Jump to content

CPU/GPU choices for arch student


Recommended Posts

Hello,

 

I am building a workstation for a friend who is an architecture student. He is not very tech savvy, whereas, I'm the engineer/ hardware enthusiast who knows very little about CAD/rendering programs. I'm looking at a budget-conscious GPU for the following programs (not sure about the older versions of the programs, I assume that's what he has access to.)

 

Adobe CS5 (Acrobat, Photoshop, Illustrator, InDesign)

 

Autodesk AutoCAD 2011

 

Autodesk 3DS Max 2012

 

Autodesk Revit Architecture 2012

 

Autodesk VIZ 2008

 

Google SketchUp 8

 

Rhinoceros 5

 

Right now, I'm waffling between a gtx760 4gb (~$300), a $200 HD7950, or a cheap quadro card. I heard that the quadro cards have much better driver optimizations for viewports etc, but lack raw compute power. Also, is Nvidia vs AMD a concern for the applications listed above?

 

Last but not least, how useful is hyperthreading for the programs above? I am wondering if its worth $100 more for an i7 over an i5, but I am not sure how well multithreaded the applications are.

Link to comment
Share on other sites

GPU compute is "relative" only for progressive GPU renderers like iray/Vray RT/Octane etc. If you friend doesn't use those (only available for 3DS Max for the programs listed above, and last 2 weeks for Sketchup too), you shouldn't care about GPU compute that much...

 

That immediately makes the GTX 760 4GB irrelevant also, as 256bit cards like the 760 cannot really address more than 2GB of RAM simultaneously, thus for viewport performance - including games - 2GB vs 4GB versions of this card (and any 6xx/7xx other that the 780) are performing identically.

Yes, in progressive renderings the 4GB buffer will allow it to load bigger scenes / textures, but won't make it any faster in any meaningful way. An the price difference between 2GB and 4GB versions is not that small.

 

That said, there are really no "cheap quadros" to buy retail...there is the ultra-affordable K600 that is kinda underwhelming and will set you back $150 or so, and then the K2000 jumps to $550 or so...

 

So, you are pretty much leaving the Quadro options "out", unless you are very budget conscious and you (your friend) cannot spend more than $150 for a new card, when a small quadro will probably offer you the best all-around performance for this price - completely forgetting about gaming performance ofc (something that maybe is still a desired quality for your friend?).

Overall the 7950 will give you slightly better performance in OpenGL based programs like Rhino and Sketchup, and more or less equal in the Direct3D based Autodesk products over the 760. It will also run slightly hotter and draw more power, but note that unlike most ppl in the webs, I when I say "hotter" I refer to absolute values - I don't mean that the card will run out of specs and self distract/burn out.

Should your friend get access to the newer Adobe CC suite and uses OpenCL GPU compute routines with it, the 7950 is absolutely unbeatable by anything nVidia has to offer - including GTX Titans, K6000 or w/e...

 

As far as HT goes, yes, all the above programs when rendering and only when rendering (or applying certain PS filters or transcoding video with Premiere / After Effects) do support and benefit quite noticeably from more threads.

For regular operations, photo editing and modeling, all of the above are single threaded, and whoever claims that overall can notice a difference using a i7 over an i5 or even a similarly clocked i3 would be kinda under the placebo effect and/or trying to qualify his higher investment.

 

That said, I would go for an i7 - probably more than half way there if you opt out of the 4GB and get the 2GB 760 or just go with the 7950.

Link to comment
Share on other sites

I'd go with a decent i7 quad core - whatever fits the budget - with 16GB, and a reasonable GPU like a midrange Radeon, and run the programs in Direct3D mode. Also, since this is for a student, he can upgrade the Autodesk programs at no cost by signing up at students.autodesk.com. A recent version of 3DSMax or 3DSMax Design (which replaces Viz) will do quite well on a midrange consumer video card.

Link to comment
Share on other sites

Thanks for all the help! To clarify, gaming performance is absolutely not a consideration for this workstation. This will be for architectural work only. I'm also trying to be fairly budget conscious, so while the target is to spend $250 on the GPU, any savings would be completely welcome. Conversely, if it really makes sense, we can spend $100-$150 more (probably at the expense of motherboard / air cooler / SSD) - really loathe to part with the SSD though. But the key theme here is perf / price. My friend is also extremely concerned about rendering performance, and is willing to sacrifice viewport fps for that (although that is desired as well).

 

I would like to clarify a few more points:

 

1. Is rendering on the above programs a pure CPU task? I was under the impression that much of it could be hardware accelerated via OpenGL or CUDA. If that its only CPU bound, is there any point to getting anything other than a cheap Quadro 600? What would I gain or lose, say, from a midrange Radeon 7xxx series vs a Quadro 600 for the same price? Or, should I go with a relatively beefy gaming GPU in the $200+ range, even at the expense of losing viewport performance, in case some applications need it in future?

 

2. Any benefit to going with an Nvidia card, (say, gtx 760 2gb) over a AMD 7950? The impression I get is that the Nvidia cards are some what favored by professionals, probably from their historical reputation of better drivers and of course for CUDA support. Is this important enough to pay more for?

 

3. Based on microcenter bundles, going from an i5-4670 to a i7-4770 is going to cost me (him) $80. Is this money well spent? Or should it be spent on a GPU?

 

4. Is it a good idea to look into the Firepro W5000 that have a promotion now for ~50% off? I can get it for slightly over $200.

 

Basically, the options I'm looking at are right now:

 

1. Quadro 600 for $150.

2. Radeon 7870 or lower for $100-$150

3. Radeon 7950 for $200

4. Firepro W5000 for $210 or so

5. gtx 760 2gb for $250 or so

6. Dropping down to an i5, giving up SSD, and getting a $400 Quadro K2000.

7. Anything else you'd suggest?

 

What is your opinions on this?

Link to comment
Share on other sites

Rendering on CPU vs. GPU - this depends on the software he's using. There are a few programs that use CUDA (e.g. Arion, mental ray's iray) and there's Vray RT-GPU which lets the user choose CUDA or OpenCL. But the normal modes of Vray and mental ray, which is what most people are using for their production renders, are CPU only.

 

So if he's not going to be using one of those CUDA/OpenCL renderers, it's best to put money in the CPU and not the GPU. For non-GPU-rendering use, look at video cards that are around $150 - IMO the 7870 is more than enough.

 

If he does want to use one of the GPU renderers (which, mind you, aren't "better" or anything, just different as a matter of taste) an nVidia card with a good amount of memory is best. Don't bother with Quadros, go with a decent Geforce.

Link to comment
Share on other sites

1. Renderers are predominatly CPU bound.

GPUs have nothing to do with it once the modeling phase is done.

 

Some renderers are evolving in using CUDA and/or OpenCL (not openGL), and offer progressive GPU accelerated (actually GPU computed, it is not like CPU is working hard and GPU "Adds" to it, if anything it is the opposite) rendering capabilities.

Other applications - like Adobe CC suite - offer OpenCL accelerated tasks in Photoshop, Premiere etc.

 

 

That said, most production work is still revolving around CPU rendering, with GPU renderers acting more as a mean for "previewing" the final result, setting up lighting and materials etc. Very powerful tool, but not replacing CPU requirements.

 

2. nVidia has a vastly bigger market share, spends more money on marketing and had CUDA developed and pitched to developers with proper support for far longer than OpenCL. Thus, for certain GPU accelerated tasks written around CUDA, nVidia is the "only way".

As far as viewports go, AMD has taken big steps in improving driver support for most programs, and its doing quite well.

nVidia drivers for GTX cards are pretty weak in the OpenGL API (Maya, Rhino, Sketchup, Solidworks, Cinema 4D), thus as far as gaming cards go, most mid-range and upwards Radeons will do better than a GTX of similar or even higher class.

 

For apps that actually are properly written for OpenCL (not that many atm, just Adobe CC Suite I think), the GCN architecture of the higher end 79xx Radeons (and Wxxxx Firepro equivalents) is at least a generation ahead from GTX cards - i.e. the 7950 is something like 40% faster than a GTX Titan. For properly translated apps that used to be CUDA (like VRay RT), AMD cards are performing as if they are 2-3 generations behind, and for CUDA only applications are ofc useless as accelerators.

 

3. i7 is money well spend for whoever deals with any short of rendering tasks.

 

4. Don't know it is still going. Surely it is a great value for the W5000.

 

Now, from the listed cards:

Quadro 600/K600 - both are decent for the price. I would get the K600 if available. Should easily do better than any GTX in most OpenGL apps listed above.

7870: decent all around card. Will do great in D3D and ok in OpenGL

7950: supersized 7870, with vastly better OpenCL compute (when available/compatible). Viewport performance will be more or less the same.

W5000: great viewport all around, much better compute than the K2000 it competes against.

760: "ok" for all around viewports. Might trade blows with the 7950, being slightly faster in D3D and slightly slower in OpenGL. Big deal = opens the way for CUDA accelerated apps and GPGPU renderers like the VRay RT GPU (great with nVidia cards, sucks with AMD) or iRay & Octane (CUDA = nvidia only).

K2000: decent mid-range Quadro card. The minimum if you were seriously into heavy models from Maya or Solidworks (W5000 ofc plays in this league). GPGPU performance is pathetic tho...no accelerated "anything" worth writing about from this card (considering the price point ofc).

 

VERY important point: all this talk about cards, refers to COMPLEX models. REALLY complex models. Most architectural students won't be miles close to the 10ths of millions of polygons that actually help seperate the "strong from the weak". They might get there before they figure out proxies for trees and whatnot, but that's it. Most architecture studenst I know, "manage" to work with laptops. Some of them manage to do it with MacBooks - i.e. not even "mid-range" mobile GPUs, against which a Quadro K600 should rate as "amazing".

There is a lot of "the tools make the man" mentality around the CG world, which ofc is false: getting a quadro "cause I need it" is more often false than true, more of a "want" than a "need", just like getting a Porsche or whichever luxury car.

 

EDIT: SSDs are fun, but of VERY little value imho. Many say that it is impossible to do without etc, but who does use them in the professional world outside large scale corporate servers and small CG offices? Fast CPU and lots of RAM is vastly more important than an SSD.

Edited by dtolios
Link to comment
Share on other sites

I think I'm not understanding something - I had previously assumed that besides rendering and viewport performance, there wasn't much GPU usage. And therefore if a program does not use GPU rendering, there are no GPU compute tasks? Is this wrong? If the Quadro K600 is superior to gaming cards for viewport, is there any point in the extra compute from the gaming cards, assuming I am mostly using the CPU to render?

 

Basically for roughly the same price - what am I getting out from the 7950 that I won't get in a Quadro K600?

Link to comment
Share on other sites

More and more apps are trying to tap into the great performance gains of GPGPU. Too many wrongfully claimed for too long that their expensive gpu was accelerating tasks outside viewport and games, but soon that might be the norm for serious professional apps.

 

Personally I would go for a 7870 or 7950 over a k600, for the all around flexibility and the gpgpu potential.

Link to comment
Share on other sites

Any recommendation on a cheapish Nvidia card to keep the CUDA angle open? Or do you think the 7950 is too good price/perf for now?

 

So far the system looks like this, for almost exactly $1300 total:

 

i7-4770k CPU

ASRock Z87 Extreme4 Mobo

Team 16gb (2x8gb) DDR3 1600 RAM

Powercolor PCS+ 7950 GPU

Samsung 840 128gb SSD

Seagate Barracuda 7200RPM 1TB HDD

Corsair CX500M Bronze Modular PSU

Corsair Carbide 300R Case

Asus VS24AH-P 24" IPS monitor

Edited by xavierzhu
Link to comment
Share on other sites

For the prices 7950 is available at, your options are GK106 based GeForces...something ala GTX 660 2GB. Not a bad card, but nothing to write home about vs. a 7950 which is "2 tiers" up the foodchain as a gaming card and simply untouchable as a compute card (again, when they get it to work - but that's not a "AMD Driver" issue, but OpenCL coding issues).

 

As far as viewports in 3D CAD apps, I would expect the 7950 to be better, but I cannot judge on "how much".

I would expect either to be "ok" for the most part, and when things get too complicated, both would be equally mediocre to "bad". It is this point where I don't think anything within your budget would do much better: gaming cards have crippled drivers and lots of raw power, affordable Quadros/Firepros have low raw power and good drivers, so when things bring gaming cards to their knees, "workstation" cards are not that happy either...just "better".

 

But again, I think industrial design and mech engineering projects / apps get there faster than an architecture student will probably get.

At least for school. If at his spare time he goes crazy with CG, good, maybe, but even in this forums we have tons of artists that do professional, high quality work using GTX 5xx/6xx cards for their viewports.

Link to comment
Share on other sites

Honestly for a student I think this is pretty well price/performance optimized. One thing I usually recommend to people who are going to be heavy CUDA rendering users is to keep one video card that's not for CUDA that runs the display and another that runs the CUDA work - because CUDA renderers are GPU intensive over sustained periods and will slow all other operations to a crawl as they compete for GPU time to refresh their windows. So, so long as there's enough PSU and case capacity to support a second card that could be left to a later upgrade. (This might require a larger PSU but I can't do that math right now.)

Link to comment
Share on other sites

Honestly for a student I think this is pretty well price/performance optimized. One thing I usually recommend to people who are going to be heavy CUDA rendering users is to keep one video card that's not for CUDA that runs the display and another that runs the CUDA work - because CUDA renderers are GPU intensive over sustained periods and will slow all other operations to a crawl as they compete for GPU time to refresh their windows. So, so long as there's enough PSU and case capacity to support a second card that could be left to a later upgrade. (This might require a larger PSU but I can't do that math right now.)

 

Yes, the truth is the 500W PSU is "pushing" it with an i7 and a 760 or 7950 class GPU, unless we are talking a VERY good quality 80+ Gold or so unit, that can actually handle being loaded above 60-70% for prolonged times. Corsair CX line is not bad, but since the 7950+i7+rest of stuff combo can easily spike above 400W, I would get either a larger CX, or a better rated unit (which actually might end up being more expensive than a CX650) if I was expecting heavy compute workloads and/or 2nd GPUs - even if it was a 50W Quadro K600. Just to be safe.

 

In general getting those 400W would require CPU+GPU working @ 100% simultaneously which is very rare for CG workflows (unless you do GPU+CPU progressive renderings). General renderings with CPU only would probably taxi in the

 

So as long as we are talking CPU renderings, casual modeling and/or casual gaming, where CPU and GPU are not @ 100% all the time, a cheap yet decent quality 500W PSU like the CX does the trick just fine. If your actual workflow will be pushing your PSU above 50-60% of its rated capacity for prolonged times, then make sure you get the best possible quality, or upgrade to a higher capacity decent quality PSU.

Edited by dtolios
Link to comment
Share on other sites

There is a AMD 7850 now available for $110:

 

http://www.newegg.com/Product/Product.aspx?Item=N82E16814127727&nm_mc=AFC-C8Junction&cm_mmc=AFC-C8Junction-_-na-_-na-_-na&cm_sp=&AID=10440897&PID=3332167&SID=u00000687

 

At this price, do you think I would be better served in stepping down from the 7950? (~$90) difference. The gamer in me wants all the hardware I can spec out, but I want to be very conscious of throwing money at limited returns, especially if he might need to add a CUDA card say 1-2 years from now.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

×
×
  • Create New...