Archives Discussions

[tkoptur1]

It's been so long time that there is no support for PCI 3.0 and memory 2133 without o.c. when it comes to amd mobo's. Will we see new mobo's coming out soon and supports these in early 2013? I would love to use new ati radeon hd 7900 series and a good 16gb with 2133 but with amd mobo's, I dont think it is possible for now which is very sad!

Thank you,

[Brane2]

2133 makes little sense on existing CPUs.

But high freq DDR3 is nice for APU, so this is why Trinity can use 1866 and Richland will alegedly go as high as 2133 stock...

[avk]

AFAIK, even the very latest and most powerful graphic cards don't really need PCI 3.0, because PCI 2.0 is still enough for them.

As for 2133 MHz RAM: I can assure you that the speed of your RAM is not as important as someone can think, especially if you would try to compare 2133 and 1866. I'm sure that you're aware that every RAM module has it's own parameters, so you can easily find a 1600 MHz RAM module that is faster than a 1866 MHz counterpart - just because of their timings.

[Brane2 · Accepted solution]

Yes, on classic CPU lower frequency with less latency might be same or better.

But APU is different beast. GPU part relies on bandwidth part MUCH more heavilly. Also picture generation needs it.

So, fast RAM can be substantially better on APU, even with higher latencies.

PCI-3.0 might be beneficial in multiple card setups or with GPU computing, but for simple usual apps it probably won't make measurable difference.

[avk]

Please look at the first post: the man didn't mentioned APU, instead - he mentioned high-end GPU, Radeon HD 7900 Series.

[tkoptur1]

If there wont be much difference between 1866 and 2133, then it should be fine, but most of the intel products are getting alot more high benchmarks even compared to the latest amd cpu's and this is very sad.

[Skysnake]

Sorry, but this is 100% wrong.

Ok, not 100%, but it depends extremly on your Problem.

There are lots of problems out there that are heavely limited bye your transfer speed from CPU to GPU Ram. For example all small problems with small datasets are heavely limited bye the PCI-E bus, because this is allways a overhead, and with small problems you have only small benefits from the faster GPU, so it can very fast happen, that only the data transfer over PCI-E took longer than the calculation on the CPU.

But not only this! There are also problems that have very little datareuse. Then it can happens that you are completly limited from your PCI-E bandwidth. Sorting should be sometimes such a problem.

Thats one point why K20 is not soooo cool. It just have PCI-E 2.0.

But the most funny part is: AMD have the only HPC DoublePercision Accelerator with PCI-E 3.0 and Intel the only platform , hey and nVidia.. hey, they just do there job with the consumer cards... and the funny K10.

[avk]

Sorry, but the topic starter was never said something about HPC and nVidia's K10&K20. He was talking about the consumer graphic card, namely Radeon HD 7900 Series. And, speaking about that model, there is almost no difference between PCI 2.0 and 3.0.

[avk]

It's just because of the weak memory & cache subsystem in the AMD CPUs, comparing them with the Intel ones. As soon as AMD will improve it, then its CPUs will use high speed RAM more efficiently. Until then, we should just wait or buy the rival's production. Sorry .

[Skysnake]

When AMD can do one thing, than build very very good Memory-Controller!

The memory controller have more or less nothing to do with the cache system... And also the cache system is not really bad. Only the L1 Cache in the BD-Modules is not really a golden egg.