AMD Fx Processors

[RainColt]

Hi,

i want to upgrade my PC, I have always been on the Intel side, and now I saw the FX-8320 (8 core - 3.5 GHz) for about €150, thought it was awesome an ectacore for such a low price, I plan to use this for gaming and graphics(photoshop, video editing, CAD, ..). But watching benchmarks has the same, or even lower, performance of i5. I have some questions:

 

1. Is the CPU good for what I mentioned before? I don't want it become obselete in a year.
2. Can I really get the benefits of the eight-core? I read it is a kinda of 4 core, 8 thread.
3. Is this MotherBoard Asus M5A78L-M/USB3 good for overclocking, aroung 4,1-4,3 GHz? I found a good CPU Cooler, so it isn't a problem the temperature. 
4. I chose a GTX970 as GPU, I want to work and play in 4K (my cousin would give a 4K monitor for a low price), there could be are any bottlenecks?
5. Given the CPU and GPU, which PSU wattage would be good? I have a new 600W, can I still use it, or have I to change it?

Thanks for your patience, btw this is my 1st post in LTT 

For your application its kind of is a toss up between an i5 and an 8320, In editing photos and video, you'll get a slight edge from the 8320 granted that the applications are multithreading ready. In games the i5 doesn't always beat out the FX but has the potential to smoke the 8320 if the graphics bottleneck is removed which is not too common @1080 (and I don't think currently possible @4k ) unless the game is  very demanding on the CPU and has not alot of difficult things to render for graphics (MMO, Top down RTS, Some large scale FPS multi player) in these cases the Intel can push out almost 2x the frames of an FX processor.

 

1.The 8320 is already 2 years old, It's a good value but I don't think it will continue to trade punches for much longer it's that old.

2. Briggsy explains this perfectly

like @Sauron said, there are 8 cores, but paired into 2 cores per module. CPU instructions are sent to a module, where the work is divided between two cores. Intel CPUs have instructions sent to each core individually, which means that the FX processors can only receive half as many instructions per core as an Intel CPU, clock for clock, within a set time period. Really, its slightly more complicated than that, and not all software is compiled the same, but this is the underlying reason why FX processors are weaker in some cases.

3. I've never used it but I wouldn't think it would be acceptable for overclocking.

4. @4k your 970 would be the bottleneck.

5. I wouldn't think you'd be pulling more than 500W while running a torture test on both your overclocked CPU and GPU

[Faa]

What power supply actually? Surprises me that nobody asks which one it is. 600W by itself means as good as nothing
 

 

In editing photos and video, you'll get a slight edge from the 8320 granted that the applications are multithreading ready. 

Photoshop or any photo editing software will be much faster on a i5 than on a 8320, you're nothing with multithreaded performance there. Besides the workload of compiling a pic is significantly different than modeling one. The render times will be probably 5-10% faster than the i5 but the difference in responsiveness in a CAD is extremely huge regardless of what your synthetic benchmarks are saying. Same goes for Windows, multitasking is much better and again your OS feels more snappier. The i5 is just the better allrounder in single & multithreaded workload and it's just faster in almost any multithreaded workload.

Their multithreaded performance have been roughly the same and taking the most out of 4 cores is much easier than with 8. The 5960x scales sometimes so horribly to a point that it's not even performing worse than a 4790K.
 

 

(MMO, Top down RTS, Some large scale FPS multi player) in these cases the Intel can push out almost 2x the frames of an FX processor.

Hmm, thought you didn't believe that? Oo
Also adding another GPU is always risking a CPU bottleneck, potential is there as well.
 

 

2. Briggsy explains this perfectly

[spoiler=]Somewhat. Point is AMD offers per thread 2 ALU's capable of 2 instructions/cycle and Sandy Bridge offers 3 ALU's per thread capable of 3 instructions/cycle or 3 ALU's per two threads with HT. Haswell has 4 ALU's. Also there's no difference between an AMD ALU & Intel ALU, theyre physically the same. If hyperthreading now makes a difference that means you werent able to leverage all 4 ALU's with a single software thread, that's why HT (2nd hardware thread) can be useful but requires multithreading. It wouldn't make much sense having 8-way Hyperthreading if you only have 1 ALU and no FPU at all. People say AMD somewhat tried "Hyperthreading" since you can't always fully leverage all of your execution resources so AMD just duplicated the alu cluster. I find SMT more efficient than CMT though, we've seen 8-way SMT on IBM powerchips, there's no denying in that imo. AMD could have gone with a single ALU cluster (what people call a core these days) so putting the 4 ALU's in one cluster that would increase the integer singlethreaded performance. Not a brilliant move from AMD imo.

Anyways that's just ALU's, SIMD is a different story and I hate people saying it only has one FPU. It's just a 256bit FPU (they name it FlexFPU) singlethreaded or two 128bit FPU's when you have two threads pumped in, singlethreaded performance degrades slightly that way but this is gamechanging; if AMD used two Flex FPU's so two 256 bit FPU's it wouldn't have made any difference over one 128bit FPU simply because there's in many many apps and surely games no difference between a 128bit FPU & 256bit FPU. I've only covered here the back-end. The root cause of AMD's low performance is their front-end. It offers a 256bit FPU (which gives AVX support) over Nehalem and 1 more ALU and still doesn't have better singlethreaded performance. Going by the specs the back-end on AMD looks much better but its singlethreaded performance and therefore corecount vs corecount performance is nowhere near close either.

http://www.realworldtech.com/sandy-bridge/6/
This article covers the front-end issues; http://semiaccurate.com/2011/10/17/bulldozer-doesnt-have-just-a-single-problem/

The main differences I can say is; Intel has a superior front-end, dominating SIMD's completely, far better performance/watt resulting in being able to provide far more cores for a certain wattage than AMD can do and SMT makes a lot more sense than CMT.

 

 

And a "shit ton" of cores give a lot better rendering/enditing performance then fast 4 cores

A lot? It's pretty much a fact a i7 has better multithreaded performance and as far as synthetic benchmarks told us there's only a 20-30% advantage over the i5 so I'm not sure where you're getting that information from. If you haven't realized two Core 2 quads at 4.2GHz will outperform a 9590 in both multi & single. Proof: http://img.hwbot.org/u18227/image_id_1046283.jpeg

Performs a tiny bit better than a 4770K but tell me what you would prefer, nearly twice the IPC and a slightly lower multithreaded performance or those two C2Q's? Choice is made, 4770K. Unless you're rendering 24/7 like bitcoining it makes no sense to go with a 8 core that has the same multithreaded performance as a quadcore. You saw some synthetic benchmarks, who knows you haven't seen one and just copypasting what most people spray, seeing a 8350 doing better means automatically it's the better multitasker or it increases your singlethreaded performance lineairly by 8 times. That's just flawed logic. Learn a thing about multithreading then you'll understand what I'm meaning here, plenty of information available on the net.

[RainColt]

 

-snip-