CPU or GPU?

[Physics_Hacker]

Which, if I'm going to get a higher end one, is more beneficial for high body count simulations?

[TheFinnishForehead]

Unfortunately, there's no GPU acceleration in US2... Many have hoped for it, and imo it should make it to the priority list as well. You see, serial processing used by CPUs is absolutely shite for simulations of thousands of particles etc. However, parallel processing utilized by GPUs is perfect for it. As an example, an GPU from 2012 would still be far more efficient at simulations than today's most powerful CPUs. This also makes me sad.... I have so many GPUs collecting dust on my shelf. This CPU-only thing limits my simulations so badly. I have 2 GTX Titan Xs (Maxwell, not Pascal or Volta) on my shelf collecting dust, not to mention the older Kepler 780 Ti's. As for my current system, GTX 1080 Ti SLI. And yet, I can't put them to good use...

[Physics_Hacker]

I've done some digging and apparently at some point Universe Sandbox (It may have been one of the legacy versions, I'm not sure) DID support the use of the GPU but they took it away because they were changing stuff. Too bad..a badly functioning GPU acceleration is better than none .-.

In any case, if I had two CPUs instead of one, would US2 be able to utilize both? I'm considering getting two CPUs both with 8 cores (which would make 16 in total) and really high clock speed for significantly cheaper than getting one CPU with lower total specs (lower clock speed and a few less cores) but if US2 won't even utilize both CPUs then obviously I should just go with the one.

[im4space]

I think that it was version 14 of Universe Sandbox ² that last supported GPU acceleration. They are looking into ways to speed up US ² and may add GPU acceleration in the future, but they may find other ways that are faster.

According to the article in how to geek, a single CPU  is more efficient than multi CPU's. See

https://www.howtogeek.com/194756/cpu-basics-multiple-cpus-cores-and-hyper-threading-explained/

"A quad-core CPU has four central processing units, an octa-core CPU has eight central processing units, and so on.

This helps dramatically improve performance while keeping the physical CPU unit small so it fits in a single socket. There only needs to be a single CPU socket with a single CPU unit inserted into it—not four different CPU sockets with four different CPUs, each needing their own power, cooling, and other hardware. There’s less latency because the cores can communicate more quickly, as they’re all on the same chip."

[TheFinnishForehead]

Dude, if I were you, I'd go for R9 1950X. The entire system will probably be cheaper than dual-Xeon one, and u can overclock that 16 core 32 thread monster to 4 GHz. Just remember to use watercooling In terms of optimization, US2 is doing pretty good, with all my cores being utilized at around 50 ish % and threads at 30-40%. X399 platform is, in terms of price to performance, god tier. For gaming it's waaaaay overkill, but if u want a cheaper alternative that packs one hell of a punch, for editing and simulation use, go for X399 platform. Alternatively, if you are a diehard Intel fan, go for their X299 platform. Tho.... it will cost u way more. Their 10 core CPU (i9-7900X) costs the same as R9 1950X and doesn't come even close to Threadripper's performance....

[Physics_Hacker]

I think that it was version 14 of Universe Sandbox ² that last supported GPU acceleration. They are looking into ways to speed up US ² and may add GPU acceleration in the future, but they may find other ways that are faster.

According to the article in how to geek, a single CPU  is more efficient than multi CPU's. See

https://www.howtogeek.com/194756/cpu-basics-multiple-cpus-cores-and-hyper-threading-explained/

"A quad-core CPU has four central processing units, an octa-core CPU has eight central processing units, and so on.

This helps dramatically improve performance while keeping the physical CPU unit small so it fits in a single socket. There only needs to be a single CPU socket with a single CPU unit inserted into it—not four different CPU sockets with four different CPUs, each needing their own power, cooling, and other hardware. There’s less latency because the cores can communicate more quickly, as they’re all on the same chip."

But my main question is, will it work at all?
I plan to be having many thousands of objects in a simulation and run it for quite awhile. Will the decrease in performance due to the use of different chips be so significant as to make the use of a second CPU totally pointless or will it only decrease it from the exactly-double-performance one might expect? I'm willing to sacrifice that drop if it's still better than getting a single CPU that would be ridiculously expensive in comparison.

Dude, if I were you, I'd go for R9 1950X. The entire system will probably be cheaper than dual-Xeon one, and u can overclock that 16 core 32 thread monster to 4 GHz. Just remember to use watercooling In terms of optimization, US2 is doing pretty good, with all my cores being utilized at around 50 ish % and threads at 30-40%. X399 platform is, in terms of price to performance, god tier. For gaming it's waaaaay overkill, but if u want a cheaper alternative that packs one hell of a punch, for editing and simulation use, go for X399 platform. Alternatively, if you are a diehard Intel fan, go for their X299 platform. Tho.... it will cost u way more. Their 10 core CPU (i9-7900X) costs the same as R9 1950X and doesn't come even close to Threadripper's performance....

At this point I'm probably going to have to use water cooling anyway.
And, I've been looking at a CPU that I would use two of that normally clocks at 4.7 GHz and can be overclocked to 5 GHz, and has 8 cores, so I'd say that at least based on those specs having two would be crazy good, even with a drop in performance because they're separate, but I am admittedly pretty new to this, so my choice of CPU could be totally wack for what I want to do, so for your opinion...

https://pcpartpicker.com/product/pGrG3C/amd-cpu-fd9590fhhkwof

I'm not set in stone (yet) but with my limited knowledge and lack of infinite knowledge and/or pretty good researching skills it's pretty likely I'm going to make a really big mistake at some point so if this, my first attempt at finding a good CPU for this or any PC is that mistake please tell me an direct me to something better, but I am on a budget here so I can't just build an infinitely amazing PC...
Also, even with one of those guys, how could 4.7 GHz and 8 cores not be good?
Also, due to the kind of long term usage of this thing I'd rather go with normal clock speed than overclocking...call me paranoid but I don't think overclocking something from really long time periods is good for said thing .-.

[TheFinnishForehead]

Oh no no no. Not the FX series. Seriously. They may have high clockspeeds but that isn't all there is to performance. This one would get rekt by an i5. AMD's Ryzen lineup is the 1st amazing one they have made in a long time. So, R7 2700X (better go with Gigabyte mobo and get GSkill Trident Z 3200MHz CL14 RAM (2x8GB or 2x16GB sticks). Also, it MUST BE CL14), or if u want the ultimate performance, R9 1950X (or the possible Ryzen 2 based Threadripper CPU lineup that'd be coming in the future). I'd recommend R7 2700X tho. It's cheap and the most powerful consumer CPU ever made (I'm excluding the prosumer i7-Extreme, i9, Xeon and R9 lineups)

[Physics_Hacker]

Why those though? The R7 2700X doesn't clock to as high as the ones I've been looking at and both have 8 cores, and while the R9 1950X is better than that, the lowest price is $700 which is basically half the budget I have so I'm kind of hesitant to do that and I can't even seem to find any clock speed for it...
What exactly am I missing here? I am new to this, so it's not that I don't believe you, I just don't understand. I want to get the best possible for my budget, and for what I want to do. So...What am I not seeing?

[TheFinnishForehead]

First off, FX series uses clusterized threading. Secondly, the cores themselves are so much inferior to the ones Intel (and AMD Ryzen series) uses. There exists a few things that affect the overall performance of a CPU. Clock speed is merely one factor. The real deal breaker is the IPC (Instructions Per Cycle), which is how many operations your CPU can do given any clock period. This is where Ryzen and FX differ considerably. Ryzen has so much higher IPC (not to mention completely reworked caches etc) that you could OC that FX until it melts, and you still wouldn't get Ryzen level of performance. I'll find some examples for ya bro, hold on...

[TheFinnishForehead]

Ok, for US2, u want multithread performance... here u go

[TheFinnishForehead]

In these cases, higher is better (obviously). R7 2700X completely wipes the floor with FX-9590. It isn't even a fight. It just doesn't stand a chance. This is not only limited to Cinebench, R7 2700X crushes in every benchmark. As u can see, it even comes very close to Intel's newest 10-core 20 thread i9-7900X, and outperforms their older generation 10 core 20 thread CPU, the i7-6950X. Oh and here's the other difference. R7 2700X has 8 cores and 16 threads, while the FX has 8 "cores" and 8 threads. (Also, the problem with the FX CPUs is that it is a modular design. Every two physical CPUs shares a single FPU (Floating Point Unit). If both CPUs must use the FPU in each module, then one core basically needs to wait and do nothing until the other core is done using the FPU. Therefore, the FPU itself is a performance bottleneck.)

[TheFinnishForehead]

On a side not, for gaming overall, the single core performance is the key, and there's also the fact that even if a game utilizes all cores well, better single core makes for better multicore as well. So, here ya go. Single core data.

[Physics_Hacker]

Thank you, that is definitely more worth it, guess the price says more about the quality than I thought :-/

Also, would you mind PMing me about other components involved with this? The more advice I can get from people more experienced with this stuff than I, the better.

[TheFinnishForehead]

PMd ya bruh. Don't be shocked tho.... I don't wanna recommend shite parts so it may be rather expensive, but I guarantee that your platform will serve you for well over 5 years, u just have to change your GPU every now and then

[Cesare]

Developers should also optimize Universe Sandbox 2 so that it can run simulations properly on slow computers, as well as powerful computers.

[im4space]

Developers should also optimize Universe Sandbox 2 so that it can run simulations properly on slow computers, as well as powerful computers.

They are doing this on slow computers such as iPads, iPhones, and Android phones and tablets. While it may be some time before this is released, they are working on this now.

[TheFinnishForehead]

Developers should also optimize Universe Sandbox 2 so that it can run simulations properly on slow computers, as well as powerful computers.

The problem is that with all, including potato, computers, there's an absolute upper limit. To go beyond that, you'd need new algorithms developed to speed up processing and new APIs etc... For example, even if US2 used up 100% of your CPUs power, meaning it is optimized very well, and was executing the code the best it could (with perfect code too....), after that, it'd be impossible to improve. And no one has time to waste on god tier optimization, especially for smaller studios. The task would take ages even for massive studios. The optimization for US2 as of now is very good. And this won't likely change for a long long time, and with new features requiring even more processing power, there's no real need to even attempt to optimize towards perfection. Just buy a computer with Ryzen 7 2700X, Ryzen 9 1950X, Intel Core i9-7900X or Intel Core i9-7980XE if you want some really good performance.... 7980XE running at 4.6GHz for example would be literally the fastest CPU you can run US2 on, it will demolish all Xeons except for Phi's or multi-socket systems (which I doubt US2 supports anyways....)

[TheFinnishForehead]

On a side note, what they really should be doing is transferring that load to the GPUs to handle. In simulations, GPUs are literally THOUSANDS OF TIMES FASTER than CPUs, that's why all modern supercomputers are basically just huge stacks of GPUs (ish, considering they use Tesla's and Phi's or Radeon Instinct's etc....)

[Cesare]

I am glad that developers are in the current process of making Universe Sandbox 2 run on slow machines. Tablets on the other hand are very slow, so developers must build Universe Sandbox 2 suitable so that performance can go on smoothly on tablets, iPhones and slow laptops.

[joe9804]

I think you should go for GPU. The same question I was asked the same question to Netgear Router Support USA and they suggest me to buy a GPU. Because it GPU works more accurately on a specific work. SO just go for GPU or you can also contact them for a suggestion.