How Well Do Workstation Graphics Cards Play Games?

[DuncThePunk]

Thanks for this very comprehensive and helpful review. May I make one suggestion/request please? OpenGL is a very valuable performance capability in Gaming but seems to be always overlooked in benchmarking these days. Seeing that the Unigine benchmark has the ability to run in OpenGL would it be possible for Tom's to test all the cards using their OpenGL rasterisers to see what their relative performance is in this regard. A lot of (older) games still use and may use OpenGL in the future, it would be great if you have the time, to run these bench's and add an extra page of OpenGL performance to this review to make it complete. I hope you guys might be able to do it and I think it would make for interesting reading. Many thanks for your efforts.

 

[RLS2013]

Thanks for the review, Igor. It was interesting. This may be one of those if you have to ask, your case isn't big enough situations, but since you are recommending the W7000, just how long is it? The AMD web site calls it a half length card, but from your pictures it appears to be only slightly shorter than the W8000 or W9000, plus it looks like it overhangs the front of the motherboard in the test bench. Also, it looks like the W7000 exhausts into the case while the W8000 and W9000 blow out the back. Any thoughts on the difference this would make on the cooling requirements for the case, or is this offset by the reduced power consumption mentioned in the review? Thanks.

 

[JCnator987]

The i7 2600k has 8mb of l3 cache, not 6.

 

[trimarky]

Hi i liked your article, i want to buy a good workstation video card to work in 3D and render, but i can't find any comparition with the new kepler quadro edition vs the W series of AMD.

 

[FormatC]

This will be one of my next tests. I'm just benchmarking K5000 and K4000 against older Fermi Quadros, the FirePro W and V cards with a big bechmark suite (OpenGL 2D/3D, CUDA, OpenCL etc.)

 

[cravin]

I'd like to see how these 'workstation' cards compare to nvidias normal gtx cards in CUDA programs, like adobe cs6.

 

[blazorthon]

I'd like to see how these 'workstation' cards compare to nvidias normal gtx cards in CUDA programs, like adobe cs6.

For anything that involved dual-precision work, the normal GTX cards from the GTX 600 series will be utter junk. Radeon 7700/800 won't be too much better, but GTX Titan and Radeon 7900 have a good shot at doing well in many compute workloads.

 

[RaQin]

The NEXT question is: How Well Do Gaming GPU perform as a Workstation Graphics Cards?

It would be nice to see that a GTX680 or appropriate AMD card could go toe-to-toe with a serious Workstation card like the W9000. Imagine the savings on workstations if gaming cards were able to be used instead.

 

[FormatC]

The NEXT question is: How Well Do Gaming GPU perform as a Workstation Graphics Cards?

This test is ready for translation.

BTW:
In the next weeks I will finish the new workstation charts with a very large benchmark suite, 14 current workstation cards include the new Kepler Quadros and also a handful of consumer cards.

 

[Wilf Tarquin]

These cards are expensive because they DO NOT HAVE CRIPPLED DOUBLE-PRECISION MATH.

That's all. They are two generations old reference cards, underclocked to avoid heat problems when running 24/7, with slow ECS memory. My work machine has double Quadra 4000, they're slower than heck, any entry-level graphics card will run circles around them at everything EXCEPT double-precision math, and then only because all entry-level cards are heavily crippled.

It's a boondoggle, but if you need double precision math there's nothing you can do about it.

 

[khaledegy200]

The GT430 beats GT440

 

[El_Capitan]

These cards are expensive because they DO NOT HAVE CRIPPLED DOUBLE-PRECISION MATH.

That's all. They are two generations old reference cards, underclocked to avoid heat problems when running 24/7, with slow ECS memory. My work machine has double Quadra 4000, they're slower than heck, any entry-level graphics card will run circles around them at everything EXCEPT double-precision math, and then only because all entry-level cards are heavily crippled.

It's a boondoggle, but if you need double precision math there's nothing you can do about it.

Just get a HD 7970 and use applications using OpenCL API's. That's what you can do about it.

 

[Weak1ings]

Why are there two GTX 660 Ti that have different scores on the lists?

 

[folem]

Hardware failing can cause errors too you know. Even perfect software/drivers can't save you from that, but at least it'll tell you about the error rather than just silently ignore it. Also, gaming cards may not be as accurate as workstation cards since gaming cards don't usually come with double precision enabled in hardware. Even if the driver supports double precision it won't do you any good if it's not present on the hardware itself.

Cards having double precision disabled/limited is not limited to gaming cards. In fact all Kepler cards, GeForce and Quadro alike, have double precision disabled/not present excepts for the Titan, K6000, and Tesla K20.

As far as I know, the only place AMD uses a similar tactic is in the Tahiti Radeon cards (I'm not familiar with the R9 series yet so I am excluding those). The FirePro W7000 and Radeon 7870, the higher Pitcairn cards, get very similar double precision performance as expected since they are almost the same card. Comparing the 7950 and W8000, the lower Tahiti cards there is a drastic discrepancy. The 7950 scales as expected from the 7870, but the difference between the W8000 and W7000 is about 4 times what would be expected. This is because Tahiti chips have 4 times the the double precision compute power per compute unit compared to the Pitcairn chips, but this feature is disabled on the Radeon Tahiti chips (there are 20 compute units on the Pitcairn chips and 24 on the Tahiti).

 

[folem]

Hardware failing can cause errors too you know. Even perfect software/drivers can't save you from that, but at least it'll tell you about the error rather than just silently ignore it. Also, gaming cards may not be as accurate as workstation cards since gaming cards don't usually come with double precision enabled in hardware. Even if the driver supports double precision it won't do you any good if it's not present on the hardware itself.

Cards having double precision disabled/limited is not limited to gaming cards. In fact all Kepler cards, GeForce and Quadro alike, have double precision disabled/not present excepts for the Titan, K6000, and Tesla K20.

As far as I know, the only place AMD uses a similar tactic is in the Tahiti Radeon cards (I'm not familiar with the R9 series yet so I am excluding those). The FirePro W7000 and Radeon 7870, the higher Pitcairn cards, get very similar double precision performance as expected since they are almost the same card. Comparing the 7950 and W8000, the lower Tahiti cards there is a drastic discrepancy. The 7950 scales as expected from the 7870, but the difference between the W8000 and W7000 is about 4 times what would be expected. This is because Tahiti chips have 4 times the the double precision compute power per compute unit compared to the Pitcairn chips, but this feature is disabled on the Radeon Tahiti chips (there are 20 compute units on the Pitcairn chips and 24 on the Tahiti).

 

[Yokoko44]

So, are these guys implying that if I were to get the latest Mac Pro and upgrade it a little (thinking about getting the $4,000 one and getting the fire pro D700's, I would be able to play my games if I BOOTCAMPED my mac?