-
Workstation CAD video cards... marketting hype?
Posted by
8bitgasm
on 05 Apr, 2018 21:37
-
I currently run the AMD FirePro W4300 in my small form factor Dell Optiplex.... sure it's a sweet card and it can run 3440x1440 CAD nicely. But I can't help but believe I should have instead got a low profile Nvidia GeForce GTX 1050 TI instead and unlocked some gaming potential as well.
I struggle to believe the GeForce woudln't be able to smoke the FirePro at CAD work. Both are 4GB of DDR5, and the GeForce significantly outperforms the W4300 on paper in every category... Mhz, Pipeline, Transistors, etc.
Are CAD workstation cards just a marketing buzzword for office environments to sink money into? I remember reading somewhere on the AMD site that AutoDesk had optimized their software to their workstation video cards, but I just can't bring myself to really believe that; and even if they did, that a gaming card wouldn't still outperform this "optimization".
What do people think?
-
#1 Reply
Posted by
Ampera
on 05 Apr, 2018 21:58
-
Most workstation cards are really only suited to some specific situations over consumer cards. This does actually remind me of the IBM PGC which was IBM's attempt to compete with the high end RISC machines in the CAD market. I think only a few programs had any real support for it, but it had an entire 8088 on it to update the screen faster than the PC bus could.
-
#2 Reply
Posted by
TimNJ
on 05 Apr, 2018 22:29
-
For the most part,
Workstation cards (FirePro, Quadro) are good for high precision and accuracy applications, like 3D modeling/drafting, complex simulations, biological modeling, etc. In these types of "productive" applications, people want accurate results most of all.
Gaming cards (Radeon, GeForce) are good for quickly pumping out "rough" geometries. We think of new videos games as being high quality, but in reality, the polygon count is still pretty low.
Workstation cards might have error correcting (ECC) memory and additional high (double) precision floating point units. Those type of features are a waste in a fast paced video game, but not when you need to render something much more complex.
-
#3 Reply
Posted by
Bicurico
on 05 Apr, 2018 23:22
-
We prefer nVidia GeForce GTX 1060/1070/1080 for our CAD/CAM software over any nVidia Quadro.
The reasons:
1) GeForce GTX is faster
2) GeForce GTX is cheaper
3) GeForce GTX is more stable - no issues with our software
With nVidia Quadra we had a 100% ratio of problems with customers.
Our software is premium CAD/CAM software.
The nVidia Quadra are only recommended for users of specialized CAD software, where the manufacturer explicitly certified a particular model and driver. You will then get that exact Quadro, otherwise you won't get support if the software crashes. And perhaps a specific piece of software might benefit from given extra features of a Quadro card. That's about the only reasons.
Just follow the recommendation of your CAD/CAM vendor.
If you don't have a CAD/CAM vendor, then you are NOT using a premium CAD/CAM software that requires a particular graphics card in order to get support, so you are better off with a regular GTX card.
I would invest the extra money in at least an 1060, instead of an 1050.
If you are considering GAMES, then by no means get any Quadra no matter how cheap you can get them - they will suck compared to GTX cards!
Regards,
Vitor
-
#4 Reply
Posted by
Halcyon
on 06 Apr, 2018 00:55
-
Just on AMD vs NVIDIA cards in general. At work we use quite a few of them and not too long ago when we used primarily AMD cards we had a much higher failure rate than the NVIDIA counterparts. I think the AMDs just run too hot. It's like have a space heater in your PC.
-
#5 Reply
Posted by
Ampera
on 06 Apr, 2018 02:38
-
I'll agree with you for once there, Halcyon, AMD graphic cards are cheap workhorses, but little else, and VEGA is the worst yet.
-
#6 Reply
Posted by
NiHaoMike
on 06 Apr, 2018 02:50
-
I think GPUs specifically designed for GPGPU use are built with better power and cooling to better handle full load 24/7. The price premium is high enough that you could just buy multiple "standard" GPUs, so it would only make real sense if downtime is expensive.
-
#7 Reply
Posted by
Halcyon
on 06 Apr, 2018 03:07
-
I think GPUs specifically designed for GPGPU use are built with better power and cooling to better handle full load 24/7. The price premium is high enough that you could just buy multiple "standard" GPUs, so it would only make real sense if downtime is expensive.
Not exactly. Nvidia prohibits the use of GeForce in data centers except for clock chain applications by putting this restriction in driver software EULA. As a small player, nobody cares. But bigger companies may care about software license.
You'll find even the large players don't care. A EULA isn't worth the paper it's printed on (so to speak) unless it's enforced in court, which is extremely unlikely to ever happen. Even if NVIDIA suspected that their cards were being used in a data centre, how could they prove it? It would be as simple as turning them away at the door saying "You aren't allowed in here".
Large company, Government or Individual, once you purchase hardware/software, it's free to do whatever you want with it (within the law) and I suspect any court would find that "reasonable". Imagine you bought a car but the manufacturer stipulated where you could or could not drive it, you would tell them where to go.
-
#8 Reply
Posted by
NiHaoMike
on 06 Apr, 2018 03:20
-
Also, PCB design of Tesla and Quadro are 100% identical to official GeForce reference design for many generations.
The only differences are in license terms, ASIC quality (Quadro and Tesla usually can run slightly faster than stock speed of GeForce and also run cooler) and enabled features, as well as after sale service.
The PCB design might be the same, but what about the components on it? I have seen some GPUs that left out a few parts in the power supply section to save on cost.
-
#9 Reply
Posted by
Ampera
on 06 Apr, 2018 03:26
-
Not to mention, considering most of these places run Linux, there are already free and open source drivers available for most of these cards. So if someone was really afraid of nVidia bringing them to civil court, they could hide behind the fact that they weren't breaking any license agreements.
-
#10 Reply
Posted by
Halcyon
on 06 Apr, 2018 03:43
-
You'll find even the large players don't care. A EULA isn't worth the paper it's printed on (so to speak) unless it's enforced in court, which is extremely unlikely to ever happen. Even if NVIDIA suspected that their cards were being used in a data centre, how could they prove it? It would be as simple as turning them away at the door saying "You aren't allowed in here".
They can simple refuse to offer any warranty, service or even future purchase orders for you if they discovered you have violated EULA.
I don't know if it's legal in Australia, but in China, a monopoly supplier can request all their distributors to boycott a buyer.
Now if you can't buy the cards, how can your keep your data center business running?
Those sorts of dodgy practices aren't legal in Australia at all. A company can write anything they like in their EULA, it doesn't make it legally binding. They also can't refuse warranty or override consumer rights (which are written into legislation). Legislation trumps anything a company says.
Some ridiculous examples of some things companies have put in their licensing agreements:
Apple iTunes:
"You also agree that you will not use these products for any purposes prohibited by United States law, including without limitation, the development, design, manufacture or production of missiles, or nuclear, chemical or biological weapons." I guess Kim Jong-Un isn't using Apple iTunes then?
Google Chrome:
"You give Google a perpetual, irrevocable, worldwide, royalty-free, and non-exclusive license to reproduce, adapt, modify, translate, publish, publicly perform, publicly display and distribute any Content which you submit, post or display on or through, the Services." (This one has since been removed because it was not considered fair. They also admitted this clause was copied and pasted from somewhere else and Google themselves didn't proofread it.
Microchip MPLAB X:
"Microchip’s authorized representatives will have the right to reasonably inspect, announced or unannounced and in its sole and absolute discretion, Licensee’s premises and to audit Licensee’s records and inventory of Licensee’s use of the Software, whether located on Licensee’s premises or elsewhere, at any time, in order to ensure Licensee’s adherence to the terms of this Agreement." Yeah, good luck enforcing that one Microchip!
Sony PlayStation Network:
Any Dispute Resolution Proceedings, whether in arbitration or court, will be conducted only on an individual basis and not in a class or representative action.I've even come across agreements more than once where the company has left in place holders such as [INSERT COMPANY NAME HERE] etc...
-
#11 Reply
Posted by
Circlotron
on 06 Apr, 2018 03:44
-
Imagine you bought a car but the manufacturer stipulated where you could or could not drive it,
Like they do with recorded music?
-
#12 Reply
Posted by
Ampera
on 06 Apr, 2018 03:44
-
Yeah I figured about much.
I never used it because I just used the regular drivers.
-
#13 Reply
Posted by
Halcyon
on 06 Apr, 2018 03:48
-
Imagine you bought a car but the manufacturer stipulated where you could or could not drive it,
Like they do with recorded music?
Yeah they try, but Australian law allows you to make a copy of music/audio "for domestic use" even if that copy is made into an entirely different format from the original (e.g.: CD to FLAC). You just can't make that copy available to the public or otherwise distribute it. If then you cease to own the original (such as if you sell it or give it away), the copy must be destroyed.
-
#14 Reply
Posted by
Ampera
on 06 Apr, 2018 03:50
-
Same in the US. You can try to stop us copying stuff, but there are few formats that can't be copied.
-
#15 Reply
Posted by
Halcyon
on 06 Apr, 2018 03:57
-
My entire point is, companies will say just about anything in fine print, contracts, EULAs, send you mean letters, etc... etc... but it's mostly all crap. Their job is to make money, not be your friend. People need to stop point-blank believing what these people and companies say, understand the laws in their country and assert their rights when it's appropriate.
You wouldn't believe anything that came out of the mouth of Bob and Frankie Roohparvar and their Batteriser/Batteroo/ReBoost product would you?
-
#16 Reply
Posted by
Dubbie
on 06 Apr, 2018 04:15
-
We long ago stopped using quadros at work.
We had endless trouble with them and they were stupidly expensive for their performance.
We switched to gamer cards and have never looked back. 1/2 the price and 2x the performance. What is not to like?
-
#17 Reply
Posted by
TimNJ
on 06 Apr, 2018 13:50
-
We long ago stopped using quadros at work.
We had endless trouble with them and they were stupidly expensive for their performance.
We switched to gamer cards and have never looked back. 1/2 the price and 2x the performance. What is not to like?
I believe that to be true for most people.
I think if you are asking the "workstation vs. gaming card" question, then you probably should just get a gaming card. Only a very small amount of specialized applications
really benefit from the workstation cards. Those applications are usually in science and engineering. Any "glitch" or inaccuracy in a computation is unacceptable if you are modeling/designing a part that human life may depend on. Likewise, if you are modeling a complex biological process, small computational errors can accumulate to much larger errors, which may be unacceptable.
Recently, I learned that another important difference is that workstation cards (and their drivers) are regression tested to ensure computational integrity across the entire line product line. That is, if Nvidia adds a new workstation card to their lineup and updates their drivers accordingly, they will back-test all hardware to ensure no one's critical models will be changed by updating drivers.
Again, for most people...this doesn't really matter.
-
#18 Reply
Posted by
8bitgasm
on 06 Apr, 2018 17:36
-
Wow what a flood of replies! Thanks everyone!
I see AMD ran tests on the w4300 using SPECviewperf 12.0.1 (pasted the benchmark from their website below)
FirePro W4300: catia-04 = 38.79, creo-01 = 43.31, snx-02 = 45.65, sw-03 = 64.86I would be curious if anybody was so kind to run this test on a Geforce 10XX series of card on a Intel i3 (3.3Ghz) LGA1155, 16GB ram, or comparable system, to see how those tests stacked in comparison!?
https://www.spec.org/gwpg/downloadindex.html
-
-
The run rules (at a min) have to be followed when benchmarking
https://www.spec.org/gwpg/gpc.static/current.html You can still get wildly different results depending on config. Also many, maybe all, of the tests are old enough that the relevance is getting kinda questionable, for example the solidworks test is based on sw 2013, creo is based on creo 2012 etc. Be aware it doesn't test for display accuracy/glitching either.
-
#20 Reply
Posted by
Zucca
on 09 Apr, 2018 15:48
-
-
#21 Reply
Posted by
technix
on 10 Apr, 2018 05:50
-
There are some Quadro/Radeon Pro (yes that is what FirePro is called now) exclusive features:
* High-efficiency double-precision calculation
* ECC memory, in case that you need to depend on your GPU not making errors with those double-precision calculations
* Extended multi-output features, for example output to 16x 4K monitors at the same time.
* Some exotic bus interface like NV-Link, not just PCIe.
For the headless Tesla/Radeon Instinct cards, there is no display output, but you might get GPU virtualization.
Otherwise, if compared architecture for architecture, core for core and clock for clock on the same workload, a workstation card and a gaming card performs identically. There are benchmarks of those matches pairs of cards to prove this by folks like LinusTechTips and RandomGamingHD.
-
#22 Reply
Posted by
thm_w
on 11 Apr, 2018 23:39
-
It all depends on the SW-CAD, but Even Solidworks can be hacked to run on gamer cards, at least this is what I read on the www.
Interesting...
https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2018-GPU-Comparison-Monster-Sized-Model-1120/
Interesting, but I suspect not all features can be hacked and clearly the right driver path is not being used so performance is quite poor. 1080ti performs half as well as the low end quadros.
In order to include a GeForce card in this test, I had to use a workaround to enable RealView - which is normally disabled unless you have a workstation-class graphics card. This is absolutely something I don't recommend doing if you are using SOLIDWORKS professionally, and the performance of GeForce cards in this situation is not something worth pursuing anyway (as you will soon see).
https://www.pugetsystems.com/labs/articles/SOLIDWORKS-2018-GPU-Comparison-What-Is-the-Meaning-of-This-1112/With the old Radeon cards you could simply install the modified firepro driver and get decent performance and all of the solidworks features. But this hasn't been possible for a while.
Will try this soon on a 1050 ti (they only tested the 1080 ti above).
-
#23 Reply
Posted by
stevelup
on 12 Apr, 2018 07:05
-
I don't know if it's still the case, but certainly at some point in time only the 'workstation' cards could render anti-aliased lines which meant you hit horrific performance problems viewing complex CAD models in wireframe mode.