Between AMD Ryzen 9 3900X and Intel Core i9-9900KS, what's the fastest processor for SINGLE THREADED performance?

[alpha754293]

If you had to pick between the AMD Ryzen 9 3900X or the Intel Core i9-9900KS for SINGLE THREADED performance, which one would you pick?

 

I've looked at the benchmarks and the reviews/results from various sites (Anandtech, Tom's Hardware, Gamer's Nexus, etc.) and it looks like that they're almost neck-and-neck relative to each other.

 

Ignore everything else. (Assume that it's the same amount of RAM, same speed of RAM, and I'm just going to have a low end PCIe 2.0 graphics card in the system, so the graphics card is going to be irrelevant. I have no intention of using the IGPU on the Intel Core i9-9900KS in any way, shape, or form.)

 

I need the FASTEST single threaded computational intensitve workload performance. (There are reasons for that. Mathematical reasons.)

 

Also assume that cost is not a factor.

 

What's the opinion of the community?

 

Your input is greatly appreciated.

 

Thank you.

[Mnky313]

5 minutes ago, alpha754293 said:

If you had to pick between the AMD Ryzen 9 3900X or the Intel Core i9-9900KS for SINGLE THREADED performance, which one would you pick?

 

I've looked at the benchmarks and the reviews/results from various sites (Anandtech, Tom's Hardware, Gamer's Nexus, etc.) and it looks like that they're almost neck-and-neck relative to each other.

 

Ignore everything else. (Assume that it's the same amount of RAM, same speed of RAM, and I'm just going to have a low end PCIe 2.0 graphics card in the system, so the graphics card is going to be irrelevant. I have no intention of using the IGPU on the Intel Core i9-9900KS in any way, shape, or form.)

 

I need the FASTEST single threaded computational intensitve workload performance. (There are reasons for that. Mathematical reasons.)

 

Also assume that cost is not a factor.

 

What's the opinion of the community?

 

Your input is greatly appreciated.

 

Thank you.

While the 3900x is the faster multi-threaded chip and I personally would chose it,
the 9900k(s) does edge it out slightly in single threaded workloads.
idk how your workloads split cores but the edge really isn't that much and hypothetically if you can run 50% more instances (50% more threads on the 3900x) it would be faster because running 8 things at 4 seconds per thing is slower than 12 things at 3.8 seconds per thing.)
(i'm not an expert in anything tho, I have 0 idea how whatever your doing scales.)

[Fasauceome]

Current gen Intel has a very slight single-threaded performance advantage right now

[Flying Sausages]

9900ks wins in single core but 3900x wins in multi core. I choose 3900x because I don't like Intel 14nm cursed. 

[Janne]

You don't need the opinion of the community. Just looking at benchmarks you can see intel's single core performance is faster

[Mathieu9836]

Short answer: Depends on application

Long answer: Depends on application, AMD is faster in some benchmarks, Intel is faster in others. But they are really neck to neck, no one is making THE chip in single threaded application.

[Harry Voyager]

So I just bench marked my R7 3800X at about 3070 Passmark single threaded performance one the golden core. The other cores were at around 2940ish. A 5.1Ghz Intel gets 3090, and a 5.2 chip gets 3190. A 4.4Ghz Zen chip can get 3130.

 

I think right now we are in the margin of error, but I ended up going Zen 2 because I expect it to get better as it gets sorted out, and I expect there to be even better AM4 chips before the socket hits end of life. 

 

I have also read that with an improved power control system, the 3900X and other dual chip cpus are probably going to get a bigger bump in CPU speed, since it should have two chiplettes to bounce the load between. 

 

Make of that what you will. 

[alpha754293]

Thank you everybody for your input.

 

It was interesting to read people's perspectives on this topic.

 

As I mentioned, yes I know that Intel slightly edges out, but I what I don't know is if you were to test it over and over again, would Ryzen 9 3900X trade places with Intel or if Intel CONSISTENTLY was able to edge out over the same.

 

To answer some of the questions about the workload, it's simple and stictly single process and to the best of my knowledge (I mean, I haven't profiled it), doesn't use much in the way of multi-threading, if any at all.

 

It's computational finance, so it's strictly single process, and mostly single threaded/lightly threaded.

[Falkentyne]

Under what test criteria here?
No one even mentioned this.

Both chips running at the exact same clock speeds?
(even though you can't compare different manufacturer's clock speeds directly)

Or both running at their max stable frequencies?

 

Plus there's the trick of disabling hyperthreading or even disabling cores to push clocks higher for even more performance....

[1van]

This topic is so lame There are tons of tests out there, performed by trusted reviewers, and just ordinary users.

It's fun to see people replying about multi-threaded performance when the topic explicitly states that the single-threaded performance is of interest.

11 minutes ago, alpha754293 said:

It's computational finance, so it's strictly single process, and mostly single threaded/lightly threaded.

Btw, you should contact your software manufacturer to see what CPU they optimize for, programmers for the computational finance usually know this. I'm going to assume it will be Intel.

[Mr. Smiley]

3 minutes ago, 1van said:

This topic is so lame There are tons of tests out there, performed by trusted reviewers, and just ordinary users.

It's fun to see people replying about multi-threaded performance when the topic explicitly states that the single-threaded performance is of interest.

Btw, you should contact your software manufacturer to see what CPU they optimize for, programmers for the computational finance usually know this. I'm going to assume it will be Intel.

Bouncing off this you should also see if there is anyway to run multiple instances of what ever script/program you are running. The single core performance between the two isn't very much. But if you can run multiple instances of your script/program at once the 3900x is going to edge out due to having additional cores meaning more instances of your script/program running at one time. 

[1van]

2 minutes ago, Mr. Smiley said:

 

I'd still recommend to contact the software manufacturer to find out for sure. It's highly unlikely that they're going to hide this information from a customer.

[Harry Voyager]

Another thing to bear in mind here, if you are only running a single thread, you should be able to get the same or similar single thread Zen 2 performance from the 3900X, 3800X and 3600X, as the differentiator between those is total core count rather than core clock speed.

 

At this point I actually suspect that the 3800X has the better single thread speed than the 3900X does. 

 

But I will also echo the advice to contact the software developer and ask them directly. This performance assumes equal levels of software optimization for both parts, however since the Sky Lake architecture has been around far longer than Zen, we can expect better software optimization for it, and hence higher performance for it, where optimization has been made. 

[alpha754293]

8 hours ago, 1van said:

This topic is so lame There are tons of tests out there, performed by trusted reviewers, and just ordinary users.

It's fun to see people replying about multi-threaded performance when the topic explicitly states that the single-threaded performance is of interest.

Btw, you should contact your software manufacturer to see what CPU they optimize for, programmers for the computational finance usually know this. I'm going to assume it will be Intel.

Well..the software itself is either a general linear algebra solver or equilvalent.

I haven't decided yet if I am going to try and port some of the existing stuff from a proprietary/closed source to using open source linear algebra solver to do the same.

But if I go open source, then I can, in theory, recompile it so that it would maximize the benefits of teach target unto itself, respectively, and the "thing" that counts is run time at the end.

 

 

8 hours ago, Mr. Smiley said:

Bouncing off this you should also see if there is anyway to run multiple instances of what ever script/program you are running. The single core performance between the two isn't very much. But if you can run multiple instances of your script/program at once the 3900x is going to edge out due to having additional cores meaning more instances of your script/program running at one time. 

 

If I were running different or multiple models, maybe/probably. But I'm not going to be. Or if I do, it won't be for quite some time until I am in a state to perform a Monte Carlo simulation on this fundamental, basic building block of an analysis or analysis type before expanding it. But yet, I would think that if I spawn a Monte Carlo, then the Ryzen 3900X, by virtue of it having just more cores, would benefit from it. But at that point in time, if I got to that point, I can just use my existing HPC cluster, which runs significantly slower on single threaded applications, but having so many more of them, will ultimately run out the Monte Carlo faster. But this is also why I wanted to narrow people's focus and scope because yes, there are a lot of other things (and as people's questions have asked) that can have an impact on it, but I wanted people to specifically ignore all of that and only narrowly focus on what little information was initially given.

 

Again, I purposely designed the question this way because yes, there are a lot of reviewers that has tested this, but what wasn't certain was whether AMD and Intel were just trading punches at who's #1 or if there was still a statistically significant result where one edged out over the other, despite the fact that the difference is probably within the noise space/margins of error.

 

Thanks.

[1van]

4 hours ago, alpha754293 said:

I haven't decided yet if I am going to try and port some of the existing stuff from a proprietary/closed source to using open source linear algebra solver to do the same.

...

Again, I purposely designed the question this way because yes, there are a lot of reviewers that has tested this, but what wasn't certain was whether AMD and Intel were just trading punches at who's #1 or if there was still a statistically significant result where one edged out over the other, despite the fact that the difference is probably within the noise space/margins of error.

Do you pretend to be a software developer who does not know what CPU microarchitecture his software is optimized for? It's starting to sound like trolling to me.

I know only one computational finance software guy, and he is well aware of which CPU is better for his tasks.

 

I would speculate that there are very few people today, who know how to optimize for Zen 2. If I were you, I'd pick Intel, or I'd built 2 systems (Zen 2 and Core, does not have to be 3900X and 9900KS) to test it in your specific application, then choose. A single-threaded performance in gaming may or may not correlate with that in your software.

[BruceWayneofLosSanto]

For single threaded the 9900KS is easily the fastest(are there any type of single thread loads that the ryzen is faster?).

[alpha754293]

10 hours ago, 1van said:

Do you pretend to be a software developer who does not know what CPU microarchitecture his software is optimized for? It's starting to sound like trolling to me.

I know only one computational finance software guy, and he is well aware of which CPU is better for his tasks.

 

I would speculate that there are very few people today, who know how to optimize for Zen 2. If I were you, I'd pick Intel, or I'd built 2 systems (Zen 2 and Core, does not have to be 3900X and 9900KS) to test it in your specific application, then choose. A single-threaded performance in gaming may or may not correlate with that in your software.

That isn't what I am saying at all.

 

What I am literally saying is that I know that they know (or that they'll give some generic BS response along the lines of "it's optimized for both/neither platforms" in order to avoid being sued) -- what I am literally saying is ignore it.

 

This statement is further cemented by the fact that if I port the existing code/software platform over from proprietary/closed source and switch over to using open source software to do the same computations, then I can recompile the open source software with the architecture specific target flags at compile time in order to optimise it to whatever architecture I end up going with. In other words, by downloading the raw source and compiling said open source software myself, it would render the statement/question irrelevant, which is what I was urging/guiding people to do in the first place.

 

re: "single thread gaming"

I've actually been using the Cinebench single thread results as the comparator so far, but again, with it being so close, again, I don't know if that difference is within the noise space where they're trading punches with each other, or if there is a statistically significant (e.g. you run it 100 times, and > 50% of the time, Intel will beat out AMD on single threaded performance, with that application.)

 

This is where asking the community comes into play. Yes there are people who have reviewed it, but what I am trying to ascertain is whether AMD and Intel is just trading punches with each other for the top in single-threaded performance, or whether there is stastically significance in the results (e.g. Intel beats AMD in single threaded applications more times than not).

 

To be even more explicit, if I download the source code for GNU Octave, then I can compile it for whatever target.

 

I've thought about buying both systems from like CyberPower or something like that (so that it's just pre-built) and testing it myself, but I want that to be a last-resort option if I can do the research online, via the community, and collect enough data to already help answer that question, knowing that people aren't running my application-specific workload.