Stereo Tool

A few days ago I went to a PC shop to order a PC where I would also use Stereo Tool, I told them that I wanted an i5 or i7 on it, but they insisted on using an AMD (probably because their provider gives them cheaper or something), but I insisted that I wanted an Intel Processor. They said "why? they both follow the same instructions, where is the difference". I didn't know what to answer, although I know that in whatever multimedia proccesing application I use, Intel proccessors are much faster/stable. So basically this is a more theoretical question. Why Intel? Where is the difference?

As every damn PC store/shop, they first offer things where they have biggest profit. AMD is cheaper then intel. Also motherboards are cheaper for AMD. So they have less profit with intel.

Intel is just better for this. You don't need to tell them "Why".

I think you just answered your own question: Speed. They just seem to run faster (although I don't know if that's still the case with the latest AMD generations, I actually have the impression that I see fewer and fewer AMD's...)

http://www.cpubenchmark.net/cpu_list.php

Sort on Rank, then search for AMD. And skip all 8 and 16 core CPU's. You won't find anything that comes close in speed to the recent Intels, unless indeed you go for an 8 or 16 core CPU. (And if you wonder what's wrong with that: Using more cores makes writing software more difficult. In case of Stereo Tool, it makes sense to split the load by working on 2 cores, one for the left and one for the right channel. And for nearly all Intel CPU's that's sufficient - they either only have 2 cores, or if they have 4, those cores are usually fast enough to only need 2. There are exceptions... But if I want to process everything on 4 cores, the only way to do it is to process different chunks of audio simultaneously, causing more latency and I would have to build in a lot of things to synchronize the processing.)

Sorry to revive this thread. I'm new to this forum but I think many will find it interesting. Over the years I've build many PC's, starting from Athlon XP's, up to newest Intel Haswell and AMD FX platforms, and I got my personal, general impression on all of them.

I was able to do a direct comparison of Sandy Bridge/Ivy Bridge/Haswell builds against various AMD builds, including Athlon II (dual and quad cores) and FX 8320. I did my comparisons using unregistered version of Stereo Tool 7.24.

My rough estimation would be that using ST, the real world single core performance of latest Intel generations is at least 40% higher as that of the latest FX series per clock. I found performance of the FX 8320 (125W) similar to mobile i5-3210m (Ivy Bridge, 35W) and low power i3-3220T (also 35W). Remember, that heat dissapation can become a big issue (especially in a closed, tight environments) and there is a huge difference between 125 and 35W. FX can get very, very hot in a second.

FX 83xx series is marketed as 8-core, but in fact it is not. They are 4 module constructions, where each module contains one full core and one lacking FPU unit. So in total, "8 core" AMD processors contain 8 ALU and only 4 FPU units. It's nowhere near 8 core circuit, is it? It's just a 4 core construction with very power hungry SMT implementation. That leads to mediocre performance per core, which is crucial for Stereo Tool (as it uses only 2 threads, and here FPUs are in use).

It is worth noting, that SMT implementation in AMD processors is also badly supported by Windows (regardless the version, but WIndows 7 has something wrong with the scheduler, I will discuss this in other thread). On linux it works better. I also found that even though on paper Athlon II x4 processors should do just fine, like Core 2 Quads, they simply do not deliver the power needed to process with Stereo Tool (Especially when you crank clippers up). I didn't test Phenom's though, they should do the trick (L3 cache might help).

When Intel is concerned, one thing should be discussed. Heat spreader of the newest Haswell and older Ivy Bridge generation is attached to the core by thermal material, which will wear off after certain period of time. Moreover people report that this thermal material in Haswell is just of low quality, possibly causing heat issues when fully loaded. In that case, I would recommend Sandy Bridge processors, as they are just a notch slower, but their heatspreader is soldered to cores permanently.

When you consider i7 over i5, I would say that hyperthreading would make difference, when using other software in parallel with ST, such as multiple encoders and/or servers for streaming.

Kind regards,
Slawomir

Dang I wish I would have looked here first. I just got a used AMD Phenom X4 9750 Quad-Core 2.4 GHz in a desktop with 8GB Ram plan on putting ESI Juli@ in it next week. If I had known, I would have went with core 2 duo. Hopefully it will run the program, as my pentium 4 3.0 ran it just fine.

In fact, for you (in my opinion) even an old LGA 775 E8400 dual core would be better for ST. According to passmark site, x4 9750 gets about 28350 points on average (4 threads), while E8400 gets roughly 2160 (per 2 threads). That gives quite a noticable difference of processing power per thread (in favor of Intel). It gets worse, when you compare cheap-as-dirt newer series, such as Celeron G1630 (LGA 1155), which matches the Phenom of yours in passmark (and this takes into account all threads the processor is able to process simultaneously).

Recently I have had 3 rigs, two with AMD processors (Athlon II x4 620, FX 8320) and one with Intel, i3-3220T. This little i3 matched FX 8320 effortlessly as far as ST is concerned. So did my mobile i5 3210m (2 cores, 4 threads) in laptop. Notice, these dual cores i3/i5 are also equivalent of Phenom II x4 945, which is quad. See my point here? The faster the single thread of your processor is, the better it will perform in ST. This is important, as Hans mentioned numerous times, ST uses only 2 threads, at least for now.

It doesn't mean that AMD processors are bad overall. They're not, but AMDs are better for something else, that doesn't rely on single threaded power (transcoding for example). The only really bad thing in latest AMD processors, is their power consumption and thus relatively high heat dissapation requirements.

My FX 8320 was a beast in that matter. Marked as TDP 125W processor, when fully loaded (and not overclocked) drew over 140 Watts from socket (yes, checked that twice using wattmeter between idle and loaded states) and needed a huge radiator for the long run (it's hard to dissapate that amount of heat, even though the case was well ventilated with 4 fans, 12 cm each). Using it only for ST would be very impractical, as it has 8 threads (ST can't benefit from that) and more importantly, if you want the things to work stable, you don't want them to get excessively hot, right? When running rig for 24/7 power consumption starts to matter as well.

I could talk about it forever, as all of the above matter a lot and thus for ST running device, when reliability is crucial, components should be picked up with care. BUT: If your hardware runs just fine and is sufficient - stick with it until it gets old enough to replace the whole platform (or hardware requirements increase significantly, then you'll have to upgrade anyway).

Regards,
Slawomir

I used to have an old Athlon 64X2 Machine. It would have been interesting to test ST on it.

I'll try a more modern, yet obsolete Athlon XII circa 2010 if I can though to see what happens.

Those Athlons II are still doing their job for most of the tasks, but as far ST is concerned, when you have all the bells and whistles on - it's a bit slow.

For comparison: On a high-end i7 (I've tested a 4770s) the CPU load for Stereo Tool with everything enabled (FM, including Composite Clipper and Stokkemask) is about 6-7%.
I haven't seen any recent AMD numbers so I can't compare, and also you need to keep in mind that the reported CPU load on Intels is generally skewed towards lower numbers because of Hyperthreading (if you use all cores - but without Hyperthreading - at 100%, you see a CPU load of 50%, but if you also fill the Hyperthreaded cores you only gain about 30% in performance, but the reported CPU load will double to 100%.)