i9 7900X overclocked

[Magnum_design]

I?m about to buy a new workstation and looking at the I9 7900X with 10 cores at 3,3 Ghz. And it?s possible to overklock to 4.0 easily It?s pretty tempting I?have no experience with overklocking at all though. I?s this possible for stable use ?

[Alex_M]

What is the reason to pick the 7900X when for the same money and even less you can get the 16 core/32 threads AMD Threadripper 1950X which is ~40% faster than 7900X in multi-threaded workloads? Except for fluid simulations (Phoenix FD) where calculations are always done on only 1-2 cores, Threadripper 1950X will always be faster by like ~40%.

1950X does ~3000 CB (~3300 with 3.9 Ghz overclock) in Cinebench, while 7900X does ~2200 CB (~2500 CB with 4.5 Ghz overclock).

[Magnum_design]

Because I?ve read in a test that Threadripper pretty much is overclocked to the top already, and the I9 had much more to clock. But I?ll gladly have a look at some tests if you could post a link

[Alex_M]

Are you sure the article was about 7900X and not 7980XE for example (the 18-core counterpart)? Because in all rendering tests I've seen be it Vray Benchmark or Cinebench, Threadripper 1950X is always faster.

Take a look at the Cinebench scores for example. Or Vray Benchmark where most/average results for 7900X is around the 1:00 min. mark as opposed to around the 0:45 min. mark for 1950X. There are also 3 times more Vray benchmarks posted with 1950X than 7900X which I think shows which is the most popular choice of most users. The system in my signature which I keep OC'ed to 3.9 Ghz for 24/7 work with a simple air cooler (Noctua NH-U14S) finishes Vray Benchmark in 0:41 min and scores 3300 CB in Cinebench. If I had a decent liquid cooler I bet that I'd be able to push it even further. I once tried to test the absolute best Cinebench score I could get with my cooler and I got 3500 CB at 4.1 Ghz (screenshot attached below) but due to my cooling solution I could not keep it at these clocks for stable 24/7 work and reverted back to 3.9 Ghz for a score of 3300 CB. Maybe with a liquid cooler down the road.

I would say the main reasons to get 7900X over 1950X is the single core performance which is ~10% faster. But that should only matter if your main workload doesn't run on more than 1-2 cores. Such would be situations like simulations (fluids, smoke etc.).

Also, if you are not in a rush, I would wait for reviews and tests of Threadripper 2 (2950X) which will be released in a few months and will have even better single core and multi core performance. Recent leaked benchmarks show that the 2nd generation of Zen CPUs such as Ryzen 2700X are ~10% faster than 1700X. So I would expect Threadripper 2950X to be about 10% faster than 1950X too.

[Magnum_design]

I belive you that it?s faster in pure rendering. I?m not entirely after speed though. Is it stable for 5 years you think ? I mean. if it?s pushed to the limit all the time...? I edit a lot of photos in Lr as my main workload. I need to do more reserch on this. Thank you for the help

[Alex_M]

I don't see why it wouldn't be stable for a lot more than 5 years if it operates within the manufacturer's thermal limits which is 85-95C. But if you're going to mainly edit photos, I really doubt it will go anywhere above 65C. The Ryzen line of CPUs from AMD is absolutely nothing like their previous CPUs (Phenom, Bulldozer etc.). It's a whole new architecture created completely from scratch, something that in the CPU world happens not more than once a decade.

Temps wise Ryzen run as cool as any recent CPU from Intel. I would even argue that an AMD Ryzen CPU will outlast an Intel CPU, because Intel are using thermal paste (like the thing you put between your CPU and the CPU cooler before mounting it) between the CPU chip and the metal cap as opposed to AMD which are directly soldering these two elements with indium and gold. So to put it into perspective, Intel cheaps out on their CPUs (even the $2000 18-core 7980XE!) by using a low-cost alternative to soldering which with time will dry out reducing it's effectiveness in heat dissipation and will inevitably lead to climbing temperatures over the years to which the only solution is to remove the metal lid from the CPU, scrubbing out the old paste, putting new thermal paste and then putting the lid back on which not only voids the warranty but also you risk the possibility of completely breaking your CPU.

By the way, are you sure that you need a $1000 CPU if your main workload is going to be editing photos? A $300 CPU such as Ryzen 1700 (or 2700/2700X which are going to be released very soon) will be a far more clever and cost effective choice, not to mention that the motherboards for these non-HEDT CPUs are 2-3 times cheaper than the ones for Threadripper and i9 which is another big cost saving. You can save ~$1000 just from the CPU and motherboard alone. Looking at this article, Lightroom doesn't seem to give you any speed improvements with more than an 8-core CPUs so something like Ryzen 1700 (or Ryzen 2700 which is to be released any moment now) will be the sweet spot. They are good enough even if you're going to be doing a little rendering on the side from time to time.

[Magnum_design]

Thanks for the info on this. I really appreciate My main workload is editing photos, yes, but that doesent mean a do little renderings on the side. It?s about 30% of my payed workload and a hobby as well. And I?m not a big fan of waiting for renders to finish. I?t s probably a bit overkill for what I do, but I can afford it and enjoy having a powerfull machine Thanks agan I will think about Amd !

[Svetlin.Nikolov]

Except for fluid simulations (Phoenix FD) where calculations are always done on only 1-2 cores,

Unless you are using Phoenix 1, this is definitely a false statement. For a while, since Phoenix 3.0, there was only one stage of the entire simulation which was single-threaded, and that was the interaction with the scene, which is now multithreaded in phoenix 3.11.

Note that some Max operations, such as acquiring the meshes, interacting with Max forces, etc., are single threaded as required by Max, so that it won't crash and will return the correct data.

Also note that if you have your cache files saved or written over the network, this could also cause the CPU to idle while waiting for the data.

Please, if you have any scene where you don't have the CPU loaded well, send them over to Support and we'll see what's going on, but don't spread false information

Cheers!

[Alex_M]

Hi Svetlin,

Good bit of information, thanks for the the insight! I didn't know that the newer versions of Phoenix are multi-threaded as I've not used the plug-in for a very long time now. Good to know that it can utilize the full resources. Sorry for my ignorance. So one more reason to opt for an AMD based system.

[urbanite]

Well, in Phoenix Intel is still better because of NUMA I am afraid.

[ijzerman86]

Doesn't Numa only matter on a multi CPU setup?

[Svetlin.Nikolov]

Multi CPUs are NUMA, but some regular CPUs, such as some Xeons and the Threadripper are also NUMA. Fluid sims in general need to move a lot of data back and forth between RAM and CPU, so the bus speed is extremely important. In NUMA, some cores have faster access to some data and slower access to other data in RAM, and this makes NUMA a nasty scenario for fluids. Good thing is that if you split your processor affinity, you can simulate and render simultaneously without the two tasks slowing each other down too much, but you could also simulate on CPU and render on GPU just as well...