Gflop ratings are theoretical, but the fact that a 7870 is as fast as two 7770s is not. Similarly, a 7970 is twice as powerful as a single 7870. The number of compute units scales linearly between each level, however the 7700 series in particular is limited in memory bandwidth drastically compared to the 7800 series due to the small bit width. If you want to see numbers of how powerful each 7000 card is, you should be looking for bitmining hash rates to see this play out as stated here.

Your thread supports what I stated, so what's your problem? These users also don't seeming to use an app_info.xml file, so you should advise them to do so to improve GPU utilization and maximum work throughput. And yes, you do have that flexibility with a single card, you just don't have that flexibility with low grade cards. I run diablominer and this project at the same time, while also watching 1080p videos with my 7950. The 7950 is supposed to be 3x the cost of a 7770, it is 3x the hardware, your logic doesn't make any sense. You think you can pay less for more?

Let's reiterate shall we?

HD 7770 = 10 Compute Units.
HD 7950 = 32 Compute Units.

That's when you buy a second motherboard and processor for a new machine. Isn't this standard practice with typical BOINC hobbyists? If you are using an AMD Opteron server however, you could better utilize the card, however you can also reduce it from 4 work units to 2 work units. It isn't as if using four work units puts twice as much load on the graphics card, the load is the same, but it helps alleviate the wait time between completed work units a bit. It is also recommended to have very high frequency RAM in your system for when the GPU needs to flush contents to RAM and more data is sent to it.

Oh, so you are just going to ignore all the facts and an actual professionals word, and instead substitute your own theory based on a few results from people not effectively utilizing their cards to their full potential? Really now.

These cards have been tested time and time again over the last year, 7970 is 3x faster than 7770 at any kind of computations.

https://en.bitcoin.it/wiki/Mining_hardware_comparison
Results don't die, neither does the facts straight from AMD's website. Do you really think a piece of equipment with a 128-bit memory width lane and 10 compute units, can compare to the model with 32 compute units and a 384 bit interface? This project, and others, demonstrate this perfectly. How fast your kernel time is depends on the frequency of the graphics card, the number of compute units, and how much of the GPU is being utilized. Because HCC does not fully utilize a single 7970/50 (realistically a single work unit uses half of my card), you instead write up an app_info to run 2-4 work units at the same time so that it does utilize the entire graphics card. Therefore, simply going by the kernel time of one work unit is really silly. Do your research next time.

I've been on every distrubuted computing project for graphics cards, and was a particularly avid tester active in POEM forums during its early beta. What you call 'resources' are things such as memory speed, bit width, PCIE link width, memory frequency and timing, dual/tri/quad channel system memory, and the CPU that brings it all together. Within a single family of graphics cards, a compute unit is an absolute unit that remains the same in all models. The more compute units you have is how much more power the graphics card has. A single compute unit in a 7970 is the same as a single compute unit in a 7770. However, resources such as the problematic memory width are devastating for projects that make frequent access to memory. There arent many OpenCL programs that have a lightweight imprint with memory, bitming is one of them It seems you don't know about hardware. What you posted is about as informational as random rumors off the street. If you are going to ignore facts, you might as well back off from technical discussions. I already demonstrated in your prior post how your link was worthless as a benchmark of how fast these cards are in this project. But you are going to ignore that as well too.