Quadro vs Geforce

[Thor]

Hi, I am hoping some of you bright minds can help me.


How come the Quadro cards are ridiculously underspec compared to GTX, but for some reason they still outperform them in design.

This made me think:

All nVidias cards use the same chips, but as far as I can tell nVidia has *severely* crippled the consumer line GTX cards in order to sell Quadro cards, but the Quadros are _also_ crippled in order to sell Tesla cards. nVidia's wettest dream.

The Quadro cards are clearly very weak, but still probably manage to outperform the *far* more capable Geforce hardware with intentionally crippled firmware and drivers.

This seems to be what all the "Quadros suck at games" and "get a Quadro for graphics work" is really about: the Quadros are slow as heck, but their maths capabilities less crippled than the faster Geforce's. nVidia is abusing the market because it has no serious competition.

 

[Fulcrum29]

It had been discussed here:

http://mybroadband.co.za/vb/showthread.php/676669-Little-system-I-built-last-week

 

[Thor]

After reading that my suspicion was correct.


Nvidia deliberately cripple their cards on a firmware level.

 

[LCBXX]

[video=youtube;U5Npt1BSF04]http://www.youtube.com/watch?v=U5Npt1BSF04[/video]

 

[Fulcrum29]

After reading that my suspicion was correct.


Nvidia deliberately cripple their cards on a firmware level.

Where do they cripple the GTX cards, in what exact essence? Your one quote:

The Quadro cards are clearly very weak, but still probably manage to outperform the *far* more capable Geforce hardware with intentionally crippled firmware and drivers.

Crippled drivers? Really? The price coupled to Quadro cards are essentially driver and application support. Besides the OS drivers, CAD, Adobe etc. all have their own Quadro driver add-ins to utilise the VRAM, CUDA etc optimally.

See the Linus video, as posted by CataclysmZA in the other thread.

[video=youtube_share;U5Npt1BSF04]http://youtu.be/U5Npt1BSF04[/video]

 

[Thor]

If you program the Better Specced GTX cards they should outform the weaker spec quadro

 

[Fulcrum29]

If you program the Better Specced GTX cards they should outform the weaker spec quadro

Then you will need to add driver and application support to that better spec'ed GTX cards, beside the 3D gaming orientated drivers. Then you need to add more VRAM, additional module integration, then you will end up with an even more expensive better spec'ed GTX card than the weaker spec'ed Quadro card.

 

[furpile]

If you program the Better Specced GTX cards they should outform the weaker spec quadro

We use Quadro cards at work for CAD. One of the main reasons for using these cards is application support, as mentioned by Fulcrum29. The big software companies certify certain hardware platforms that will work with their software. That is not to say it's the only setup that works, but these have been tested. These high end CAD packages and other graphic software can also be very temperamental, so even using the wrong driver version can cause a lot of problems.

For example, PTC have a list of workstations that are approved for CREO (CAD package). These include DELL Precision workstations, which use only Quadro and Firepro cards. From their website:

Certified hardware
PTC certifies complete configurations (processor, graphics, operating system, etc.) provided by PTC's enterprise platform partners. We do not certify graphics cards by themselves. A certified configuration is one that PTC has thoroughly tested, specifically for the current release.

So you can use a GTX, but if you have any issues you are on your own.

 

[Chevron]

Where do they cripple the GTX cards, in what exact essence? Your one quote:



Crippled drivers? Really? The price coupled to Quadro cards are essentially driver and application support. Besides the OS drivers, CAD, Adobe etc. all have their own Quadro driver add-ins to utilise the VRAM, CUDA etc optimally.

See the Linus video, as posted by CataclysmZA in the other thread.

[video=youtube_share;U5Npt1BSF04]http://youtu.be/U5Npt1BSF04[/video]

Nvidia cripple fp64 on GTX for one. Anything double precision actually.

 

[Maverick Jester]

I don't buy the idea of the cards being "crippled", as more a case of them being purposed differently. That the underlying hardware is pretty similar is irrelevant if the use cases vary.

A one size fits all approach doesn't work for specialist applications.

 

[Chevron]

I don't buy the idea of the cards being "crippled", as more a case of them being purposed differently. That they underlying hardware is pretty similar is irrelevant if the use cases vary.

A one size fits all approach doesn't work for specialist applications.

The fact the underlying hardware is similar is everything. It means they could give the awesome software performance of the one, to the other, but choose not to.

That's crippling.

 

[Maverick Jester]

The fact the underlying hardware is similar is everything. It means they could give the awesome software performance of the one, to the other, but choose not to.

That's crippling.

Or differentiating. Much like vehicle manufacturers give different models of the same underlying vehicle.

 

[Rickster]

Quadro cards will do well with animation but Geforce is better optimised for games and vice versa.

 

[Fulcrum29]

Nvidia cripple fp64 on GTX for one. Anything double precision actually.

The fact the underlying hardware is similar is everything. It means they could give the awesome software performance of the one, to the other, but choose not to.

That's crippling.

Yes, but do gaming need double precision at its highest capacity? Then also to do computing at the required rate, double precision essentially provides longer calculations (coupled to error correcting technology), more VRAM, larger compute sizes, herein where CAD and modelling applications take advantage.

Nvidia has three cards designed to do computing above graphics, 1) GTX Titan, 2) Quadro and 3) Tesla.

Video games rely on FP32, very limited FP64 hardware resources are utilised. It is also stated in NVIDIA’s white paper that the reduced FP64 hardware resources convey power and FP32 computing more optimally, thus the blocks are better utilised to suit gaming than computing.

Then again, the Quadro pricing allows on-going driver and application support. I really cannot see where they are crippling the card when the purposed FP64 limitations are applied to enhance a user's gaming experience and given optimal system utilisation.

 

[CataclysmZA]

Nvidia has actually been lasering off or completely removing FP64 units in their yields and architectures, so even installing Quadro drivers for most Geforce cards won't automagically improve performance. That only happens for the GTX Titan, Titan Black and Titan-Z.

No game that I know of uses FP64. Also, because Nvidia doesn't want consumers to buy Geforce cards to try use them for professional stuff, they've been crippling double precision ever since the GTX 580.

 

[Chevron]

Yes, but do gaming need double precision at its highest capacity? Then also to do computing at the required rate, double precision essentially provides longer calculations (coupled to error correcting technology), more VRAM, larger compute sizes, herein where CAD and modelling applications take advantage.

Nvidia has three cards designed to do computing above graphics, 1) GTX Titan, 2) Quadro and 3) Tesla.

Video games rely on FP32, very limited FP64 hardware resources are utilised. It is also stated in NVIDIA’s white paper that the reduced FP64 hardware resources convey power and FP32 computing more optimally, thus the blocks are better utilised to suit gaming than computing.

Then again, the Quadro pricing allows on-going driver and application support. I really cannot see where they are crippling the card when the purposed FP64 limitations are applied to enhance a user's gaming experience and given optimal system utilisation.

CataclysmZA has it right. Having good fp64 doesn't magically make your fp32 worse.

They just don't want you using cheap gaming cards instead of paying more for quadro. End of story.

 

[oldhat]

Here's an interesting hack:
http://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/

 

[TJ99]

NVidia is capable of anything, as I've recently been reminded of.

Huang's speech about the ****ed VRAM on the GTX 970 being a "feature" is the most dishonest, backpedalingest, spinniest response I've ever heard out of any tech company that isn't Apple.

 

[Fulcrum29]

CataclysmZA has it right. Having good fp64 doesn't magically make your fp32 worse.

They just don't want you using cheap gaming cards instead of paying more for quadro. End of story.

I said:

ideo games rely on FP32, very limited FP64 hardware resources are utilised. It is also stated in NVIDIA’s white paper that the reduced FP64 hardware resources convey power and FP32 computing more optimally, thus the blocks are better utilised to suit gaming than computing.

As with this article:

http://wccftech.com/nvidia-gm200-gpu-fp64-performance/

Snipped:

What Nvidia’s GM200 Weak Double Precision Performance Could Mean For Pro Graphics

Before going into the ramifications of this potential decision by Nvidia we must remind you again that we couldn’t verify this report by 3DCenter ourselves and thus will treat it as a rumor for the time being. Now that we got that out of the way, to fully understand what this development means for consumers we must understand how FP64 compute has traditionally been added and why it’s important.

You can deduce the difference between double precision floating point (FP64) and single precision floating point (FP32) from the name. FP64 results are significantly more precise than FP32. This added precision in the results is crucial for scientific research, professional applications and servers. And less so in video games. Even though FP64 is used in games in a very limited subset of functions, the bulk of video game and graphics code relies on FP32. As such this added precision in turn requires more capable hardware which would net higher costs by increasing the size of the chip while simultaneously increasing power consumption.



Double precision (FP64) compute performance has always been lower than single precision (FP32) in GPUs for that reason. Normally there’s a fixed ratio between the peak single and double precision floating point capability of a given GPU. This ratio varies between different GPU architectures and different GPUs within the same architecture as well. In the latest enthusiast class chip from Nvidia the ratio between FP32 and FP64 peak performance sits at 3:1. This is true for Nvidia’s GK110 GPU which powers the Quadro K6000, Titan and GTX 780/780 Ti graphics cards among others. Although the ratio has been artificially restricted on the 780 and 780 Ti cards to 16:1.

For AMD the ratio is a more aggressive 2:1 in its latest enthusiast class GPU Hawaii which powers the company’s flagship FirePro W9100 and Radeon R9 290 series products. Although the ratio is artificially restricted in the 290 series to 8:1.

So, since the GTX Titan Black has a peak of 5.1 TFLOPS single precision floating point performance, a 3:1 ratio means that double precision compute goes down to 1.7 TFLOPs. And with AMD’s Hawaii XT which has a peak of 5.6 TFLOPs of FP32 compute performance, a 2:1 ratio means that it will go down to a more respectable 2.8 TFLOPs of FP64 compute performance. This advantage in FP64 compute is why AMD succeeded in capturing the top spot in the Green500 list of the world’s most power efficient supercomputers with it’s Hawaii XT powered FirePro S9150 server graphics cards.

The FP32 to FP64 ratio in Nvidia’s GM204 and GM206 Maxwell GPUs, powering the GTX 980, 970 and 960 is 32:1. Which means the GPU will be 32 times slower when dealing with FP64 intensive operations compared to FP32. As we’ve discussed above this is mostly OK for video games but downright unacceptable for professional applications.

If Nvidia’s GM200 does end up with a similarly weak double precision compute capablity the card will have very limited uses in the professional market. However in theory the reduction of FP64 hardware resources on the chip should make it more power efficient in games and FP32 compute work. Even though I’m not entirely convinced that it’s a worthwhile trade off. Especially for a card that is poised to go into the next generation Qaudro flagship compute cards.

Note the article is related to FP64 in the GM200. Yes, the author isn't convinced with the FP64/FP32 ratio in the GM200 as an acceptable 'cripple' as you and others stated. I didn't post the entire article and only snipped the middle, but we will have wait and see the FP64/FP32 ratio in the upcoming pro series, GFX Titan, Quadro and Tesla.

Yes, the standard GFX cards can be modded to ID as a Quadro, Tesla or other enthusiast card, and assume its role. However the modded card isn't a 1:1 result. As I also previously said, the pricing is to allow on-going driver and application support.

 

[CataclysmZA]

They just don't want you using cheap gaming cards instead of paying more for quadro. End of story.

They also want to avoid idiots using a Geforce with Geforce drivers on any kind of mission-critical project that needs drivers that are 100% stable for particular workloads.

Here's an interesting hack:
http://www.eevblog.com/forum/chat/hacking-nvidia-cards-into-their-professional-counterparts/

It definitely works, but it only changes the device IDs, so the performance of the cards are still dismal compared to their Quadro counterparts. These days FP64 units are soldered off and no amount of BIOS hacks will bring them back. I remember flashing a Geforce 6200LE into a 6600GT, which was awesome.

Tom's Hardware did an interesting experiment a while back, attempting to change a HD6570 into a FirePro V3900. They didn't succeed, but they could have done it right with a few soldering mods on the back. Still, the cards are identical down to the last transistor in the core, so the article turns into a comparison of Radeon and FirePro drivers.

Huang's speech about the ****ed VRAM on the GTX 970 being a "feature" is the most dishonest, backpedalingest, spinniest response I've ever heard out of any tech company that isn't Apple.

Well, it is a feature of Maxwell. Plus the card still goes blisteringly fast and better drivers will solve any lingering issues with prioritising access to that second pool of memory. They made a huge stuff-up, agreed, but it is very out of character for the company to try hide the details of how Maxwell works with memory and L2 cache these days.

Let it go, man.