Question What's the current best CPU for consumers on the market regardless of price?

[mujmuj]

i9-12900KS ? According to this link (Intel® Core™ i9 Processors ) , i9-12900KS has the best Max Turbo Frequency and the highest core numbers simultaneously.

But what does that "S" mean at the end? Last time I made the mistake of buying i9-12900KF, which later I learnt from this forum that "F" means it doesn't have the integrated graphics. So perhaps "S" also mean something I didn't expect?

And what about the other companies like AMD?

My only use is computation power for my big data (>30GB) statistical analysis.

 

[kanewolf]

i9-12900KS ? According to this link (Intel® Core™ i9 Processors ) , i9-12900KS has the best Max Turbo Frequency and the highest core numbers simultaneously.

But what does that "S" mean at the end? Last time I made the mistake of buying i9-12900KF, which later I learnt from this forum that "F" means it doesn't have the integrated graphics. So perhaps "S" also mean something I didn't expect?

And what about the other companies like AMD?

My only use is computation power for my big data (>30GB) statistical analysis.

A Xeon or Threadripper. More memory slots, more memory bandwidth than any desktop CPU. Which model depends on budget.
But it depends on the scalability of the software.

 

[Eximo]

Just a slightly higher binned processor, the boost profile is a little faster.

An argument can be made for the 5950X with its 16 true cores. Depends on the workload whether the single core IPC on the 8 cores plus the 8 efficiency cores is any better.

And in a few weeks time the 7950X is certainly going to be the top CPU with some competition from the 13900K when it comes out (they doubled the efficiency cores to 16)

 

[drea.drechsler]

The question is invalid without stating a use-case.

It's like asking 'what's the best car on the market'. Some would say a Tesla because it doesn't use gas at all...but some who need to drive a thousand miles in a week would find that a terribly limiting choice. Some would say Ford Excursion because you can haul a lot of passengers and their gear...but others only drive themselves.

Any processor like a 12900K is overkill if the most you do is play some light web games. A Threadripper system will spit out video renderings like nobody's business but it's equally overkill to read emails and shop Amazon...and is middlin' at best for playing games. They're both gonna sit in horribly expensive systems that may not provide anymore benefit than a system costing a lot less. What's 'best for consumers' is what offers the best value for money.

 

[Nighthawk117]

i9-12900KS ? According to this link (Intel® Core™ i9 Processors ) , i9-12900KS has the best Max Turbo Frequency and the highest core numbers simultaneously.

But what does that "S" mean at the end? Last time I made the mistake of buying i9-12900KF, which later I learnt from this forum that "F" means it doesn't have the integrated graphics. So perhaps "S" also mean something I didn't expect?

And what about the other companies like AMD?

My only use is computation power for my big data (>30GB) statistical analysis.

If you go by benchmarks then the 12900KS, but I prefer the 5950X. All cores are the same and it's much more power efficient. I wouldn't buy either though, AMD's 7000 CPU's will be available at the end of this month with Intel following the month after.

The 'S' means 'Special Edition', they are usually clocked slightly higher and higher binned chips. In the case of the i9 12900KS, 'K' means it has an unlocked multiplier and the 'S' means it's special.

 

[hotaru.hino]

My only use is computation power for my big data (>30GB) statistical analysis.

For this use case, a many-core processor would be better here. Which looking at PCPartpicker, the best one for the job would be a Threadripper 3990X, as it's a 64-core processor. Though if you don't mind the pre-built route, you can find Threadripper Pro workstations.

Basically for any sort of work on a large amount of data, and assuming it doesn't alter the data as it works on it, higher core counts is favorable here.

 

[wyliec2]

Determining the ‘Best CPU’ is entirely dependent on the task(s) to be performed.

For example I have a 5950X CPU (16 cores/32 threads) and 3080 Ti GPU which I mainly use for video encoding.

If I use H265 for 4K video encoding, it will use nearly all threads at >90% CPU utilization.
If I use H264 for 1080p encoding, all threads are not as fully utilized at 50%-60% CPU utilization.
If I use NVENC (GPU hardware encoding), the process is much faster and the CPU is utilized at 40%-60%. It should be noted that the GPU hardware encoding is dependent on the GPU generation (ie. Ampere) and not on the actual GPU graphics engine (ie. a 3070 will perform similarly to a 3080 Ti).

My primary purpose doesn’t make use of the 3080 Ti – it’s more to play with; a 3060 would have sufficed.

For a statistical analysis/modelling, again, the key will be scalability/parallelism to utilize multiple cores/threads. Assuming your data is all in RAM and all cores are in use, then an intensely iterative function will depend on the CPU frequency and IPC (instructions per cycle). RAM speed would likely factor in as well.

FWIW, since I deal with an application utilizing all cores, I’m a bit leery of the Intel combination of Performance and Efficiency cores versus the normal AMD core architecture.

 

[kanewolf]

For this use case, a many-core processor would be better here. Which looking at PCPartpicker, the best one for the job would be a Threadripper 3990X, as it's a 64-core processor. Though if you don't mind the pre-built route, you can find Threadripper Pro workstations.

Basically for any sort of work on a large amount of data, and assuming it doesn't alter the data as it works on it, higher core counts is favorable here.

Only if the software scales. If data reduction with Excel is a large part of the workflow, it doesn't scale much beyond 8 cores. We need more information about workflows and software suites used for the "30GB" data.

 

[hotaru.hino]

Only if the software scales. If data reduction with Excel is a large part of the workflow, it doesn't scale much beyond 8 cores. We need more information about workflows and software suites used for the "30GB" data.

True

But depending on how the data is stored or accessed, you could just open up another instance of Excel to get around this. Or whatever program you're using.

EDIT: As an example of this, I made a tool at my last job to parse through data logs and convert them into CSV files. I did experiment with multithreading to speed up the process, but in the end, since I had a mechanism to start and end at any point within the log file, it was faster to get this "feature" by simply launching multiple instances of this tool at various start and end points. The only downside was there multiple sets of CSV files, but those can be collated easily by hand if desired.

 

[Paulie walnuts1888]

w-3175x with a 1kw cooling system, possibly 10 of them depending how wealthy you are

 

[mujmuj]

The question is invalid without stating a use-case.

It's like asking 'what's the best car on the market'. Some would say a Tesla because it doesn't use gas at all...but some who need to drive a thousand miles in a week would find that a terribly limiting choice. Some would say Ford Excursion because you can haul a lot of passengers and their gear...but others only drive themselves.

Any processor like a 12900K is overkill if the most you do is play some light web games. A Threadripper system will spit out video renderings like nobody's business but it's equally overkill to read emails and shop Amazon...and is middlin' at best for playing games. They're both gonna sit in horribly expensive systems that may not provide anymore benefit than a system costing a lot less. What's 'best for consumers' is what offers the best value for money.

That's right. As I wrote in my post, "My only use is computation power for my big data (>30GB) statistical analysis. " I will not play games or anything else. It's just for my work, which is for big data statistical analysis. It loads all the big data onto RAM first, and then calculates stuffs.

 

[mujmuj]

For this use case, a many-core processor would be better here. Which looking at PCPartpicker, the best one for the job would be a Threadripper 3990X, as it's a 64-core processor. Though if you don't mind the pre-built route, you can find Threadripper Pro workstations.

Basically for any sort of work on a large amount of data, and assuming it doesn't alter the data as it works on it, higher core counts is favorable here.

Threadripper 3990X is amazing, but it only support DDR4, but not DDR5, correct? I have DDR5 RAM already, and my big data software loads all the data to RAM, so I guess RAM functionalities are important.

 

[mujmuj]

w-3175x with a 1kw cooling system, possibly 10 of them depending how wealthy you are

w-3175x only supports DDR4, but not DDR5, correct? I have DDR5 RAM already, and my big data software loads all the data to RAM, so I guess RAM functionalities are important.

 

[mujmuj]

Just a slightly higher binned processor, the boost profile is a little faster.

An argument can be made for the 5950X with its 16 true cores. Depends on the workload whether the single core IPC on the 8 cores plus the 8 efficiency cores is any better.

And in a few weeks time the 7950X is certainly going to be the top CPU with some competition from the 13900K when it comes out (they doubled the efficiency cores to 16)

A Xeon or Threadripper. More memory slots, more memory bandwidth than any desktop CPU. Which model depends on budget.
But it depends on the scalability of the software.

In short, is 7950X better than Threadripper 3990X on the clock speed and DDR5 support, but worse on the number of cores?


And in this link, does "32.00GB DDR5 SDRAM 2998 MHz" mean that a computer with 7950X CPU cannot support RAM space bigger than 32GB? I have two 32GB RAM so in total it's 64GB.

 

[kanewolf]

In short, is 7950X better than Threadripper 3990X on the clock speed, but worse on the number of cores?

Unknown. You are comparing an as-yet unreleased CPU. The best answer is "probably", but that question can't be answered for several weeks.

 

[mujmuj]

Just a slightly higher binned processor, the boost profile is a little faster.

An argument can be made for the 5950X with its 16 true cores. Depends on the workload whether the single core IPC on the 8 cores plus the 8 efficiency cores is any better.

And in a few weeks time the 7950X is certainly going to be the top CPU with some competition from the 13900K when it comes out (they doubled the efficiency cores to 16)

I see. It seems their release dates will be similar, and I have some time to wait for enthusiast-experts testing both products and deciding which one is better for which purpose.

"Intel's 13th Gen Core 'Raptor Lake' CPUs reportedly launch on September 28, with availability starting from October 10, according to recent leaks."

"The release date for the Ryzen 9 7950X is the 29th of September 2022."

 

[wyliec2]

That's right. As I wrote in my post, "My only use is computation power for my big data (>30GB) statistical analysis. " I will not play games or anything else. It's just for my work, which is for big data statistical analysis. It loads all the big data onto RAM first, and then calculates stuffs.

The key question I haven’t seen answered is what application you’re using and/or how many threads can it effectively use concurrently???

 

[geofelt]

What is that app that you use for statistical analysis?
Can it make use of all available processing threads?
If you can effectively use all available processing threads, the R9-5950X and I9-12900KS are comparable.
There are workstation systems that can use multiple cpu's that could do more.
You did not ask about that.

Soon, next gen processors from both Intel and AMD will launch.
Think of about a 15% added boost from either.
Many apps do not lend themselves to multi threading.
You may want to read about "amdahl's law"
https://en.wikipedia.org/wiki/Amdahl's_law

If you are running a system now, try an experiment to see how sensitive your work is to the presence of many threads.
For example, disable half of them and see if the processing time doubles.

Or, use windows power management to cut the max processor from 100% to 50% and see if your run time doubles.

 

[Paulie walnuts1888]

w-3175x only supports DDR4, but not DDR5, correct? I have DDR5 RAM already, and my big data software loads all the data to RAM, so I guess RAM functionalities are important.

it does, i guess threadripper 5995x with the chiller would be better

 

[mujmuj]

@wyliec2 @geofelt

It's Stata MP. It says it's "24 cores" version of Stata.

It says "You can purchase a Stata/MP license for up to the number of cores on your machine (maximum is 64). For example, if your machine has eight cores, you can purchase a Stata/MP license for eight cores, four cores, or two cores. "

 

[mujmuj]

it does, i guess threadripper 5995x with the chiller would be better

These two links seem to suggest only DDR4-2666 with w-3175x. You mean there is a way to upgrade this?


Intel Xeon W-3175X vs Intel Core i5-12600K Benchmark, comparison and differences (cpu-monkey.com)

Intel® Xeon® W-3175X Processor

 

[geofelt]

@wyliec2 @geofelt

It's Stata MP. It says it's "24 cores" version of Stata.

It says "You can purchase a Stata/MP license for up to the number of cores on your machine (maximum is 64). For example, if your machine has eight cores, you can purchase a Stata/MP license for eight cores, four cores, or two cores. "

The question remains as to how much faster a job will run.
Look at amdahls law:
https://en.wikipedia.org/wiki/Amdahl's_law

For your particular app, it might be better to consult the support forums for the app.

 

[mujmuj]

The question remains as to how much faster a job will run.
Look at amdahls law:
https://en.wikipedia.org/wiki/Amdahl's_law

For your particular app, it might be better to consult the support forums for the app.

This is what Stata website says about how fast it becomes as cores are increased. Basically, it can scale up greatly as cores are added.

If the Threadripper 3990X supported DDR5, I would have bought it right away.

And another lingering question is why many people are saying 7950X is now the best CPU when there is AMD threadripper, which has more cores.

Because 7950X supports DDR5 whereas Threadripper 3990X doesn't?

Or because people simply don't count too expensive things like Threadripper 3990X?

Advanced multiprocessing capabilities

Almost every computer can take advantage of the advanced multiprocessing capabilities of Stata/MP. Stata/MP provides the most extensive support for multicore computers and multiprocessor computers of any statistics and data-management package.

www.stata.com

"Speed is often most crucial when performing computationally intense estimation procedures. A few of Stata's estimation procedures, including linear regression, are nearly perfectly parallelized, meaning they run twice as fast on two cores, four times as fast on four cores, eight times as fast on eight cores, and so on. Some estimation commands can be parallelized more than others. Taken at the median, estimation commands run 1.8 times faster on 2 cores, 2.9 times faster on 4 cores, and 4.1 times faster on 8 cores.

Speed can also be important when managing large datasets. Adding new variables is nearly 100 percent parallelized, and sorting is 61 percent parallelized.

Some procedures are not parallelized and some are inherently sequential, meaning they run the same speed in Stata/MP."

 

[Nighthawk117]

And another lingering question is why many people are saying 7950X is now the best CPU when there is AMD threadripper, which has more cores.

It would depend on what your doing but I would have said the best consumer desktop CPU is the 7950X. It will give better performance than Threadripper for typical consumer desktop workloads because most don't scale with extremely high core count CPU's. If you want the most powerful though and your workload can scale to 64 cores then the Threadripper Pro 5995wx is the most powerful available.

 

[geofelt]

"Speed can also be important when managing large datasets. Adding new variables is nearly 100 percent parallelized, and sorting is 61 percent parallelized.

Some procedures are not parallelized and some are inherently sequential, meaning they run the same speed in Stata/MP."

This means that not all work can be run in parallel.
In particular sorting 39% can not run in parallel.

Early passmark ratings for the 7950x with 32 threads show an impressive
rating of 64572 when all threads are fully utilized.
The single thread performance rating is 4362 which is also very good.
You could do worse.
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+7950X&id=5031

By comparison, the I9-13900K also has 32 threads and a rating of 54433/4833
https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i9-13900K&id=5022

fwiw, threadripper 3970X with 64 threads is 63548/2686
https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+Threadripper+3970X&id=3623

As you can see, how well you can do is very dependent on the possibility of being able to fully use all threads vs. the need to handle the single thread component.

I might add that there is supposed to be a binned I9-13900KS that is even stronger.

How many threads do you plan to purchase for the Stata license?

I would not worry much today about DDR4 and DDR5 differences.
Yes, the speed differences are impressive, but the associated cas latencies are correspondingly high which negates most of the speed advantage.
Currently, only a specialized few apps show any significant difference in actual performance between DDR4 and DDR5.

Do you have a budget??