Well I did do my Numerical Analysis homework back in the days.. and it tells me that there is no such thing as a 512bit application. you planning on writting any CUDA applications with your card? Hardly has anything to do with graphics now does it....

Wow this guy really is clueless.

Memory bus width on it's own is not a measure of bandwidth and therefore not a performance metric.

Who said anything about bandwidth? You can have all the memory bandwidth in the world, if the data still has to get through that narrow bus what good is it. Do you have both a GTX 280/285 and 4870/4890 like many of us here do. Is this another

Who said anything about bandwidth? You can have all the memory bandwidth in the world, if the data still has to get through that narrow bus what good is it.

It is clear you are not knowledgeable about that which you speak. Bandwidth is a measurement of how much data can be transferred from one device to another. Bus width is only part of the equation, which I've already shown you. Don't believe me? Lower the clockspeed of your GTX 285's memory to some nominal value significantly below stock and see what happens. All the bus width in the world doesn't mean a thing unless you have adequate clockspeed, and vice versa.

Do you have both a GTX 280/285 and 4870/4890 like many of us here do. Is this another "I have to justify the fact that I can't spend the money so the cheaper video has got to be better" thread?

It's not as though one needs to own both cards to determine their performance, that's what the internet is for. Hardware review sites publish these numbers for us.

Apparently heavy weather in FSX is not the only game where ATI gets clobbered. I just tested the 4870 with "Rise of Flight" and as soon as I turned the camera toward cloud coverage the performance dropped right off. I've experienced the same thing in certain racing sims when particle effects come into play such as tire smoke.

And you think this is a function of bandwidth? More likely to be a function of ROP capability, which ATi still lacks and NV does not.

It is clear you are not knowledgeable about that which you speak. Bandwidth is a measurement of how much data can be transferred from one device to another.

Is the memory bus not used between the GPU and the video cards memory? So then where is the weak link with the ATI? How come the 4870/4890 falls down with certain games and particle effects?

It's not as though one needs to own both cards to determine their performance, that's what the internet is for. Hardware review sites publish these numbers for us.

I don

Nvidia's ROPs are much more capable than ATi's with the lone exception of AA samples/cycle where ATi wins, as you can see in benchmarks where ATi cards lose to their competing Nvidia counterparts until you turn on 8x AA. Were this a function of bandwidth then surely the NV parts would win, as NV has an advantage here. Not only does NV have more capable ROPs, but they simply have more of them. 32 vs. 16 in the case of the GTX 280/285 vs. HD 4870/4890. Smoke/particle effects are usually a function of Z fill rate, which NV dominates. See the first post in this thread (click on show spoiler) at Beyond3d for real world numbers of the maximum fill rates of each various architecture. Note: testing is of HD4870 and GTX280 products. You'll see that NV's Z fill rate is nearly 3x as high as ATi's. This discrepancy was even greater before the HD 4000 series was released, as ATi's previous ROP architecture only allowed for 2 Z pixels per cycle whereas the current architecture doubled this number. NV's architecture allows the use of up to 8 Z pixels per cycle and with 32 ROPs this means 256 pixels/cycle in a Z-only fill rate, or when used in conjunction with AA up to 128 AA samples and 128 Z pixels can be output per cycle.

Nvidia's ROPs are much more capable than ATi's with the lone exception of AA samples/cycle where ATi wins, as you can see in benchmarks where ATi cards lose to their competing Nvidia counterparts until you turn on 8x AA. Were this a function of bandwidth then surely the NV parts would win, as NV has an advantage here. Not only does NV have more capable ROPs, but they simply have more of them. 32 vs. 16 in the case of the GTX 280/285 vs. HD 4870/4890. Smoke/particle effects are usually a function of Z fill rate, which NV dominates. See the first post in this thread (click on show spoiler) at Beyond3d for real world numbers of the maximum fill rates of each various architecture. Note: testing is of HD4870 and GTX280 products. You'll see that NV's Z fill rate is nearly 3x as high as ATi's. This discrepancy was even greater before the HD 4000 series was released, as ATi's previous ROP architecture only allowed for 2 Z pixels per cycle whereas the current architecture doubled this number. NV's architecture allows the use of up to 8 Z pixels per cycle and with 32 ROPs this means 256 pixels/cycle in a Z-only fill rate, or when used in conjunction with AA up to 128 AA samples and 128 Z pixels can be output per cycle.

Excellent information, thanks.I do remember a discussion over at Anandtech a while ago that did talk about ROP and particle effects.What