Hi there, new Houdini coming from Softimage here. I'm currently finishing my training and will more than probably jump into the fray very soon.

However, one thing I quickly, painfully learn while practicing at home, was that my computer just couldn't handle Houdini. As soon as the load increases, everything will freeze.
I have an I5 3570k, 8gb ram, and a GTX 780. Ram is way too low, granted, but being limited by 4 threads is a pain too.

So here I am wanting to upgrade.

- On one hand there are options to get a used Xeon based workstation at a fairly decent price, upgrading part of my I5 for side tasks / gaming (I'm also a compulsive gamer )( upgrading one component could led to upgrading everything though, thanks to CPU sockets having very limited compatiblity )

- On the other hand I could get a brand new I7 w/ GTX970 and 32gb ram which could do a lot of things for a probably high price, keeping my I5 untouched on the side but also potentially useless, even for rendering.

Both options have pros and cons, obviously. I just can't decide what would be best for me at home - I essentially work elsewhere, but I hate being powerless when I have to do something from home for whatever reasons.

One thing worthy of mentionning is that I do not know the first thing about Xeons and what they could offer me. It's just tales of greatness I heard and read.

So any idea how to get the most bang for my buck ? Also I was told Houdini wasn't able to use neither hyperthreading nor a second CPU. What gives?
Oh, and does the Apprentice edition of Houdini allows network rendering anyway ?

If anyone would be willing to help and give advices…
Thanks !

If you have the budget for a multi CPU system (two or more CPU sockets) go for Xeons since it's pretty much the only option for that.

Otherwise, i7 for a single CPU setup.

It's really a matter of coin. If your budget is limited, there is really no point in going for a single Xeon platform since its performance is often lower than an 8-core i7 (which higher clocked), unless of course the Xeon has more cores. Last one has 18 if I remember correctly, but to get it you'd have to rob a bank. Or own it.

Wolrajh
One thing worthy of mentionning is that I do not know the first thing about Xeons and what they could offer me. It's just tales of greatness I heard and read.

So any idea how to get the most bang for my buck ?Also I was told Houdini wasn't able to use neither hyperthreading nor a second CPU. What gives?

Xeons give you the ability to address more than 64GB of VRAM, and to have 2 or more physical CPUs (dual or quad socket). If you're not interested in either of those, an i7 5960 or 5930 will suit you just fine.

I'm not sure who told you Houdini doesn't support hyperthreading (SMT) or dual-sockets, but it does (in the sense that it simply “sees” more logical processers to run on). Maybe you're thinking of multiple GPUs? OpenCL-compute and graphics can only utilize a single GPU for each.

I am using a dual xeon OSX system at work which presents 16 cores, 8 which are hyper-threaded, to Houdini @2.26 Ghz. I have 28GB ram and I can only preview about 60-70 frames of a pyro sim before the blue bar starts chasing the playback head. This is quite disappointing to me. (2 seconds of realtime playback)

I get about the same experience from my home system which is a Hexcore AMD running @ 3.1Ghz with 8Gb ram.

So I am not really sold on the hype that Xeon is the best. What seems to be the best for Houdini, and I am mainly concerned with simulations, is a fast CPU.

So go for a 4Gz i7 if possible. The faster the better.

Hi, and thanks for the answers so far !

Well I was just as surprised as you hearing about the whole Houdini doesn't support HT/Dual CPU. Especially when I was reading here or on odforce's forums people discussing I7s and Xeons - none of them would make sense if Houdini weren't able to use their special features, right ?

Anyway yeah, what is really propelling me toward Xeons is the fact it IS possible to get some cheap ones. Again, I do not know Xeons specifications, or rather names. I don't know which is which and which is better.
But seeing things like that [ebay.fr], one can start wondering !

I'm very interest into acquiring an I7 anyway, but in the end it's all about upgrading my I5 with cheaper components or replacing it entirely with high end components. I suspect both options would cost me as much - upgrading my I5 means upgrading means ( probably ) upgrading RAM, but keeping HDDs / drives, GPU ( since the 970 is so “cheap” though I'm tempted too ), while getting a brand new comp and pairing it with my I5 means building it all from scratch, no salvage. In the end, I want two computers side by side and I will either use a display switch or a remote desktop access.

At first I was all about getting a 2k euros I7 config ( probably can lower the price with a smaller SSD or things like that ), but seeing the above link I started thinking that maybe a used workstation was a good place to go too. I clearly do not plan to buy and build a brand new dual xeon rig. CPU and MB alone cost more than the full I7 rig

twod

Wolrajh
One thing worthy of mentionning is that I do not know the first thing about Xeons and what they could offer me. It's just tales of greatness I heard and read.

So any idea how to get the most bang for my buck ?Also I was told Houdini wasn't able to use neither hyperthreading nor a second CPU. What gives?

Xeons give you the ability to address more than 64GB of VRAM, and to have 2 or more physical CPUs (dual or quad socket). If you're not interested in either of those, an i7 5960 or 5930 will suit you just fine.

I'm not sure who told you Houdini doesn't support hyperthreading (SMT) or dual-sockets, but it does (in the sense that it simply “sees” more logical processers to run on). Maybe you're thinking of multiple GPUs? OpenCL-compute and graphics can only utilize a single GPU for each.

Even if it was so that Houdini itself didn't take advantage of the extra cpu, it could still be that the general system load would decrease in other areas, hence keep Houdini from crashing. But this is anyway logged by the os , so the source of the problem shouldn't be hard to trace.

Also running it via a debugger would yield info…

Don't go with Xeons if you're just learning and not yet making money from using Houdini (I assume you don't have money to burn).

The only reason to get a dual Xeon workstation is if you need >64GB RAM (you don't need dual for this, but it's the MBs), or you do intensive rendering (or any other process that benefits from multithreading).

There are so many single threaded processes in Houdini (and most applications) that 30+ threads would be completely wasted, and the lower (not over-clockable) CPU speeds will mean a very measly performance unless you're in a position where it makes sense to buy into the vastly more expensive Xeons (think $2k+ per CPU). I've posted on another thread asking the same questions, it's jaw grinding when Task Manager shows an application using just 3% CPU processing power.

There are other Xeon gotchas such as more expensive EEC RAM.

Even the octo core i7 X series are in a similar ballpark for price:performance… a single threaded process isn't using 1/8th of your CPU, it's using 1/16th. Just get a nice fast i7, a motherboard that will let you fit a decent amount of RAM (you don't need to fill it straight away), and a good GPU…spend what you can save on a second GPU if you're thinking of doing anything with OpenCL.

SSDs - system, and scratch (for cache files) - are a good idea.

Regarding overclocking, it gets more difficult (unstable) the more RAM you add, so don't rely exclusively on that.

My few cents anyway.

Hey, thanks for this analysis. It's really helpful.
The training has concluded last Friday, and I already have a few missions scheduled, so actually I am going to make money using Houdini within a few weeks

I will keep apprentice at home since we have fully licenced stations at work - in a nutshell, I'm not going to do any commercial work from home. Meaning I don't need some overly powerful workstations of my own anyway. The closer to ‘commercial’ I'd do would be R&D before taking a job, something that shouldn't be needed anymore overtime as I grow more experienced using Houdini.
What's more we vaguely talked about single-threaded nodes vs multi-threaded ones the last day, but yeah, everything seem to be pointing at getting a faster clocked, single CPU, hexacore i7.

The only real benefit from getting a used Xeon workstation from, say, ebay, was the price, but I fear this kind of rig actually wouldn't be much faster than my current i5 ( to which I just added 16gb ram, making my life a looot easier already ).

So yeah. I guess I'll slowly build a second hybrid multimedia/gaming rig, maybe complete it by the end of summer/beginning of fall, when the next generation of GPUs should be out and - maybe - prices will fall in line.
I'm not in a hurry.

Since you will use apprentice hd license, go for i7.

Buy a 4790k and 32G ram. That's max for consumer 1150 socket.
I won't recommend 2011-3 socket yet because of DDR4 Price.

Use this rig for a year, you might make enough money from it to buy dual xeon rig. Then you can decide whether you need a dual xeon or not. Or if you got a big contract.

I hace 48 cores dual xeon e5-2697 and 64G ram at work. And ram never be enough for work. But for rnd i7 would be perfectly capable.

Hey,

I have i7 4960x (12 threads) and dual xeon 2630 v3 (32 threads) , the i7 is most faster than the dual xeon in flip simulation, im not sure why…


cinebench R15:
i7 4960x = 1070 points.
2x xeon 2630 v3 = 2000 points.

someone know why the dual xeon is more slow in simulation?

thanks in advance!
Carlos

A Xeon 2630 has a lot of cores but they are each clocked at 2.30ghz each, meaning any calculation that cannot be multithreaded is going to be approx %50 slower compared to an 4960 per core which runs at 4.0 ghz

I am not sure but if we calculated the raw power of these cpus it could be sth like this

Xeon 2630 * 2X = 2.3GHZ * 12 (physical cores) = 27.6 GHZ
Single i7 4960 = 4.0 GHZ * 6 (physical cores) = 24 GHZ

Almost identical, , an I7 system doesn't have the overhead of syncing with the second CPU in this case would give it an advantage also.

So in my opinion, you will only get slightly faster mantra renders as it can multithread very well. For any simulation that is not %100 multithreaded the 4960 will be %50 faster than the Xeon. A raw flip or pyro simulation without collisions should multithread well until you start adding custom nodes etc.

I've been having the same questions. I'm going in circles debating if I should spend the additional money to create a dual xeon setup or single cpu.

Options i'm considering for dual vs single cpu:

dual xeon e5-2630 v3 2.4Ghz, 8 core cpus
single i7 Skylake or 5960x, or even a higher Ghz e5.

With the dual xeon setup, it will cost more, but have lots of upgradability (in terms of more ram over time)… Will plan on getting 64gb to start. And also, the cpus could get upgraded, when the higher e5s become a bit cheaper.

As for the single setup,
It would be cheaper, but have much less upgradability… Unless however, if I find a dual socket mobo, and only utilize one socket temporarily until I can match the cpu. This would allow for more upgradability as well.

Something like the i7 skylake won't give much options for upgrades because, it can have max of 64gb ram, and it's currently the best version of the processor. Therefore, that system would just have to last, and then buy a whole new system at another time.

What's been really getting at me though is if the dual xeon is really going to give me “THAT MUCH MORE” performance than the single setup. As noted, the single, higher clock cpu, may win out in more single threaded scenarios, whereas the dual xeons, will win in the multithreaded cases.

For an all around computer, for big sims, rendering, and essentially the entire pipeline, it still feels the dual xeon is the better “and longer lasting” workstation to go with. However, I am still looking for more info on that.

itriix
I've been having the same questions. I'm going in circles debating if I should spend the additional money to create a dual xeon setup or single cpu.

Also look at the ark.intel.com site for the main information. When comparing benchmarks in Houdini, I've noticed these to be important: Cache size, QPI link speed, Base and Max Turbo Freq, Max Memory Bandwidth.

Not so sure if the Instruction ‘Set Extensions Available’ is though.

What's a good i7 CPU with good performance (high Ghz and cores) but won't break into the 1k range?

pusat
What's a good i7 CPU with good performance (high Ghz and cores) but won't break into the 1k range?

The 6-core i7 5930K (3.5Ghz, $580us) or the 6-core i7 5820 (3.3GHz, $390us). Both are socket 2011-v3 CPUs.

Rumour has it that they're good overclockers as well, if you're willing to do such a thing to a workstation But you can probably get a few hundred more MHz out of them without much trouble.

twod
The 6-core i7 5930K (3.5Ghz, $580us) or the 6-core i7 5820 (3.3GHz, $390us). Both are socket 2011-v3 CPUs.

Rumour has it that they're good overclockers as well, if you're willing to do such a thing to a workstation But you can probably get a few hundred more MHz out of them without much trouble.

Thanks Mark. I am glad to see there are such options for us too Do you know a good mobo that goes with either CPU? I guess there are many options for those.

Keep in mind that max turbo frequency does not mean much since it is the max frequency a single CPU core can climb to, when it comes to multi core performance base frequency is your benchmark. CPU's quickly climb up to their thermal design power once pushed in all cores and usually float around base frequency if maxed out for prolonged amount of time.

Honestly 64gb ram with a decent i7 cpu is freelancers best bet imho. We all almost use 64gb single cpu fast xeons at work for the fx, for anything extreme the task is divided in as many sub tasks as possible for farm distribution. I am on holiday and far away from my pc, and i did some small tests on my mac mini with 2gb ram and i was surprised how much i could do on this 2core 2gb ram machine with H13 …. So optimisation is the key guys

Hey guys, just wondering whats best these days to go for to use for Sims? Considering buying a used HPz840 for around 2k but wonder for that price (and whats actually faster for Sims?) to go for a relatively new i7?

Cheers!

mandrake0
If you are a indie user get some threadripper with 128gb and split the sims. Can't see a better solution that works for that price and over 512gb ram is much cooler( for a indie mostly overkill but it's possible ). GPU are cool but driver issues are sometimes not funny.