Chaos Corona Forum
General Category => General CG Discussion => Hardware => Topic started by: shortcirkuit on 2023-03-06, 01:04:29
-
Hi all
Ive had a local PC build company quote me on a build - just wanted to get your thoughts on it to see if theres anything that can be changed/improved?
CPU - AMD Ryzen Threadripper Pro 5995WX
CASE - Fractal Design Meshify 2 XL
MOBO - Asus WRX80E-SAGE Pro SE - WRX80 -
GFX CARD - Gigabyte RTX 4090 Gaming OC - 24GB
COOLER - Cooler Master ML360 Mirror TR4 | 360mm Radiator + 11 x 120mm fans
POWER - 2000W Cooler Master M2000 Platinum
RAM - 32GB Crucial 3200MHz (1x32GB) (8 OF THESE - 256GB)
HARD DRIVE #1 - 1TB Gen4 Samsung 990 PRO M.2. NVME (R:7450MB/s | W: 6900MB/s) - used to install windows and main applications
HARD DRIVE #2 - 4 X 2TB Gen4 Samsung 990Pro M.2. NVME (R:7450MB/s | W: 6900MB/s) - raid zero set up to maximise read and write speeds. Will have all my 3D assets on here.
what are your thoughts on this setup? any help would be appreciated.
-
32GB RAM is too low IMHO. Would recommend 64GB absolute minimum
-
Seems pretty similar to what I got including, to an extent, the memory (I'm using Ballistix, 8x16GB) and it works really well for me. I did opt for an air cooled system with a Noctua U14s.
Unless you want to slot a second GPU in there I do think the 2000W PSU is overkill. I would guess it'll be below 50% load so efficiency would take a hit.
@davetwo, I think shortcirkuit plans on populating all 8 of his slots with 32GB sticks.
-
thanks for your replies guys - good to know.
Do any of you have this raid set up though? does it make it quicker?
-
I mean I got a couple of WD SN770s (Gen 4, 5GBps) for my main drives and honestly for the main tasks that I do I find these perfectly adequate. Wouldn't ever think about RAIDing them. The speedup I get when transferring files (animations etc.) or the tutorial recordings is pretty great compared to a regular SSD but it does kind of end there - outside of that I personally wouldn't be able to notice the difference. So I guess as subjective as this statement might be, I personally don't think RAID is necessary. Unless you are constantly transferring over a _crap_ ton of data. :)
-
Thanks for the repliy mate.
So correct me here if i am wrong, the 13900KS on the benchmarks is around 30 seconds. I currently have a 3970X TR and these results are around 27-29 secs.
Am i missing something here or is the 13900KS just a little slower than what i currently have? the price of a 13900KS is just over 1K AUD. Massive price difference compared to the 3970x - with me now wanting the 5995xw which according to the benchmarks is around 11 seconds and costs over 11K AUD. Price vs Performance wise, am i missing something or am i better off getting 2 or 3 13900KS instead?
-
1) Give up Raid. That's just nonsense nowadays. And it won't make any read/write of your files faster in reality. I have Enterprise Optane drives, it doesn't make 3dsMax read textures even fraction of second faster and Optane is magnitude better than the sequential monstrosities with NAND :- ). You will never utilize the sequential advantage over single drive, they're already absolute over the top. Save the money, save the complexity, be more safe.
2) I have no issue with 2000W PSU :- ). But since adding second GPU is almost impossible nowadays, high-quality 1600W PSU might be better choice. That will still be overkill (but overkill is good for transient spikes and stability), but closer to ideal efficiency curve and it will be of higher quality.
3) Yes, 13900K, 7950X/7950X3D are in fact, faster for working during 3dsMax than current Threadripper/Xeons. With Threadripper/Xeons, you get the multi-threaded advantage, which is good if you also use your own PC a lot for rendering final files as well (like overnight).
Sadly Threadrippers due to PRO tag got lot more expensive, so they are the same price as upcoming Xeon-S (expect my testimonial video soon lol). A lot of money, for less improvement than you might like. If you only need single PC, buying them still makes sense. If you own studio with employees, I would buy everyone 13900K/7950X and only buy Threadrippers/Xeons/Epycs as Render-farm.
Also, you have to account for the price of components, buying 128-256GB DDR5 and 2TB of SSDs and current motherboard and PSU prices really offset some of the sting of Threadripper/Xeon build. Don't buy 3 Mainstream PCs for personal use :- ). It's either 1x Mainstream platform + 1x Enterprise platform, or just 1x Enterprise platform.
HERE COMES THE MOST IMPORTANT ADVICE :- ) Right after you build the PC, INSTALL NEW FIRMWARE for the Samsung 990 Pro. Like the first thing you will do.
-
Thanks for the repliy mate.
So correct me here if i am wrong, the 13900KS on the benchmarks is around 30 seconds. I currently have a 3970X TR and these results are around 27-29 secs.
Am i missing something here or is the 13900KS just a little slower than what i currently have? the price of a 13900KS is just over 1K AUD. Massive price difference compared to the 3970x - with me now wanting the 5995xw which according to the benchmarks is around 11 seconds and costs over 11K AUD. Price vs Performance wise, am i missing something or am i better off getting 2 or 3 13900KS instead?
I would probably take those 13900KS results with a grain of salt. Those are probably overclocked entries so the 13900KS I would still expect to be slower than the 3970x although by all accounts definitely not by a very noticeable amount. My impression is that the 13900KS probably falls somewhere between 34-38s on the benchmark scale. That said, the benchmark doesn't necessarily account for all types of scenes so there might be some additional variation in there going either way imho.
I would most certainly say the same for the 5995wx. My build, as you can see on the entry list, clocks in at around 16-17s. Given that I'm running it without an OC my impression is that this is quite a realistic position for it. It is after all very much like the 3990x but with higher clocks and IPC which should probably account for that extra ~15-20% performance. I definitely can't imagine the 5995wx scoring 11s. That is dual Zen 3 Epyc territory I'd say.
For what its worth I pretty much agree with Juraj on all points except maybe on number 3. I do tend to consider myself to be somewhat fussy for when it comes to waiting for the UI to draw and all those wonderful single threaded tasks where a 7950x or a 13900k is just plain faster than a 5995wx is but when I'm working I really super appreciate the IR speed and I don't quite feel the need for more single threaded performance. I mean it would be great by all means its just that to me it seems like a good in between combo. I haven't used a 13900k or a 7950x yet mind you but the drawback there is that you're just waiting for noise to clear compared to a 5995wx. Depends on your style and needs I guess.
One thing to consider with these new Threadripper Pro builds is that they are incredibly expensive. Especially for arch-viz or product-viz where you don't need 9123412312 PCI-E lanes. So unless you know you'll put it to good use its very much worth considering investing in a mainstream platform and reaping some of that extra single threaded goodness.
-
Well I had to keep SMT(/Hyperthreading) off on my 3990X for past half year to even make IR smooth for me :- (. So it might be twice as fast as mainstream CPUs, the whole start/close of IR and everything associated with the transient workload around multithreaded one is just slower.
I am testing the latest to be released Xeons for Intel and I have to devise some way to start finally measuring these things. There's just something about 3dsMax or 3dsMax&Corona that is still unfriendly towards these high-core behemoths, whether the issue is still scheduler or intra-core communication is mystery to me.
But my laptops are faster, even for IR.
I mean, on paper, I still love Threadrippers and the new Single-socket Xeons-W are effectively 5xxx Threadrippers just with PCI 5.0 and DDR5. And same price.
But I think there needs to be more optimization done of workstation software side to utilize them which will probably never happen in non-Apple space. And def not for 3dsMax. Even damn Photoshop is just slower and slower.
3990X vs 5990WX is not 20perc ;- ). My 3990X had 17seconds stock config on launch day. It's more like 5 perc. Which is what makes 7xxx Ryzens and Raptor-Lake Intel so much more advanced. The enterprise workstation sector is heavily lagging behind these days. Hopefully the new Xeon-W will even speed up AMD's 7xxx Threadrippers, and hopefully a return of non-Pro versions.
I feel like went full circle in past 6 years. From super expensive enterprise/HEDT CPUs, to very affordable ones, back to super expensive.
GPUs are not even worth mentioning with any positive word.
-
gents - thanks for your responses.
So, instead of raiding the 2TB (x4) nvme drives, are you suggesting that I just buy an SSD instead and save some money?
I have a 3970X - so i will use this as a render node & as a machine i can work on while the other is rendering.
@Juraj - so with your 3990 - are you saying IR works bad? how do you even turn off hyperthreading?
-
3990X is long discussion :- ) SMT/Hyperthreading is turned off in bios. It did make 3dsMax overall experience smoother. I will be testing this from now on in newer machines as well.
Regarding Raid, advice is simply not to to raid the SSDs. Buy the same SSDs and in the same capacity that you need. Just don't raid them.
-
thanks for that - but turning it off, doesn't that defeat the purpose of having such a beefy PC? does it not render as fast? what are the downsides of turning it off?
-
thanks for that - but turning it off, doesn't that defeat the purpose of having such a beefy PC? does it not render as fast? what are the downsides of turning it off?
I've noticed about 20-30% of a performance drop on my end here - at least in all the various benchmarks. So my 5995wx went from being a bit faster than a 3990x that has HT enabled, to being slower than it :)
-
thanks for that - but turning it off, doesn't that defeat the purpose of having such a beefy PC? does it not render as fast? what are the downsides of turning it off?
It does.. but it made IR much smoother experience. Honestly worth some deep dive from some knowledgeable authority (like L1Tech Linux dude) what the underlying issue could be.
-
thanks guys ive really appreciated your input - the order has been placed. Cant wait to get it.
-
Hi Juraj,
I see you said "GPUs are not even worth mentioning with any positive word." I thought you liked Fstorm GPU renderer or maybe I'm mistaken. Just wondering why you are saying that about GPU's?
-
Hi Juraj,
I see you said "GPUs are not even worth mentioning with any positive word." I thought you liked Fstorm GPU renderer or maybe I'm mistaken. Just wondering why you are saying that about GPU's?
I think you read too much into my words, I meant it quite literally, in terms of cost/value :- ). It's hard to say positive words about 4xxx nVidia if the only good model (4090 with 24GB Vram) costs 1800+ Euros while you can buy 3090 for 600 Euro second-hand when lucky.
It doesn't mean anything about GPU technology or GPU rendering.
-
Oh okay, gotcha. I was just curious to see your view on the subject if you wasn't fond of GPU rendering. Yeah the 4090s are only worth it if you're going to use it for a compute.
-
If Corona was running on GPU, I would be fond of GPU rendering :- ).
-
Hi guys - i wrote a post here if you have any input id love to hear it?
https://forum.corona-renderer.com/index.php?topic=40073.0