R520 = Dissapointment
- Thread starter Hellbinder
- Start date
OpenGL guy
Veteran
It runs cooler than my R480 board.Waltar said:
Joe DeFuria
Legend
DemoCoder said:
Agreed....To be clear, I was not trying to imply that in an absolute sense, the 520 has more "shader power" than G70 given complex shaders. Only pointing out that generally, the more complex shader you have, the "more powerful" that R520s shaders should appear.
Whether or not we could generalize one card having more "shader power" than the other would best be left to a future exercise with a history of a bunch of diverse shader apps / games to test with.
Jawed
Legend
It'll be interesting to see if any games ever actually use FP textures and filter them.Sigma said:
Humus's discussion:
http://www.beyond3d.com/forum/showthread.php?t=23276
seems to suggest that FP16 texture filtering in dedicated hardware is the stuff of dreams.
But then again, FP16 blending with AA in the render target/back buffer was a bit of a dream until recently...
Yes, but it seems it's only supersampling of transparent areas - not the "dithered" technique that 7800GTX also supports.
Jawed
Yay! Monarch! That's where I got my X800 pro from.SugarCoat said:
Shipping date Oct 12th...
Really? Do R480s run a lot hotter than R420s? Or are they about the same? (Just trying to get an idea)OpenGL guy said:
The card delivers pretty much what I expected out of it. Wins some benchmarks and loses others. Pretty much par for the course with this generation. Unfortunately it will probably be 5-6 months late in terms of quantity in the channel. That is just abysmal if you think about it.
When the R580 comes out it will be interesting to see what Nvidia has planned to counter it.
When the R580 comes out it will be interesting to see what Nvidia has planned to counter it.
OpenGL guy
Veteran
Not sure off the top of my head and I don't have an R420 handy. Didn't some web site already post this info some time ago?digitalwanderer said:
so the XL cant beat the 7800GTX and the 1800XT wont be out until mid-November and will be a 2 slotted beast with a captain crunch plastic fan noisemaker.
I wonder what the prices of the 7800GTX will be mid November? Right now they are at $419 on newegg.com The fact that the 1800XT is using 512MB of ram and that ram is faster than the 7800GTX makes you wonder how much performance is just from the ram itself. If nvidia has a 7800GTX "ultra" with 512MB of fast ram I wonder if it would not be faster than the 1800XT
I wonder what the prices of the 7800GTX will be mid November? Right now they are at $419 on newegg.com The fact that the 1800XT is using 512MB of ram and that ram is faster than the 7800GTX makes you wonder how much performance is just from the ram itself. If nvidia has a 7800GTX "ultra" with 512MB of fast ram I wonder if it would not be faster than the 1800XT
Bouncing Zabaglione Bros.
Legend
Junkstyle said:
It's very a very quiet fan according to the reviews, and it exhausts hot air out of the back of your case.
Last edited by a moderator:
broke 10k in 3Dmark05 with a single card, GPU overclock only. pretty sure its MSI based. that didnt take long.
http://www.xtremesystems.org/forums/showthread.php?t=75973
I don't know if I'd qualify it in that way. It's not necessarily the complexity of the shader that is important, but rather how instructions are ordered and grouped.Joe DeFuria said:
Basically, nVidia's hardware will do best if the texture reads are distributed at nearly even intervals throughout the shader, as are the special functions. ATI's hardware doesn't much care which order the instructions come in. In other words, the decoupled texture and ALU units result in a dramatic reduction in the number of scenarios under which the pipelines will stall. This isn't related to more complex shaders, just different ones.
Some of this may be managed by working more carefully on optimizing the shaders, either through developer effort or through better compilers. Some of the time it just won't be possible.
Register usage is still the crunch on NVIDIA's parts.
ChronoReverse
Newcomer
According to Hexus
What, you expect me to hunt it up?OpenGL guy said:
I think they are about the same actually, but I'll poke around and check. My 420 idles around 37 and peaks around 61 I think.
I doubt it's that much of an issue. From what I can tell, register usage only relates to how many of the NV4x's functional units can possibly be active for any given clock cycle, and the minimal usage is two FP32 operations per clock. NV's parts were typically ahead of the R4xx on a per-clock shader op basis because they have more functional units. The fact that these units aren't always able to be used due to register pressure just places a maximum limit on the possible amount of performance that nVidia can get out of the architecture.Dave Baumann said:
No, I'm really convinced that the reason that the R5xx sometimes tramples the NV4x is due to register pressure. If it were, enforcing FP16 for all operations should relieve this pressure and bring the NV4x back up....I doubt that it would. I'm willing to bet that it's due to pipeline stalls that the NV4x is experiencing that the R520 is not.