Now one of the Newest Quadro cards to be released; the Quadro 2000 is based on Fermi architecture. Listed under â€œmid-rangeâ€ $500 price point from Nvidia, this card has much more going for it than Nvidia would like us to believe. The card is to replace the Quadro 1800 FX. Specs show it may be nearing the performance of last generation Quadro 3800 FX which is listed under â€œHigh-Endâ€. Being based on 106 GPU it shares the same chip as the GTS 450 but optimized for the workstation environment. The Clock Speed was lowered to reduce power intake and ultimately leaving it without a 6-pin power connector. This is a good sign for a quiet and cool running video card but bad if you were thinking about overclocking it (More on that later). What stands Quadros apart from the consumer cards is the stability and performance gain over a Geforce card. No different this time around, the Quadro 2000 is well rounded with 1GB of GDDR5 Ram and 625/1251Mhz clock. Letâ€™s take a closer look.
Closer Look: Box art & Size
PNY is the only Quadro card maker which gives them the power to choice how to package the cards. They go with a simple look this time around but I was disappointed to find the card was more or less â€œfloatingâ€ in its plastic holder. A $400 card should have a little more care to it and maybe a good form holder would be nice.
This was a very disappointing thing to find this on a band new card. This is a picture left over residue from the soldering process all over the back of the card. Itâ€™s like someone forgot to clean the card up. Iâ€™m also not really sure at what point you would avoid this (before or after the soldering?). In any case itâ€™s a very sticky card making it not nice to handle while installing.
Quadro Drivers are a little different than any Gerforce one. The options under â€œGlobal Settingsâ€ has expanded greatly for every sort of 3D and 2d Professional app. Gone is presets for video games, but that is expect since Quadros arenâ€™t meant to be gaming on.
Graphics Processing Clusters: Unknown
Streaming Multiprocessors :Unknown
CUDA Cores :192
Texture Units :Unknown
ROP Units :16
Graphics Clock (Fixed Function Units) 625mhz
Processor Clock (CUDA Cores) 1251mhz
Memory Clock (Clock rate / Data rate) 652mhz
Total Video Memory 1024MB GDDR5
Memory Interface 128-bit
Total Memory Bandwidth 41.7 GB/s
Provided by Nvidia.com
Next-Generation NVIDIAÂ® CUDAâ„¢ Architecture
Breakthrough NVIDIA Fermi architecture incorporates CUDA parallel computing capabilities and advanced visualization to deliver performance that greatly accelerates professional workflows.
NVIDIAÂ® Scalable Geometry Engineâ„¢
Dramatically improves geometry performance across a broad range of CAD, DCC and medical applications, enabling you to work interactively with models and scenes that are an order of magnitude more complex than ever before.
Large Framebuffers with Ultra-Fast Bandwidth
Large GPU memory with fast bandwidth for accelerating the display of complex models and scenes, as well as performing complex computation on large datasets.
NVIDIAÂ® Parallel DataCacheâ„¢
Supports a true cache hierarchy combined with on-chip shared memory. L1 and L2 caches drive exceptional throughput, accelerating features such as real-time ray tracing, physics and texture filtering.
NVIDIAÂ® GigaThreadâ„¢ Engine
Provides up to 10x faster context switching compared to previous generation architectures, concurrent kernel execution, and improved thread block scheduling.
30-bit color fidelity
30-bit color fidelity (10-bits per color) enables billions rather than millions of color variations for rich, vivid image quality with the broadest dynamic range.
NVIDIAÂ® Mosaic Technology
NVIDIAÂ® Mosaic TechnologyÂ¹ enables transparent scaling of any application across up to eight display channels.
Full-Scene Antialiasing (FSAA)
Up to 64X FSAA dramatically reduces visual aliasing artifacts or â€œjaggies,â€ resulting in unparalled image quality and highly realistic scenes.
Testing: Setup & Overclocking
â€¢ Processor Intel Core I7 920 [email protected]
â€¢ Cooling: Stock Cooler
â€¢ Motherboard: AsRock X58 Extreme
â€¢ Memory: 6x2gb DDR3 1066
â€¢ Video Card(s): NVIDIA GTS 450, ASUS GTX 285, Evga GTs 450
â€¢ Power Supply: Seasonic x650 Gold
â€¢ Hard Drive: 1x Seagate 500gb 7200.12
â€¢ Optical Drive: HP DVD
â€¢ OS: Windows 7 Professional 64-bit
â€¢ Case: Cooler Master HAF 922
Comparison Video Cards:
â€¢ Quadro 2000
â€¢ Asus GTX 285
â€¢ Evga GTS 450 Superclocked
1. SPECview 11
2. Cinebench 10 64x
3. Cinebench 11 64x
4. Photoshop (Beauty Box Plugin)
5. Final Fantasy 14 Benchmark
6. Unigine Heaven Benchmark 2.1
7. 3DMark 06
8. 3DMark Vantage
9. Crysis Benchmark
Nvidia Offical Score Test Rig (SPECviewperfÂ® 11: for more information www.spec.org. Tested on single proc 3.3 GHz Quad Core XeonÂ® w5590, x58, Windows 7-64, Driver Release rel 260+)
As you see here in the SPECviewperf benchmark it shows off the difference between a professional card and customer card. The scores are based on a average from a mean. It does not represent frames per second but it feels like it when you see objects moving slowly. The Red and Yellow is Nvidia official score from their website. The only problem was I could not find what resolution it was running at. Itâ€™s a guess but it matches more or less. I did find however the Maya-03 ran with issues as test #5 and 6 just more for less froze and went avg of .5fp. After Reinstalling the drivers with no luck the issue may be in the newest drivers. Lastly as you can see SLi for comsumer cards has zero effect on score. In some ways it seemed slower. If you are thinking about buying two gaming cards to outperform a professional one, think again.
Update: Using newer drivers I was able to resolve the Quadro Maya issue. It was about 6 months later when the (at the time) current drivers came out and i re-ran this benchmark once again. unfortunately I did not record the number but I remember it was 50~ range.
Running it at 1080 doesnâ€™t change the score much, itâ€™s simply a way to show another resolution you may be working at. One thing that is out of place is GTS 450 on ensight-04 which is keeping up with the Quadro. Once again, I redid the drivers and ran the test a few times and nothing changed. I currently have no proper conclusion until I can find out how demanding ensight really is.
During the initial test I thought the AA option in the benchmark was working until I was playing around with the new Quadro options to go up to 64x. I stumbled across that fact that AA only worked when I forced it for the Spec benchmark. Until I am able to re-run the GTS 450 and GTX 285, all you get is the proper Quadro scores. The performance drop is not huge and I prefer the quality of 4x making the geometry smooth in wireframe.
With Cinebench 11 the same idea stands. It really doesnâ€™t matter which card you have. Even a entry level card like GTS 450 pulls in with 36fps. What I did find interesting was when you overclock the CPU the FPS goes up. during the test you can see the car run over cones with physics from the CPU. Once being CPU bound is eliminated itâ€™s even less of issue which card you have. Lastly it looks as the Quadro 2000 wasnâ€™t CPU bound, even at 4ghz it gain nothing. Same goes for the GTS 450, itâ€™s just not powerful enough to be held back by the CPU.
The newest addition of Adobe Photoshop has added even more support to GPU usage. CS5 has options to move the workload of resizing, filters, masking, etc to the GPU over the CPU. While this can cause corruption in the final product, this is something you donâ€™t have to worry about while using a Quadro card. If you fear the loss of quality on some crucial work you can also turn off GPU acceleration. The Beauty Box plugin is a great example of what is to come in offloading workloads. This plugin operates like any CUDA app would. As you can see if you donâ€™t mind a crash once in a while, any card will do and the more cores it has the faster it will render the final product. The only reason to stick with a Professional card would be to ensure stability while working. I canâ€™t stress stability enough as when I used a GTX 285 on a daily bases I could guarantee a crash at least once a week in Photoshop. Itâ€™s never fun to redo an hour of hard work.
Final Fantasy 14
Heaven Benchmark 2.1
Originally I ran everything at DX11 forgetting the GTX 285 doesnâ€™t support it. While Tessellation looks great, itâ€™s unfair to compare since it has a larger impact on performance. Not much to report on this since they are performs as I expected. I will say that the Quadro had really good color repetition something that made me go check the Nvidia settings to see if I left something on. One of Quadros strong point is they keep the colors accurately. Even with my monitored already calibrated something is just lost when a Geforce card is rendering and I wish all users could see games â€œHow itâ€™s Meant to be Playedâ€
While Vantage was heavily criticized during launch for what games will look like in the future over what was being shown at the time. We now can see with many games are coming out with Physx and HDR lighting. It is not too far off from what is currently being put out these days and worth a closer look. We see the Quadro lagging behind again but the main propose is to show it can stay within a playable frame rate.
But can it play Crysis?
An ongoing joke across the web and forums, I had to poke fun at it. Other than showing its unplayable the main reason was to see what kind of workload it can take. These daysâ€™ games like Crysis are made in an open Sandbox world where hundreds of artists and programmers can work in real time on the same project. ID software is a good example as â€œRageâ€ is being made with an open Sandbox engine. Things will literally pop in and out of existence as you are working on something. By looking at the benchmark unless you lower the settings or get a better Quadro itâ€™s going to be hard to make a level at 12fps. This is future of making games in a large firm I believe.
The Quadro 2000 may have a small heatsink but doesnâ€™t produce enough heat to really warrant much bigger. Personally I would have liked a two slot GTS 450 heatsink but you canâ€™t have everything. The temps under Furmark are around 70c~. After about 5 minutes of slowly climbing up, it doesnâ€™t go any higher. This is very unrealistic environment because nothing can stress a card out that much but itâ€™s good to see it sticking well under the 95c limit.
The Quadro 2000 isnâ€™t as quiet as Nvidia tries to play it off. Rated â€œOfficialâ€ at 28db I canâ€™t hear it over any stock fan. What I can hear is the high pitch squeal when the fan runs at full speed. Itâ€™s not the same as when a card is running Folding@home which gives a similar squealing sound. Simply put if you have a loud computer you wonâ€™t even know it is running. Otherwise I suggest investing in the silencer form to eliminate this high pitch sound.
N/A Simply donâ€™t have the right measurement tools. Easily countered by the fact that the card on paper only draws 63watts without a 6pin connector.I would recommend even with a fully decked out system (as in the test rig) a Quality 400watt psu would not break a sweat.
The Quadro 2000 is a well rounded card for its price. Keeping close to the last generations higher end cards and well outperforming the old mid range, this card has something going for it. I was however very disappointed to find in small print that only Quadro 5000 and 6000 have ECC memory. Nvidia was selling the idea left and right that I didnâ€™t think to check until I noticed the option wasnâ€™t present in the control panel. Very sneaky of Nvidia and I donâ€™t see a reason not to include ECC ram in all the cards since it can only cost dollars more to them per card. This true is a downside since you canâ€™t count on CUDA not producing any errors during a large task. Overall I can only suggest this card to a few select cases. That being (A) Already looking at the price point being the maximum you can spend. (B) Looking to upgrade from entry level to something that can keep up and not struggle. (C ) Not working on critical workloads. That being said the Quadro 5000 is currently $1,700 on Newegg. A very steep price point for true stability in the sense ECC ram is just a redundancy to a already quality product and drivers. I would not expect a Quadro card of any kind to crash because of a driver issues. Unlike my time using a consumer card for many years, Maya and Photoshop constantly crashed during the oddest times and just came natural to Ctrl+S every minute.
â€¢ Great performance
â€¢ Low noise
â€¢ Low Power Requirements
â€¢ Small Heatsink
â€¢ No ECC Memory
â€¢ Lacking SLI
Update: Since the update I have been able to run a GTX 570 and Qaudro 3700. Since it is not on the old test bed I cannot officially include them into the review. I will in the coming weeks re-run the benchmarks to include them on the correct platform. I also found it interesting for observations that the GTX 570 was faster than Quadro 2000 in and ONLY in the viewport 2.0. Because i cannot benchmark this, it's purely my own observations. Moving a 25 Million Polygon model was fast and responsive with the GTX 570. Using the same 3d model the Quadro 2000 was slow to move and respond in the viewport 2.0.
Here is a look at what is to come.
Using a i7-3930K, Asus P9X79 WS, Quadro 3700, SSD and 64GB (8x8gb) G-skill 1600 Here is the Specs scores for the Quadro 3700, GTX 570 and Quadro FX 4800
Quadro FX 4800
Viewperf 11 - 1080x1920 0xAA
GPU: 46.74 fps
Viewperf 11 - 1080x1920 0xAA:
Viewperf 11 - 1280x1024 0xAA:
maya-03 8.84 (slower than 1080)
GPU: 28.9 fps
GTX 570 ASUS DirectCUII
Viewperf 11 - 1280x1024 0xAA:
GPU: 51.72 fps
Edited by hornybluecow, 07 December 2012 - 07:53 PM.