Viper 770s at Stock Speed
| Benchmark Type |
16 Meg |
32 Meg |
| Quake 2 Demo1 640x480 |
73.2 |
74.1 |
| Quake 2 Demo1 800x600 |
48.6 |
49.0 |
| Quake 2 Demo1 1024x768 |
38.1 |
42.0 |
| Quake 2 Crusher 640x480 |
47.0 |
47.3 |
| Quake 2 Crusher 800x600 |
42.8 |
42.9 |
| Quake 2 Crusher 1024x768 |
35.0 |
34.8 |
| Quake 3 Demo 1 |
32.8 |
33.0 |
| Quake 3 Demo 2 |
54.3 |
54.3 |
| 3D Mark 99 Max 800x600 |
5096 |
5093 |
As you can see, the scores are
pretty much a wash. The 16 meg Viper V770 is every bit as fast as the more
expensive 32 meg version. But would this hold up when we overclock the two
cards?
To overclock the Viper V770s I put a
50mm fan on top of the heatsink and some PCI strips on the memory chips. Since
both card are the same except for memory, they should overclock to the same
amount. It turns out that the 32 meg Viper can have its memory overclocked
higher than the 16 meg Viper. The 16 meg Viper can have its core overclocked
higher than the 32 meg version.
Ready to rock!
I was able to get the 32 meg Viper
running at 160Mhz core and 185Mhz memory. This was lower than the last 32 meg
Viper V770 I had which could run at 170/200. The 16 meg Viper was able to run
its core at 170Mhz but its memory can only do 175Mhz. This is most likely due to
the production variances of each card and should not be taken any other way.
Just because my 16 meg Viper can run its core higher than the 32 meg Viper
doesn't mean all 16 meg Viper can do that.
Viper V770
Overclocked
| Benchmark Type |
16 Meg 170/175 |
32 Meg 160/185 |
| Quake 2 Demo1 640x480 |
73.5 |
75.8 |
| Quake 2 Demo1 800x600 |
70.3 |
68.1 |
| Quake 2 Demo1 1024x768 |
45.0 |
43.9 |
| Quake 2 Crusher 640x480 |
47.0 |
48.1 |
| Quake 2 Crusher 800x600 |
45.8 |
44.5 |
| Quake 2 Crusher 1024x768 |
39.0 |
42.6 |
| Quake 3 Demo 1 |
41.1 |
41.5 |
| Quake 3 Demo 2 |
60.9 |
61.1 |
| 3D Mark 99 Max 800x600 |
5113 |
5110 |
Again, there is no clear winner
here. Both cards posted nearly the same scores across the board.
Really
High Texture Benchmarks
| Benchmark Type |
16 Meg |
32 Meg |
| Quake 3 Demo 1 800x600 |
19.7 |
41.0 |
| Quake 3 Demo 1 1024x768 |
12.7 |
27.3 |
| Quake 3 Demo 2 800x600 |
27.4 |
65.0 |
| Quake 3 Demo 2 1024x768 |
14.7 |
47.9 |
The benefits of 32 megs of ram
finally shows itself when we really crank up the textures and resolutions. I ran
the same tests again using Quake 3 Test but this time I max out EVERYTHING. As
you can see, the 32 meg Viper just destroys the 16 meg Viper.
This kinda shows that 3dfx was
correct when it stated that AGP 2X was not fast enough to replace on board ram
which is why 3Dfx cards don't have AGP texturing. The purpose of AGP texturing
was that if the card should run out of memory it can texture out to the AGP
port. This works well with most games but as we can see in the above, games are
coming where AGP 2X will not be acceptable.
Given the results I got, I don't
really see a reason to get a 32 meg Viper over a 16 meg one. The 16 meg Viper is
every bit as fast as the 32 meg version when running today's games. Quake 3 is
another animal all together. By the time the game ships, NVIDIA's NV10 and
3dfx's Voodoo 4 will be out ready to run it better than TNT2 can ever dream
possible.
While the 32 meg Viper has a big
advantage in ultra high texture games, the difference isn't enough to make up
for the savings. Quake 3 running with everything maxed out doesn't look all that
much better than default "high quality" mode which runs very well on a 16 meg
Viper.
If it was up to me, I would get a 16
meg Viper now and a NV10 later. Since the NV10 is just a few months away and
will be up to 4 times faster than a TNT2 Ultra, I think that would be the better
upgrade path.
