What parameters determine the single-threaded performance of a processor ?

[Amyrro]

Hello everyone

I am considering buying a new machine, and which I am going to use primarily for Engineering CAD. I searched about the software I already use, and turns out that most are single-threaded (Solidworks or CATIA, Ansys Fluent, & maybe Comsol + others).

I had a quick research on CPUs, and I have the option to get one of AMD FX-9590, AMD FX-8370, Intel i7-6700K, i7-6700 OR Intel i5-6600. Of course the i5 is cheaper than both, and as I have been through some threads on this forum, I found that many recommend the Intel i7s over the AMDs CPU.

http://www.tomshardware.com/answers/id-2801571/intel-47...

http://www.tomshardware.com/answers/id-2669858/intel-46...

http://www.tomshardware.com/answers/id-2691990/amd-9590...

http://www.tomshardware.com/answers/id-2196981/intel-49...

Q.1
I found some people online recommending i5 processors over i7 processors for gaming, especially games which do not need many cores, when the difference in the clock speed is minimal (as in the case i7-6700k with 4.2 GHz or i7-6700 with 4.0 GHz, compared to i5-6600 with 3.9 GHz), and does not worth the difference in price (more than 200$).

From what I know, single-threaded performance relies heavily on the clock speed, which makes some i5s in many cases better than i7s because of the difference in the clock speed, or in some cases when the clock speeds is comparable or identical, the i5s become a more economic option.

Does an i7 processor process data better, or more efficiently than an i5 processor of the same generation, architecture and clock speed in a single-threaded task ? Or is the single-threaded performance related to the processor itself with its specification rather than its class?

In this sense, does this mean that the primary difference between i5 and i7 processors is the number of cores and the hyper-threading, plus the cache memory associated with class of processors (i7s tend to carry more cache memory than i5s) ??


Q.2
From the benchmarks shown below in the links, I noticed that although i5-6600K has a clock speed of 3.5 GHz, it is slightly lower in single-threaded performance than the i7-6700 which has a clock speed of 3.4 GHz, which implies that there are some other contributing factors.
http://www.cpubenchmark.net/singleThread.html
http://www.cpubenchmark.net/high_end_cpus.html

Also, I noticed that although the clock speeds of the AMD processors FX-9590 and FX-8370 are way higher than the Intel CPUs, the single threaded performance is lower than the Intel CPUs. Does this imply that the single threaded performance depends on the different parameters of the CPU, with the clock speed and instruction per clock being the main one , excluding the number of cores and threads ?


Q.3
I got a little bit confused on the parameter that I have to consider when looking at different CPUs and look for single-threaded performance. What would define the single threaded performance CPUs ? How can I estimate the single threaded performance of CPUs ? Especially when comparing i5 to i7 CPUs , or Intels to AMDs.

Other than the clock speed and instruction per clock, doesn't the lithography and cache memory have an effect on the single-threaded performance of a CPU ?


Regards and thanks in advance

 

[Robert Cook]

Q1
Pretty much an i5 is an i7 with no hyper threads. the per core performance not counting hyper threads are equal at a given clock speed. they are the same core designs per generation, and it is simply whether or not you get hyper threads. Also the cache is lower on the i5 chips, but that is really a deal buster. On a single threaded application they should perform equally if they are both at the same clock speed.
Q2
I would view those score differences as next to nothing.
As of right now AMD sadly does not have very powerful chips in the case that they might run fast but they need that extra speed top perform near equal. in other words as of now a single AMD core is not as powerful.
Q3
Pretty much just know that Intel CPUs as of right now ruin AMD at single threaded tasks and in raw power on core to core basis. Lithography dikes play a role, as an older architecture implies older technology and thus less performance. Note that the core two quad can run as fast as a modern i5, but the i5 almost doubles it.
Cache does not always have much if any of an effect. Unless a task is known to need lots of cache, anything above 4Mb generally is enough.

 

[k1114]

Q1, i7 and i5 is merely marketing branding. A cpu is just going to have more cache with more cores or threads to more efficiently feed info to all of them at once.

Q2, the reason the i7 is a bit higher is because you didn't take into account the turbo speed, the i7 is faster 4ghz vs 3.9ghz so scores a tiny bit higher.

Q3, the main factor is the architecture. With each gen they usually increase ipc. Other than that, you look at speed and don't forget turbo. Other specs are so minimal it doesn't matter. Amd hasn't released anything new for awhile. You should only be looking at skylake cpus.

 

[Robert Cook]

^ glad you mentioned turbo, I forgot ti think of that.

 

[Amyrro]

Q1
Pretty much an i5 is an i7 with no hyper threads. the per core performance not counting hyper threads are equal at a given clock speed. they are the same core designs per generation, and it is simply whether or not you get hyper threads. Also the cache is lower on the i5 chips, but that is really a deal buster. On a single threaded application they should perform equally if they are both at the same clock speed.
Q2
I would view those score differences as next to nothing.
As of right now AMD sadly does not have very powerful chips in the case that they might run fast but they need that extra speed top perform near equal. in other words as of now a single AMD core is not as powerful.
Q3
Pretty much just know that Intel CPUs as of right now ruin AMD at single threaded tasks and in raw power on core to core basis. Lithography dikes play a role, as an older architecture implies older technology and thus less performance. Note that the core two quad can run as fast as a modern i5, but the i5 almost doubles it.
Cache does not always have much if any of an effect. Unless a task is known to need lots of cache, anything above 4Mb generally is enough.

Thank you guys for the replies. For some reason I am not able to quote K1114's comment or reply to it. Anyway!

Q.4
What Mr. Cook mentioned is interesting.
Does this also apply to i3 CPUs also ? Would a i3 processor perform like an i7 at a given clock speed and for the same generation and architecture under a single-threaded task ?

Q.5
What could be possible reasons behind the lag of AMD in terms of single threaded processing in comparison to Intel. The lag of AMD becomes clear when considering the superior number of cores and clock speeds which AMD offers on its processors, while it still lags behind Intel in terms of performance, which reflects the lag in the individual core performance.

Q.6
What k1114 mentioned on the increase of cache with the increase of the number of cores is also interesting. I thought about upgrading the processor on a laptop I got from Intel Core 2 Duo P7350 (2.0 GHz, 3 MB Cache) to either P8800 (2.66 GHz, 3 MB) or P9500 (2.53 GHz, 6 MB) from the same family (same socket compatibility and same wattage rating). The link below shows a comparison between the three.
http://ark.intel.com/compare/36750,40380,35566

The overall performance benchmark of P9500 is slightly higher than the P8800, and if not I am not mistaken it could due to the cache memory difference.
http://www.cpubenchmark.net/midlow_range_cpus.html

In contrast, the difference single-threaded performance is negligible, but for the sake of comparison, we can see that the P8800 has slightly higher score because of its clock speed.
http://www.cpubenchmark.net/singleThread.html

Does this mean that the increase in cache memory can lead to improvement in overall performance, as data is fed more efficiently to the CPU cores (both have the same number of cores) ??

Q.7

I would like to ask about virtual machines and 3D design, especially rendering. Are these two processes considered to be single-threaded or multi-threaded processes ?

Thanks again

 

[k1114]

Q4/6 are about the same thing. Cache makes little difference and yes even a pentium will compare to an i7. Don't get too caught up in the marketing names. You don't see many benchmarks because it's been proven over and over. They still test gen vs gen often but never the same gen vs the same gen. Older article but this is proof of same gen, they clock them the same and compare. The increased cache on the i7 is hardly anything vs the i5 and architecture is what matters. http://www.tomshardware.com/reviews/processor-architecture-benchmark,2974-14.html It then comes down to cost, is 0.1% performance worth $100 just for a larger cache with i5 vs i7. A lot of R&D goes into cpus and more cache is going to make the cpu cost more. Look at the cpu comparison you linked to. See the price increase for the larger cache? They have to balance cost vs performance.

Q5, Going a bit back in older gens, amd has just gone with more cores vs ht and bumped up their clocks to stay comparable in single threaded. Back with phenom II x6 to compare with a quad core with ht. The fx cpus came out when sandy bridge came out, 2nd gen i series, 4 years ago. They were comparable then too but look at the news and amd is losing lots of money. Go farther back in the news and you see intel is the cause and all the antitrust lawsuits. All they get is a slap on the wrist and a few million dollar fine but that's nothing compared to the market share they took and the profits they made from it.

Amd stays in there by dropping prices, you get an 8 core fx the same price as an i3. Amd has been releasing apus every year in the meantime but nothing to compare or replace the higher end fx. They will finally release something new with zen coming out this year which will catch up some; supposedly 40% ipc increase.

Q7, While those cad software are mostly singlethreaded which may just need a lower end cpu, you have other workload considerations that will make you go for a higher thread count anyways. You won't find much of anything that isn't more multithreaded than rendering. But the biggest factor is the software being used. While modeling is mostly single threaded, this is largely gpu work with with viewport performance if you are doing 3d. Rendering can just easily be separated so takes advantage of parallelism very well. If you are going into the higher end, you have to look for how many threads it can use, but that shouldn't be a concern unless you are going multi cpu. Then there's gpu rendering where you get 1000s of "cores" and upwards of 10x faster than cpu rendering. The software you mentioned don't come with gpu renderers unless you get a third party renderer. As for virtual machines, the work is being divided up between the environments and if you have multiple at the same time which usually does happen, that is literally multitasking so it all lends itself to multithreading.

 

[Robert Cook]

^