Why would I ever buy an entry workstation card?
I was always skeptical about Quadro cards…I mean, what is so special about them? We all knew for ages now that Quadro’s are not that different from their GeForce siblings, in many cases sharing 100% identical hardware, yet selling for many times the price.
How can nVidia get away with this?In fact, even for basic workstation graphic cards like the nVidia Quadro 600, retailers ask you investing as much as you would for a much more powerful and current GeForce GTX, and configuring your Dell or HP workstation with a “professional” card, would set you back even more…outrageous when you look at the underwhelming specs. It was mediocre back in 2010, and certainly it doesn’t’ look any better now:
Based on a version of the GF108 40nm Fermi processor, the Quadro 600 is nearly identical with the GeForce GT 420. 92 Stream processors (also referred to as CUDA cores) with only 8 of them FP64 capable, 1024MB of non ECC 128bit DDR3 VRAM, and two display outputs: DisplayPort (DP), and Dual Link DVI (DVI-D).
Both of the above are desirable and often supported by professional grade monitors, but users that would prefer HDMI would have to opt for an adapter or a HDMI to DVI conversion cable. The DVI-D is welcome and a “must-have” for some 1440p or bigger panels. If bought retail, the Quadro 600 will also come with a handy DP to DVI adapter.
On paper the Quadro 600 is surely not a card capable for high computation performance. Unlike the top 3 Quadro cards of its generation (4000, 5000 & 6000) based on the “big-Fermi” GF100 architecture that offers good double precision computation, wide memory buses and of course much higher clocks across the board, the puny GF108 leaves you wondering: Could driver and BIOS optimizations alone help it rise to its name and price tag expectations?
This short review will focus on results from the benchmarks: SPEC Viewperf 11 benchmark suite, SPECapc for Maya® 2012 and Maxon Cinebench 11.5. As the database on the site gets updated with more graphic card models and the testing process refined with more up-to-date tests, the reviews will get updated.
The card selection was an experimentation on affordable workstation GPU selection, and of course YMMV.
The testbed system was an i7-3820 @ 4.5GHz, on the Asus P9X70 Pro with 4x4GB Samsung 30nm @ 2333. OS and apps were running on a Crucial m4 256MB @ 75% capacity.
The EVGA GTX 670 SC 4GB that lives in the above system, will be representing the current generation of gaming cards. At $400 is hardly a “cheap” solution. The SC (SuperClocked) model runs at 967MHz stock, while the reference GTX 670s are 915MHz core. The OC @ 1250MHz refers to the max boost speed, which is around 1110-1150MHz for reference cards under full load.
All cards were using the latest nVidia 310.90 drivers for both GeForce and Quadro, and were connected to two monitors – one 27″ 1440p, and one 22″ 1080p through DP or DVI-D, depending on the available ports for each card. Vsync was forced off through the driver settings.
Finally, my little laptop was thrown in with the beasts, to see what a mid-range, 3yo mobile GPU has to offer: Acer 7745G i7-720 @ 1.6GHz (2.8GHz boost), 12GB RAM @ 1066 and Mobility Radeon 5650 1GB. The GPU was driving the 900p laptop screen, and a 22″ 1080p screen through the available HDMI.
The OS was Win 7 64bit in all cases.
Many of the tests included in Viewperf 11 could seem outdated for some readers, yet I know of no other standardized benchmark suite that is based on real viewport engines. Tests have been run 3 times and results have been averaged out for each program.
The little Quadro offers considerable performance gains over the much more powerful GTX 670 on most engines included in Viewperf 11. In many cases, the performance advantage is two or even threefold that of the GTX 670 SC, which retails nearly three times more than the Quadro 600, and cannot even keep up with models based on 3 or 4 architecture generations behind, featuring a fraction of the available VRam.
Ensite appears to have the only OpenGL engine that can take advantage of the GeForce drivers, giving the edge to the GTX card. Note that it is not only the Quadro cards outperforming the mighty GTX 670: the little Radeon M does appear to have much better OpenGL drivers, theoretically outperforming the GTX despite the massive raw power difference on both the GPUs and the platforms those are based on.
Overclocking the already faster than the reference GTX 670 board, yields very small benefits. With the addition of the GTX 560Ti in the charts, we can see that the Kepler card did manage to surpass it on every single test, in some case for more than 20-30%. Of course these are not directly comparable cards, but it does help you shape you expectations. The current 310.90 drivers did not allow the 560Ti to work with the Catia benchmark within SPECviewperf 11. Re-installing the drivers did not help.
SPECapc for Maya® 2012
First of all, I have to mention that running the SPECapc for Maya 2012 is extremely tedious! Each session goes through each of the 5 scenes trough an animated sequence, testing recording fps switching between all 9 viewport modes, and it will repeat it 4 times. The average time of completion for each card was roughly 40-45min. The benchmark records individual test scores and computes composite scores for graphics and CPU performance, which are normalized against a reference system – a Dell Precision 690 with Xeon 5130 2GHz, NVIDIA Quadro FX 570 graphics card, and 16GB RAM with 4x4GB ECC DDR2 667MHz. The test defaults at 1920x1200p windowed mode in our 1440p screen*, and the “usable” viewport is 1600×896.
Maya 2012 offers Viewport 2.0, which does support optimized instruction sets to streamline the way the program utilizes modern CPUs. The GTX 670 SC felt well, though like all gaming cards, the performance suffers a lot in wireframe and projected vertices modes, while it improves a lot when switching to shaded modes. The GTX 560Ti cannot keep up with the 670, but the perfomance difference is by no mean as pronounced as between a GeForce and even the cheapest Quadros.
This is a real time viewport capture of a single run under SPECapc for Maya® 2012 and the Quadro 600.
If you think the viewport with the Quadro 600 is not that responsive, you can always compare it side to side with the GeForce GTX 670 and identical settings.
Also keep in mind that the 2012 Maya benchmark is focusing and measuring a single maximized perspective view, while the 2009 version that is included in SPEC Viewperf 11 is animating 2 or 3 different views at the same time. In wireframe mode for example, the Maya 2009 test does feel very slow, and this is reflected by the numbers the GeForce card is allowed to achieve.
The test also stresses the on-board VRam with large textures, reaching around 1450MB of usage in our system with the GTX 670 4GB, while all other cards were maxed out. Keep note that the FX 570 of the reference system that would get “1.0 composite score”, is identical with the FX 1700 (same core and clocks), but has 256MB of 128bit RAM vs. 512MB for the 1700. That, along with a processor clocked 2.5x more, allows the almost identical GPU to perform ~80% better.
* The laptop, lacking 1440p capabilities through its HDMI port, was tested on 1080p.
Maxon Cinebench R11.5
The OpenGL graphics card testing procedure uses a complex 3D scene depicting a car chase (created by Renderbaron using car models from Dosch Design) with which the performance of your graphics card in OpenGL mode is measured.
During the benchmark tests the graphics card is evaluated by way of displaying an intricate scene that includes complex geometry, high-resolution textures and a variety of effects.
Talking about wheels turning…The Cinebench R11.5 OpenGL benchmark appears to favor both the gaming cards unfold their computation potential, surpassing easily the lower clocked, Quadro cards that show their age.
This is the first benchmark the older FX1800 loses to the Fermi based Quadro 600, which was trailing pretty close, taking second place in pretty much all the performance tests we have included in this review. Do not forget that the successor of the FX1800 is the current Quadro 2000, so the entry 600 is actually doing pretty well. The Fermi architecture did fare well against this test, allowing the GTX 560Ti to also surpass the much higher spec-ed GTX 670.
I have to admit that I was positively surprised. The puny little Quadro was able to smoke one of the most powerful gaming cards in the market today in the role of a 3D CAD viewport acceleration card, proving that it can comfortably drive a couple of monitors in a dedicated CAD workstation environment. It delivers the goods while remaining a half height, single-slot card that could fit in a mATX or even mITX low profile desktop, draws less than 1/3 the power of the 670, and costs nearly that much less.
Users on a tight budget, rejoice: the nVidia Quadro 600 walks the talk pretty well for a $150 card!
2013/02/08: Charts updated with GTX 560Ti 1GB Data. The card used was an EVGA GTX 560 Ti 1GB, tested on our i7-3820 machine with 310.90 drivers. More cards can be compared through PCFoo’s Resource pages.