Jump to content

PC specification.. Any thoughts?


Recommended Posts

Hi Everyone,

 

I am specifying a custom build PC for Renderworks and never have done this before, not really my strong point.. Does anyone have any thoughts on the following tower system spec:

 

AMD Ryzen 9 3900 12 core CPU 3.1-4.3 GHZ.

 

Asus Rog Strix X570F Gaming motherboard

 

32 GB Corsair Vengeance DDR4 2400MHZ RAM

 

6GB NVIDIA GeFORCE GTX 1660Ti

 

250GB seagate SSD

 

In particular if I understand correctly, the Geforce GTX is all that is needed for looking at the plan view, and the 12 core is the best way to get fast rendering, based on ; 

 

 

I am unsure about the motherboard in particular its recommended verbally by the sales guy, but £200 more expensive than many of the alternatives.

 

Thanks

 

Ivan

 

Link to comment

If you render a lot on that machine, I would recommend a Threadripper 3000. The question is whether you want to get the one that costs 1000.- or 4000.- more. That depends on how much you really render. Use the Cinebench multicore results as a guide and calculate (extra charge/life cycle)-(salary*saved rendering time).

 

The top threadripper is about 3.5x as fast as the Ryzen 9 3900. So if you are rendering all day long, you woul'd have the money back within a few months.
But if you do not render a lot, the Ryzen is also a very good choice.

 

The mainboard is a little overkill, because you don't have any components that use PCIE-4.0. A B450 Mainboard woul'd do as well. Put the money in faster clocked RAM with good timings. Bout 3200 MHz CL16.

Edited by herbieherb
Link to comment
4 hours ago, herbieherb said:

A B450 Mainboard woul'd do as well.

 

I would go with one of the new 550 Mainboards.

These are basically B450 Chips that have already a few PCI4 Lanes

but no cooling fan like 570. They are also cheaper like 450 but have

even better VRMs now.

 

There came lots of testing videos of the new Asus 550 TUF Plus Gaming

(I have the predecessor B450 TUF Pro Gamin and am really happy with it)

It has 2.5 GBit LAN + Wifi now.

All conclusions where that you normally do not need the 570.

 

4 hours ago, herbieherb said:

Put the money in faster clocked RAM with good timings. Bout 3200 MHz CL16.

 

I agree.

  • Like 1
Link to comment

 

Hi HerbieHerb and Zoomer,

 

Thanks for the great advice, I can see that you are both in agreement about the motherboard being over-specified. I think I'm gonna go for the 550 as you suggested Zoomer. 

But not the 570, HerbieHerb, good to be clear it provides PCIE 4.0 which I don't need..(or an extra fan)

 

There is such a deep level of technical knowledge here....which you folk have...

 

Threadripper would be good, I took a look, but at £1000 it doesn't really make sense, I only use Renderworks about 20% of the time. The rest of the time I use Vectorworks, (40%) Photoshop (20%) emailing the rest....

 

I am hoping this setup will be good for photoshop and Vectorworks too.

 

The entire tower should be delivered for £1,250 so its really inexpensive compared to my current Dell Precision 3530 laptop, which was listed for more than this..with far less power...

I've never ordered a custom PC before, so a bit concerned in case they go bust after I place the order or something else goes wrong... its not the same as having the Dell 3 year on site guarantee...I would buy it with a credit card, but only have a company debit card..

 

Do you have a view on the 32 GB Corsair Vengeance DDR4 2400MHZ RAM? Does this have to be matched to any of the other components..I read that I don't need to match this to the motherboard. (but would it improve performance?)

 

Thanks again to both of you for your great advice. Its very much appreciated.

 

Link to comment

570 motherboards are clearly better motherboards and have more features - if you need them

 

For PCI 4.0,

570 chipset does support PCI4,

550 (=450) chipset does not !

 

But usually 550 boards have 1 PCI slot (GPU) and 1 NVme Slot (SSD) directly

connected to the CPU, which supports PCI 4.0.

Basically enough for most users

All other ports that a controlled by 550 chipset will be PCI 3.0 only.

 

So if you think of using multiple NVme SDDs, GPUs or future PCI4 devices,

you better go 570 instead. Or better Threadripper to have more lanes.

Link to comment

I would go with the X570 because it’s going to be futureproofing your rig and there are a few good contenders in the price range of the B450/550. M.2 SSD’S are rolling out with PCIE 4 support and going by leaks, the forthcoming nvidia ampere/rtx3000 series will support PCIE 4, so the next ssd or graphics card upgrade will definitely use it. I find for me, my cpu/main board/ram does at least one graphics card upgrade before I cycle out the cpu and main board between gpu  upgrades - this spreads out the upgrade cost rather than being a lump sum.
I’d also go with an RTX 2070 super as an absolute minimum - the 1660 is a fairly low end card - OpenGL will leverage the card when navigating around your model and if you get tempted (this happens a lot by people who say they aren’t interested initially) by real-time rendering engines (enscape, Twinmotion, etc) over the long wait for renderworks renders (there are a lot of threads in the forum covering this), you’ll need at least a 2070 and you’ll kick yourself not being able to trial one effectively with hardware below min spec.

Edited by Aspect_Design
  • Like 1
Link to comment

I'm very skeptical whether a PCIE-4 graphics card brings a big advantage for Vectorworks. In my experience it is not the bandwidth of the PCIe connection that is bottlenecking, but the single-core performance of the CPU and the clock speed of the GPU. According to my measurements on a RTX 2070 Super, even on my huge test file, the bus interface load never exceeds 20%. From this point of view Vectorworks would not really benefit from PCIe-4 besides the fast hard disk, but that's exactly what the 550 boards with direct connection for M.2-SSD are for.

Of course, if you plan to use realtime renderers, if they turn out to benefit from PCIe-4.0, and if you actually want to upgrade the GPU in the future, then it coul'd turn out that the 570 motherboard was worth it.

 

On 6/19/2020 at 11:54 PM, ivan mac said:

Do you have a view on the 32 GB Corsair Vengeance DDR4 2400MHZ RAM?

For a Ryzen system I wouldn't use 2400 MHz RAM. 3200 MHz CL16 RAM costs almost the same, but gives you a good 5% extra performance in OpenGL according to my own measurements.

 

Link to comment

Yes, especially with the realtime renderers it is quite possible that they could make use of more bandwidth. I have never tested this before. My statement only referred to Vectorworks itself, where I haven't found a method to push the PCIe-3.0 bus to its limits. Maximum was 20-30%, so I would not see the PCIe-3.0 bus of the graphics card as a critical bottleneck for Vectorworks in the near future.

Link to comment
1 hour ago, herbieherb said:

graphics card as a critical bottleneck for Vectorworks in the near future

With redshift integration coming the need for a better gpu might be a consideration - but without a hint of timing it’s a question of when - hopefully sooner rather than later! I personally think it’s a good idea to think broadly about what the currently use is and what i could use it for when plotting an overhaul; most software gets more demanding on each version generally. Overspeccing a bit now just delays an inevitable upgrade - a 2070/2080 is a reasonable bump compared to the massive jump of say, an Nvidia Titan or a 32 core threadripper, which I don’t think you’d see a return on in both money spent 🤢 or performance 🤔

Link to comment

Quotations should not be shortened in such a way as to change their meaning.

On 6/23/2020 at 11:26 AM, herbieherb said:

I would not see the PCIe-3.0 bus of the graphics card as a critical bottleneck for Vectorworks in the near future.

Of course the graphics card can be a bottleneck for Vectorworks. Typically, the size of the VRAM is critical.

 

I would also not install a GTX 1660 Ti today and probably take an RX 5700 xt instead, but not because of the PCIe 4.0, but because it is a very cheap and performant card with 8GB VRAM. The RTX 2070 Super is also a good offer. For the majority of users, the GTX 1660 Ti is certainly fast enough for Vectorworks. But if you use live renderers, you'll want something faster, whereby the performance increase over the RTX 2080 Ti doesn't justify the price anymore. This is then only worthwhile for very specific workloads.


With CPU it's quite the opposite. Here you get proportionately about the same amount of additional performance for the extra charge including the very high specced Threadripper 3990wx. Therefore it is definitely worth it to do the math, because for someone who spends much time of the day with CPU renderings, even the very expensive Threadripper processors can pay off quickly:

1388065224_hardwarepricesperperformance.thumb.jpg.d6327c7eb57cba23a4be53ee2aa1aba6.jpg

hardware prices per performance.pdf

Edited by herbieherb
Link to comment

@herbieherb sorry did not mean to offend or intend to change the meaning of the quote snippet. If I had of left the PCIE bit in the quote my response would still have been the same. With redshift on the horizon rendering in VW is in for a an overhaul and more so than ever the graphics card will be the most important part of the machine (also noted here about future iterations leaning on the gpu even more) and I think it would be worth the cost to spec the machine forward with a PCIE 4 capable motherboard, as the forthcoming Ryzen 4000/zen 3 chips will als support the current Socket AM4 but only on an X570/B550 chipped motherboard. With that in mind, if redshift or another renderer is going to use all you can give it in terms of bandwidth and gpu grunt, it makes sense to think ahead - it’s easy to change a graphics card but is a bigger job to change the motherboard.

Regarding CPU’s - I have a threadripper and I won’t be buying another one.. I cant justify the time needed - nor will clientS pay for it - to set up a scene in vectorworks (not to mention the lack of good 3D trees, vehicles etc I’d have to work around or purchase), then the time to render, tying up the machine. I’m using Twinmotion for this now and in a fraction of the time it would take to do the above, crank out stunning renders that sell the job. Even factoring in the cost of buying TM and the speed of doing a scene, it’s cheaper then buying a Threadripper. There is just no way I’d ever consider using renderworks again for 3D renders - the only exception currently using custom renderworks for elevations.
In my other life as a VW trainer, when you start going over renderworks and the time to do an offline render,  it kills any enthusiasm that’s been built up over the course of the coaching when they want to see a 3D render... Throw it into enscape - even in its current beta form, and everyone is blown away. Which leads me to my other point above to the OP is to not think “I’ll never use Twinmotion or enscape..or lumion...”, chances are you might just want to try a demo... and get hooked - build the machine with bit of fat in it so you can try these things out, if you don’t it last you longer between upgrades.

  • Like 1
Link to comment

I tried Twinmotion. It was great for Video and I guess that together with VR glasses it is even more impressive. But for static Renderings i don't have time to mess around with export/import, setting textures, exchanging trees etc. I adjust the camera and render with renderworks. I don't need hyper realistic renderings, as we mainly do it to double check our ideas but in the end of the day we also want to show some of these to our customers.

 

This is what i get out of Twinmotion after wasting time with import/export, changing textures and objects etc.:

tm.thumb.jpg.9c817b3df0dc300d53f1fee66aceba71.jpg

 

This is what i get out of Renderworks with less than a minute Rendertime:

rw.thumb.jpg.e51315bb16e007d21f30858756bfa230.jpg

 

So Twinmotion is just a waste of time for me in most cases. The only case i use it is when i want to show the model to the customer in motion.

 

Of course, with live renderers the whole thing is much more accessible and simple. But if you know Renderworks there is no reason to switch to live renderers at the moment, unless you want to show the client the 3D model directly, make movies or VR.

 

I am also excited that Enscape will soon be part of Vectorworks. But even there, the PCIe bus is not a bottleneck. When rendering, the whole model is already in GPU memory anyway. The PS5 will show where the journey will take us when the model no longer needs to be in GPU memory. Maybe you're right and we'll exceed the 16GB/s capacity of the 16x PCIe-3.0 lane in the next three years. So everybody has to ask himself if you put more money into a motherboard where you don't need the feature you paid more for at least a few years, or if you put the money in things that makes your system faster now.

Especially because the 3D live renderers are currently undergoing extreme development and the future system requirements are hardly foreseeable, the second strategie could also turn out to be well.

Edited by herbieherb
  • Like 2
Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...