What computer for SU?
-
@ands43 said:
Does SU use the video card a lot in which case I'll ask them to get a high spec imac with Nvidia GeForce graphics or will I be fine with the RAM maxed out on a macmini.
I have no experience with Macs & Sketchup but a dedicated Gfx card would be my preferred choice if you plan to do some more serious (more high polycount) Sketchup modelling.
Maybe this spreadsheet with computerspecs might be of help: https://docs.google.com/spreadsheet/ccc?key=0AmovuDT9v-e-cGRWOGUtTGxkT0VwUXc2azMtb3ZpZmc%26amp;hl=en#gid=1
-
@alexandre sk said:
normally you want the fastest processor you can. skp uses only one core, so you need a fast one, and don't matter if its 2 - 4 or 6 cores.
Oh great, that means that the new dual Xeon rig, which I've practically had to sell a kidney to order, is going to be slower than the dinosaur I'm using at the moment
-
@hieru said:
Oh great, that means that the new dual Xeon rig, which I've practically had to sell a kidney to order, is going to be slower than the dinosaur I'm using at the moment
If your old rig had a higher single core performance... yes, the new one will be slower for SU. This is the problem with the focus on multicore of the last years (and future)...
The best processor for SU at the moment will be an overclocked Haswell i7 with maybe 4,4-4,5GHz or more. -
i5 2500k I run at 5ghz.
But will you render too? considered that and get as many cores as you can if you will. Some new renderers are GPU based and will want a big video card instead.
-
Numerobis,
The Xeon model I've gone with is only 2.8GHz and my current chip is an 3.2GHz i7. Not much of a difference, but enough to concern me.
Does chip architecture not make a difference, how about using a SSD for the system disk?
Otherwise I'll have to consider over clocking the Xeons.....or swap back and forth between PCs if the Xeon performance is noticeably bad.
It suppose all this now puts me firmly in the same camp as everyone else insisting that SU needs to go 64 bit.
-
@hieru said:
Does chip architecture not make a difference, how about using a SSD for the system disk?
architecture will make a difference.. depends on how old of an i7 you're talking and how new of a xeon.
(as far as doing straight clock speed comparisons goes.. as in 2GHz from 8 years ago isn't the same as 2Ghz now.. but just look at the single core benchmarks between both chips to see how they compare speed wise)@unknownuser said:
Otherwise I'll have to consider over clocking the Xeons.....or swap back and forth between PCs if the Xeon performance is noticeably bad.
you can't really overclock a xeon.. i mean, you probably can but benefits are going to be negligible..
one thing about the xeons is they'll run at their turbo speeds much better than the i7 will..
the new mac pro's 12 core uses a 2.7GHz xeon (E5-2697 V2) which turbos at 3.5GHZ.. from everything i've seen so far, if you ran sketchup on that cpu, you could expect it to stay at 3.5GHz, or very close to it, throughout a session.
@unknownuser said:
It suppose all this now puts me firmly in the same camp as everyone else insisting that SU needs to go 64 bit.
i don't think 64bit would change anything in these regards.
-
Thanks Jeff, that's really helpful and addresses a lot of my concerns.
I suspected that a current chip at a lower speed might be more efficient, so not a like-for-like comparison with an i7 from five years ago. It's also good to hear that I should get a better performance than the new Mac Pro
As for 64 bit, if 32 bit throttles SU by limiting it to one core, surely access to multiple cores via 64 bit will allow SU performance to skyrocket. Yes, I'm aware that in general 64 bit may slow down SU, but surely the ability to make use of the additional computational resources would more than make up for that?
-
the current 64bit threads around here would seem incredibly mild compared to what we'd see if it were possible for sketchup to use all cores simply by going 64bit..
likewise, the developers would of went 64bit a long time ago if that were the case.sketchup isn't unique in it's single core use for most modeling tasks.. ALL of the apps do this.
going 64bit would allow sketchup to use more ram. but the thing is, sketchup barely ever runs out of ram.
basically, say a big model is using 2GB ram in 32bit.. if it were 64bit, it would still only be using 2GB so no real improvement has happened. -
I didn't realise that - I assumed that most modelling apps would (like rendering software) embrace the computational possibilities of multi-core. It just seems crazy to only use 1 core when anything up to 24 cores could be available...or more with networking.
I'm not alone in thinking this, right?!
-
@hieru said:
I didn't realise that - I assumed that most modelling apps would (like rendering software) embrace the computational possibilities of multi-core. It just seems crazy to only use 1 core when anything up to 24 cores could be available...or more with networking.
I'm not alone in thinking this, right?!
right, you're not.. we all wish modeling apps could use all the cores.. it's just that most of the calculations are linear in nature and can't be divided up.. calculation A has to happen before B can be determined followed by C.. A,B,&C can't all happen at once on separate cores since the results of one needs to happen before the next can occur..
real basic/rough example-
8 ÷ 2 x 4 - 6 = 12
if we tried to put that on two threads (8 ÷ 2) • (4 - 6) then the result is wrong (-8)
not sure if that helps but it's something along these lines.. it's not that it's 'hard' to multithread certain processes.. it's that it's impossible.
but this is why a fast clock speed is better in linear calculations.. it can go through a line of numbers faster.
where as with rendering-- say you have 4 cores and are rendering an image the size of your screen.. a rendering program can basically divide your screen into four smaller sections and have each core work on the smaller section as if it were a single core working on a smaller image.
-
another analogy that may click.
say you have to deliver a package from point A to point B which are 300 miles apart.. if you go 60mph (clock speed), it will take you 5 hrs to get there.. if you increase your clock speed to 100mph, you'll get there in 3 hrs.
but-- you can't simply use three cars (multicore) and cut your trip to 100 miles long.. one single car needs to take the package the entire distance.
-
That makes sense Jeff. It still seems a bit of a waste of resources though.
It's a case of 'the more cores the better' with CPU rendering, so I don't think it's unreasonable for programmes like Sketchup to also make use of that computational power.
-
@hieru said:
That makes sense Jeff. It still seems a bit of a waste of resources though.
It's a case of 'the more cores the better' with CPU rendering, so I don't think it's unreasonable for programmes like Sketchup to also make use of that computational power.
it is sort of unreasonable though.. when the GHz race stopped and the multicore race began, lots of people didn't even blink an eye and just started buying more cores instead.. in many situations, doing more harm than good and spending too much on resources that will sit idle.. relatively speaking, very few processes can be run in parallel (talking about all apps even.. not just CAD apps.)
coders used to rely on hardware to speed up their inefficient code.. the hardware is sort of at a standstill now so it's up to the developers to rewrite/refine their algorithms so they'll run faster.. or accomplish the same thing with less steps.. some programs do this better than others and when we compare apps, we can sense that sketchup has some not-so-smooth algorithms in place which is why it bogs down quicker than other programs.
but if you can come up with the algorithms to allow naturally linear processes to run in parallel then you'll pretty much be an instant millionaire and instant hero.. it's not as if there aren't major incentives to figure out how to do it.. you also have to realize some of the smartest people in the world are trying to figure it out and can't. so in that regard, it is sort of unreasonable to expect sketchup to run most of it's operations on multiple cores.
-
When you put it like that, it does sound unreasonable. I suppose we just need to wait for some genius to make a breakthrough that will revolutionise the way that software can utilise the advances made in hardware development.
In the mean time I suppose we'll just have to hope that chip speeds start to increase again.
-
@hieru said:
In the mean time I suppose we'll just have to hope that chip speeds start to increase again.
heh, yeah.. but we can't let the developers off so easily
sketchup was written during the ghz race so a lot of the code was "we'll just write it like this for now and speed will double every year or so with hardware advancements.".. but it only worked out that way for a handful of years then tapered off.
a lot (most?) of that code is still in there as the back bone of sketchup. if the developers were to convert to 64bit, they'd be rewriting this legacy code and it should present an opportunity to not only update the program to 64bit but, maybe more importantly, throw out some of this GHz-race based algorithms and write smarter/more efficient code with the mindset that 4Ghz is basically max clock speed.. it's a big undertaking to convert to 64bit.. it's much more challenging to redo the algorithms. if they simply kept everything the same but made it 64bit then that's sort of a waste.. if they redo the way sketchup fundamentally works under the hood, then it's not a waste..(disclaimer- lots of speculations made in this post
) -
I have to wonder whether some of the resistance to 64 bit is really motivated by a preciousness towards the original code and a desire to avoid the complexities and consequences of rewriting outdated algorithms.
-
@hieru said:
Numerobis,
The Xeon model I've gone with is only 2.8GHz and my current chip is an 3.2GHz i7. Not much of a difference, but enough to concern me.
If your Xeon 2,8GHz is something like the E5-2680V2 you have 3,6GHz single core turbo, so no, there is almost no difference to your 3,2GHz i7 if it is a 3930K with 3,8GHz single core turbo, since the E5 V2 is Ivy Bridge based and a few percent faster at the same clockspeed. If we're talking about a old i7-9xx 3,2Ghz (960/965) then your xeon should be clearly faster.
-
Yep, I'm going with the E5-2680V2....they seem to be rare as hen's teeth & my builder has had to order them from the States.
My current chip is an i7 960, which runs at 3.46 GHz with Turbo (probably slower the way it's been performing recently), so the Xeons will certainly be faster.
Thanks for the help & reassurance. Time to breath out a long sigh of relief and crack open the first beer of the weekend.
-
@hieru said:
Yep, I'm going with the E5-2680V2....they seem to be rare as hen's teeth & my builder has had to order them from the States.
if it's not too late and you can scrape together another $250 or so, get the top end 8core instead.. that's the one-- especially for a modeling/render combo chip.
.
that aside, some people are already modding the new mac pro with the 10-core e5v2.. pretty sweet
http://9to5mac.com/2014/03/04/upgrading-a-late-2013-mac-pros-to-10-core-intel-xeon-cpu-video/
-
The 2680 comes In 8 or 10 core flavours and I've gone for the 10 core version.
Advertisement