. The fan is being temperature controlled; it will spin faster when core temperature increases to prevent overheating damage. With a duty cycle of only 70% at heavy load there is much cooling power left. As we tricked the fan speed to 100% with help of nTune, we saw how a high speed fan reduces the GPU temperatures with quite some margin.

Looking at our Calibre sample: Heatpipes... double 8cm fans... thermo electric cooling... it has it all! And just look at how it lowers the GPU temperature by more then 15°C on heavy load. It even manages to beat that stock cooler running its fan at 100% duty cycle!

Another fact is that the hot air is expelled outside the case with both heatsink designs, so thermal impact on overall case temperature is minimal.

Noise

Noise level was recorded with a SmartSensor AR824. The sensor was placed ~50cm away from the side view window of our Antec housing we mentioned above. The CPU fan was turned off to reduce noise and concentrate on the reading from the VGA cards.



The use of a big aluminum heatsink in combination with a heat pipe and a 7cm fan has proved successful to keep the 8800 GTS core temperature well within specs. Noise wise you can see that even at 70% fan speed the HSF remains quite silent, though I don't recommend tricking the fan speed to 100% for extra cooling in daily use, the added noise will certainly be noticed even if your average PC noise is already quite high. For Calibre's card we see that MACS has made the perfect balance between noise and performance. While in real life it didn't really sound noisier then other 8800GTS cards, it did offer us quite some lower GPU temperatures.

->