My question is, will Rhino benefit more from a powerful processor or more and faster RAM? Of course both, but if the budget requires I choose?
I’m limping along with my 6 year old machine, so it’s time for something with more recent technology. I plan again to build a SFF box. Rhino is my main work software, so the build should be focused toward that. I feel like I read somewhere that Rhino does not make use of multiple cores, so processor core count is not as important.
I am thinking about a Ryzen 9 7900X or 7 9700x, with 64gb DDR5 RAM. However, the ITX motherboards I’m considering are capable of holding 128gb. I’m pretty sure that might require cost cutting some where else, and the processors are the most expensive components.
Less memory than needed can result in a significant slow down.
More memory than needed with the excess memory sitting unused does not provide any benefit.
The amount of memory needed depends on the model and what you are doing with it, and what else is running on the system.
If you care currently working with models similar to what you expect to work with in the future then monitor your current memory usage and see close you are currently getting to 64 GB used. HWMonitor will monitor memory usage and can show maximum used since last reset.
You could initially install 64 GB of ram in half the available sockets, and if needed later install another 64 GB of ram. I don’t know the subtleties of doing this.
I’d say the answer depends on the kind of geometry you will be working with. If you are working with lots of sizeable parts/pieces (like the steel beams for a skyscraper) then I think the more RAM you can get the better. But if you are dealing with smaller/fewer things then CPU speed is a better place to spend your $.
Most of the GH routines are single threaded, so they only use 1 CPU core. Some of them are multi-programmed now, but not many. I use GH (not Rhino itself) for 3D printing, and all my geometry is no larger than a loaf of bread. Because of that I opted for a 3.7 GHz Intel I5 instead of a slower I7 because I don’t need lots of RAM to store lots of geometry. My PC has 16 GB of RAM and I’ve never had a problem with that, except for a couple of times when my GH script put Rhino into a disabled loop and it used up all available RAM. You have to work pretty hard to make that happen.
The McNeel GH folks are working on making GH2 multi-threaded, but that’s been going on for quite a while and I’m not aware of any forecasted due date for GH2. You can get a Beta version of GH2, so you might try that and see how it performs with the kind of geometry you typically use.
9700X has less than half the TDP of the 7900X … so it’s colder, fans go slower, less dust, quieter.
Also it’s cheaper and go better in single thread (most of Rhino computations).
7900X would go a bit faster when multithreading with more than 8 cores…
This is good info. When building a SFF box selecting the right components for good temps is more difficult than with a full tower.
I currently have this. Originally had 16, bumped up to 32, then 64. If 32 would be enough I’d be amenable. But with ITX, bumping memory up means replacing the existing modules as there are not extra slots to additional modules.
Thanks for the info. However, I don’t use GH, so at present it does not play into my decision making. I have had some lag working in recent projects. Mostly those done by others that I had to work on. Current machine is a Ryzen 5 5600x. Not a high performer.
This will be an ITX build. Only two memory sockets. I’m not sure currently, but I think it has to be two matching memory sticks. Increasing means replacing both. This I’ll need to verify with the new motherboard.
RAM should be matching sticks. Depending on your workload, 64gb should be fine for a while. I would try get the fastest CPU the budget will allow.
I built a new pc last year around the Ryzen 9 9950x with 128gb of DDR5 (4x 32).
I’m yet to max out the RAM, I could have got away with 64gb for a couple years, but then adding matching sticks in the future would be much harder.
I believe that there are performance benefits to using all of the RAM channels. So if there are 4 channels one is better off having 4 sticks of 8 GB of RAM rather than having 2 sticks of 16 GB assuming all of the RAM sticks have the same specifications and at that time 32GB total of RAM is enough.
The monetary disadvantage could be that in the prior case if you now need to double the amount of RAM. Assuming you cannot have sticks with mixed amounts of RAM per stick or other specifications, then all 4 sticks need to be replaced. Practically, 28GB RAM or 52GB sticks might not be manufactured. I just do not know. However, if these assumptions are incorrect then … Perhaps then there is not as much of a monetary disadvantage. [I think I did the math correctly]
This is not quite the case. For DDR4, the use of four RAM sticks vs two RAM sticks can result in very small increases in performance.
In this case, both of these Ryzen processors are on an AM5 platform, and hence only have DDR5 support. This is a consumer grade platform, where the CPUs have two memory controllers. Partly owing to this, it then typically becomes advantageous to have 2x16 GB vs 4x8 GB.
For 128 GB of RAM, on DDR5, I think then there is the chance that one should consider RAM with slower speeds, if you are going to run all four channels.You also risk losing the EXPO performance profiles, owing to increased chance of stability issues with DDR5. In reality, it can become a sort of lottery with the speed vs capacity of these sticks. In general considerations, I would probably settle on 2 x 32 GB of DDR5 on AM5 boards; just so I stand a chance of running a good profile.
I guess you could use 2 x 48 GB RAM, but I am not convinced that they provide much benefit. Maybe if you are dumping absolutely tonnes of detailed 8K PBR maps into scenes, and rendering on CPU… then maybe so. Maybe if you for whatever mad reason were running local AI models on CPU.
For the largest models you’ll be needing at least 400GB of RAM.
Anyway, lots of memory is useful also if you have multiple programs open. Hitting swapfiles isn’t very fun, especially not when lots of your data gets paged in and out. That becomes very slow.
In my own machine I have a 7900X and 128GB RAM. I don’t think I would want to go back again. Well, my laptop has 64GB, but that still can get close to getting filled.
I think the speed difference between 2 sticks vs 4 sticks is not going to be that dramatic during work. It is anyway going to be faster than paging data in and out.
I mean 64 is probably enough, that’s what I have for the moment, but that’s practically what they recommend for a gaming system these days. Leftover RAM gets used as a disk cache by Windows.
Nice thread. Thank you to all. Quite informative. I deliberately use language / words of e.g. believe, assumptions are correct- so that “learning mode” is turned on and because an exceptional quality to this form is that I feel that the accuracy and precision of the resultant cumulative of responses is very high of a given thread and treads respectively.
Is this exclusively a rhino machine? Or do you have other programs you use in your workflow?
Rhino is more of a pig and not a thoroughbred. But that’s why we love it. As long as there is mud it’s happy.
It sounds like much of your troubles have to do with your graphics card capabilities, and less to do with your ram / processor. This is easily troubleshot by watching your processing graphs in task manager while you were working on a complex model. If your ram is maxed out or you’re processing power is maxed out then there is your bottleneck.
If for example, you have an Nvidia card simply changing a few settings, such as dynamic streaming preset on professional level cards can quadruple your performance.
This being said if you are an exclusive rhino user. I highly recommend grabbing a MacBook Pro refurbished from Apple as your rhino workstation. Not only is apple’s silica highly adaptable to the varying needs of rhinos requests It is for some reason much faster than my $10,000 workstations on some very basic manipulations.
Another important note here is that you can go much further with a high-end laptop (windows) than a custom built PC because the bus speeds and integration of RAM and other peripherals seem to make rhino much happier than custom workstation builds.