Question Ryzen 5900X vs 5950X

[shadowafctor]

I've built computers for a long time but since my college days I've only had a laptop. I started a small business and need a computer for a sole purpose.

I will be running multiple threaded python scripts which use a selenium web driver to launch roughly 30-60 Chrome profiles at once and complete a few web based tasks. It uses screen scrapped info to process some things.

It will launch 30-60 independent sessions of the website concurrently and load through 4-5 web pages. From start to finish it takes about 1 min 30 seconds to complete each session. Once the 30-60 sessions are complete it would close the windows (profiles) and launch another 30-60 until it works its way through a configuration file.

My max budget is about $6000 on this computer but i've read that AMD has a serious advantage over intel now. I realize that these new Ryzen chips are hard to come by and I will likely need to purchase one off ebay at a considerable markup. Don't love the idea but this computer is critical to my business.



Is the 5950X worth getting over the 5900X for my purpose ? Any other thoughts and suggestions would be great. I plan on buying other high end components for the computer as well, but to get an idea of what I need to pick out. I need to start with a processor



The computer will need to run 24/7 or close too, but will be idle during alot of that time. I spend a lot of time traveling for my day job so I need to be able to remote into the computer at my house.



This is a friends setup. He is launching 20 profiles at once with a Ryzen 9 3900X, 64 GB Ram and a 1 TB SSD.

First picture is at idle and second is at load ( 20 profiles)





Thanks!

 

[OC_Nemas_OC]

I've built computers for a long time but since my college days I've only had a laptop. I started a small business and need a computer for a sole purpose.

I will be running multiple threaded python scripts which use a selenium web driver to launch roughly 30-60 Chrome profiles at once and complete a few web based tasks. It uses screen scrapped info to process some things.

It will launch 30-60 independent sessions of the website concurrently and load through 4-5 web pages. From start to finish it takes about 1 min 30 seconds to complete each session. Once the 30-60 sessions are complete it would close the windows (profiles) and launch another 30-60 until it works its way through a configuration file.

My max budget is about $6000 on this computer but i've read that AMD has a serious advantage over intel now. I realize that these new Ryzen chips are hard to come by and I will likely need to purchase one off ebay at a considerable markup. Don't love the idea but this computer is critical to my business.



Is the 5950X worth getting over the 5900X for my purpose ? Any other thoughts and suggestions would be great. I plan on buying other high end components for the computer as well, but to get an idea of what I need to pick out. I need to start with a processor



The computer will need to run 24/7 or close too, but will be idle during alot of that time. I spend a lot of time traveling for my day job so I need to be able to remote into the computer at my house.



This is a friends setup. He is launching 20 profiles at once with a Ryzen 9 3900X, 64 GB Ram and a 1 TB SSD.

First picture is at idle and second is at load ( 20 profiles)





Thanks!

Hi
The extra cores go a long way especially if you are planning to run multiple scripts and for business use then you can get a good computer for the money. Both chips will do the job nicely, but if the price is right I recommend the 5950x, but if you can wait dont pay the scalpers on Ebay else it just gives them credit. Have fun and good luck

 

[Sable]

Sounds like you're setting up a sniping bot...

 

[VirtualLarry]

Sounds like you're setting up a sniping bot...

Or his own version of "Cambridge Analytica".

 

[MrTeal]

Sounds like you're setting up a sniping bot...

Use 3900X to acquire 5950X. Use 5950X to acquire 3080.
...
Profit

 

[shady28]

Python + Zen 3 yes.

Doesn't appear to scale well past 12C/24T though. I also checked on threadripper benches, same thing.

Ditto for Selenium.

You would probably be better off getting two 3900X machines and running two sets of scripts based on what I just looked at. Otherwise a 5900X looks to perform almost the same as 5950X on this stuff.

 

[damian101]

Python + Zen 3 yes.

Doesn't appear to scale well past 12C/24T though. I also checked on threadripper benches, same thing.

Ditto for Selenium.

You would probably be better off getting two 3900X machines and running two sets of scripts based on what I just looked at. Otherwise a 5900X looks to perform almost the same as 5950X on this stuff.


View attachment 34973

I don't think that's representative of @shadowafctor's workload.

 

[shady28]

I don't think that's representative of @shadowafctor's workload.

I'm not sure about Python, but the Selenium part is probably relevant. Selenium is used a lot for stability and load testing of websites, along with bots.

 

[Cogman]

Honestly, I'd bet that network is a bigger issue than CPU power.

I doubt you need a 5xxx CPU.

You might consider getting a server CPU instead of a consumer CPU.

For example

Are you a human?

www.newegg.com

 

[Carfax83]

Personally, I would stay away from the 5950x as it seems to be severely bandwidth limited. Dual channel DDR4 memory is already being pushed past the limit to keep those cores fed, which seems to affect the scaling significantly.

DDR5 can't come fast enough!

 

[GoodRevrnd]

Sounds like you're setting up a sniping bot...

Hahaha was going to say same thing. "Which cpu works the best to snipe more CPUs to scalp?"

 

[Justinus]

Personally, I would stay away from the 5950x as it seems to be severely bandwidth limited. Dual channel DDR4 memory is already being pushed past the limit to keep those cores fed, which seems to affect the scaling significantly.

DDR5 can't come fast enough!

Can you back that up with a source? Maybe I missed the trend showing this when looking at reviews?

 

[cytg111]

Damn man, what are you doing?
If you are not a actually io bound, and it sounds like you are, the extra cores will help BUT if your bottleneck is the actual Python scripts, get the bleep off Python.
Idle most of the time... Is latency important to you? Actual execution time?
If your actual network io is not the bottleneck I would ditch python and chrome../

 

[Carfax83]

Can you back that up with a source? Maybe I missed the trend showing this when looking at reviews?

It was pretty obvious from the reviews. Zen 3 had a very large single thread performance gain when compared to Zen 2, but when you looked at multithreaded performance, the gain was nowhere near as high. The large L3 cache helps a lot, but it's not a miracle worker.

DDR4 is nearing the end of its lifespan, and AMD has been pushing core counts on dual channel mainstream CPU parts way more aggressively than Intel so this shouldn't really be a surprise. Ideally, the 5900x and 5950x would have a quad channel memory interface at the very least.

DDR5 is going to be a game changer for high core count CPUs on desktop.

From the Anandtech review:

In multithreaded SPECrate, the absolute gain was only around 10% or so, given that faster cores also require more bandwidth to main memory, which hasn’t been provided in this generation. This means that there are some bottlenecks to which a higher IPC won’t help if more cores require the same resources.

For real-world tests, across our whole suite, we saw an average +24% uplift. For explicitly multithreaded tests, we saw ranges from even performance up to +35%, while for explicitly single threaded tests, this ranged from even performance up to +57%. This comes down to execution/compute bound tests getting bigger speedups over memory bound workloads.

 

[Justinus]

It was pretty obvious from the reviews. Zen 3 had a very large single thread performance gain when compared to Zen 2, but when you looked at multithreaded performance, the gain was nowhere near as high. The large L3 cache helps a lot, but it's not a miracle worker.

DDR4 is nearing the end of its lifespan, and AMD has been pushing core counts on dual channel mainstream CPU parts way more aggressively than Intel so this shouldn't really be a surprise. Ideally, the 5900x and 5950x would have a quad channel memory interface at the very least.

DDR5 is going to be a game changer for high core count CPUs on desktop.

From the Anandtech review:

Oh, yes. That.

It seems to only affect specific, bandwidth intensive workloads. It's far from "Avoid the 5950x it's bottlenecked badly". More like know your workload and choose between 5950x or Threadripper appropriately.

 

[moinmoin]

For the 12 and 16 cores chips running all core workloads PPT, TDC, and EDC limits are also a concern. I wouldn't be so sure the lack of scaling of the IPC improvement from ST to MT is all or even majorly down to an assumed lack of memory bandwidth.

 

[cytg111]

DDR5 is going to be a game changer for high core count CPUs on desktop.

That would imply that SMT scaled better on zen3 6 and 8 cores, right?

 

[Carfax83]

Oh, yes. That.

It seems to only affect specific, bandwidth intensive workloads. It's far from "Avoid the 5950x it's bottlenecked badly". More like know your workload and choose between 5950x or Threadripper appropriately.

Specific bandwidth intensive workloads like compression, encoding, rendering etcetera..... Now whether these workloads are important to you, that's another matter. Most end consumers don't really run these kinds of workloads on a regular basis, but many power users do; regardless, that doesn't negate the fact that the scaling going from the 5800x to the 5900x and then on to the 5950x is often much less than it should be for aforementioned reasons.

I did the math (someone can check for errors if I am wrong) and for a 5950x, each core is getting around 3.5GB/s of bandwidth with overclocked DDR4-3600 which is paltry when you think about it. That's why despite the huge advantage in IPC gain, Zen 3 can sometimes lose to Skylake as the latter has a monolithic architecture and a memory controller which is embedded in the same die as the CPU cores which increases memory bandwidth efficiency and lowers memory latency. As I said before, Zen 3 makes up for much of that lack of efficiency by having a huge L3 cache, but many workloads will be too big to run from cache and will spill over into system memory.

With DDR5 4800 (the base speed), a 5950x would have 4.8GB/s of bandwidth per core. The highest projected speed for DDR5 is DDR5-8400, which would give each core on a 5950x 8.4GB/s, a 140% increase over DDR4-3600.

 

[Carfax83]

For the 12 and 16 cores chips running all core workloads PPT, TDC, and EDC limits are also a concern. I wouldn't be so sure the lack of scaling of the IPC improvement from ST to MT is all or even majorly down to an assumed lack of memory bandwidth.

It might not be, but it's definitely a big factor. The chiplet and CCX architecture doesn't help either.

 

[Carfax83]

That would imply that SMT scaled better on zen3 6 and 8 cores, right?

Yeah it would. When I was looking at the reviewers, I noticed that the performance jump from 6 core to 8 cores was often higher than for 8 to 12 core. Case in point:

x265.png (500×930) (tpucdn.com)

5800x is about 25% faster than the 5600x in this benchmark for 33% more cores, while the 5900x is 15.7% faster than the 5800x despite having 50% more cores than the latter.

I'm sure there is also inter-ccx latency and bandwidth penalties at play here as well.

 

[Justinus]

Yeah it would. When I was looking at the reviewers, I noticed that the performance jump from 6 core to 8 cores was often higher than for 8 to 12 core. Case in point:

x265.png (500×930) (tpucdn.com)

5800x is about 25% faster than the 5600x in this benchmark for 33% more cores, while the 5900x is 15.7% faster than the 5800x despite having 50% more cores than the latter.

I'm sure there is also inter-ccx latency and bandwidth penalties at play here as well.

We already know encoding doesn't scale linearly with core count regardless of memory bandwidth.

Way to cherry pick a benchmark and take it way out of context.

 

[Carfax83]

We already know encoding doesn't scale linearly with core count regardless of memory bandwidth.

Way to cherry pick a benchmark and take it way out of context.

If I'm cherry picking, then you're issuing blanket statements. Core count scaling when it comes to encoding depends on the codec being used. If I had used an x264 benchmark, then maybe you'd have a point as x264 being much older has a much lower thread cap than x265 and doesn't scale nearly as well. The benchmark above used x265, which was designed to scale well with high core counts; way more than the 16 cores of the 5950x.

Testing x265 encoder scaling on a 128 core Azure VM for 4K HDR | by Kay Singh | Medium

From the test above, it seems that x265 has linear scaling up to 32 cores, then past that it's sublinear.

 

[Justinus]

If I'm cherry picking, then you're issuing blanket statements. Core count scaling when it comes to encoding depends on the codec being used. If I had used an x264 benchmark, then maybe you'd have a point as x264 being much older has a much lower thread cap than x265 and doesn't scale nearly as well. The benchmark above used x265, which was designed to scale well with high core counts; way more than the 16 cores of the 5950x.

Testing x265 encoder scaling on a 128 core Azure VM for 4K HDR | by Kay Singh | Medium

From the test above, it seems that x265 has linear scaling up to 32 cores, then past that it's sublinear.

You can see in our own "Benchmark your CPU with Handbrake and h.265" thread it doesn't scale well from 16-24-32-64 cores.



If it scaled well and was bandwidth bottlenecked as you claim, shouldn't it have scaled better from a 3950x to a 3960x?

 

[lightmanek]

For the 12 and 16 cores chips running all core workloads PPT, TDC, and EDC limits are also a concern. I wouldn't be so sure the lack of scaling of the IPC improvement from ST to MT is all or even majorly down to an assumed lack of memory bandwidth.

This!
From my own test's is really mainly down to power and CPU clock limits. If I lock my cores to fixed freq. then scaling looks almost identical to my old 3900XT in vast majority of tests I ran.

 

[Carfax83]

If it scaled well and was bandwidth bottlenecked as you claim, shouldn't it have scaled better from a 3950x to a 3960x?

It likely comes down mostly to configuration, in combination with Zen's low per core memory bandwidth and inter CCX bandwidth + latency penalities. Handbrake is for casual encoders and uses a bunch of preset settings to achieve a desired outcome based on quality and speed. Using a lot of cores can increase speed, but reduce quality. Encoding/transcoding is complicated, so there are a plethora of settings which can affect scalability and quality.

This is actually why Intel developed their SVT line of codecs, because it allows massive parallelization without affecting quality, so the SVT codecs are eventually going to replace x265 as they continue to get better.

But if you look at the raw x265 codec, it can certainly scale to a large amount of cores as shown by that in depth review I linked to. If it couldn't scale well beyond 16 cores, then it would be an absolute failure. x265 was used to achieve real time 4K60 fps 10 bit HDR encoding several years ago by MulticoreWare on a single rack dual socket Xeon server. If it couldn't scale well beyond 16 cores, it would have never been able to do that.

Source