Logo   Banner   TopRight
TopUnder
Transparent
AMD FirePro V3900: Pro Performance, Bargain Price
Transparent
Date: Mar 27, 2012
Section:Graphics/Sound
Author: Joel Hruska
Transparent
Introduction
The FirePro V3900 is AMD's latest budget solution for the professional graphics market. We covered the card's launch last month; it's a formidable step up from the V3800 it replaces. Budget card launches may lack some of the excitement of high-end product debuts, but lower-end cards like the V3900 account for a significant percentage of AMD's total GPU sales.

The V3900 is interesting for another reason -- it's AMD's latest salvo in its ongoing campaign to steal professional market share away from rival Nvidia. Workstation-class GPU sales are overwhelmingly dominated by Team Green, but AMD has slashed its professional prices in an attempt to siphon market share. The FirePro V3900 is aggressively positioned against its predecessor, as you can see from the specs below.



AMD Fire Pro V3900
Specifications & Features
GPU
Stream Processors: 480
Memory Interface: 128-bit

Memory
Size/Type: 1GB DDR3
Bandwidth (GB/s): 28.8

Display Outputs
DisplayPort 1.2: 1
Dual-link DVI: 1
Max. Resolution per monitor: 2560x1600 @ 60Hz
Max. Combined Horizontal Resolution: 9600x1080 @ 60Hz
AMD Eyefinity Technology Support: Yes
API/Feature/OS Support
OpenCL: 1.1
OpenGL: 4.2
DirectX: 11.0
Shader Model: 5.0
OS Support: Microsoft Windows 7, Windows XP, Windows Vista, Linux (32-bit or 64-bit)

Thermal/Power/Form Factor
Max Power (TDP): <75 W
Max Power (measured): <50 watts
Slots Required: 1
Form Factor: Half Length/Half Height (packaging includes full-height bracket)
Bus Interface: PCI Express 2.1 x16

Street Price:  $119


The FirePro V3900 is currently selling for $114; the FirePro V3800 is $99. Based on specifications alone, we expect the V3900 to be a substantially better value.

Should You Go Pro?

This is one of the most common questions readers ask about professional video cards, so we're going to take a moment to set the record straight. Both AMD and Nvidia have created some FUD on this subject; there's a video on YouTube from last spring in which Bobby Miller, the ex-FirePro maanger at AMD, states that "AMD FirePro graphics cards feature a unique hardware set that differs greatly from what can be found in consumer graphics cards... If you put the AMD FirePro chip into a Radeon board, it simply would not function."

This is somewhat misleading. AMD may perform some additional validation on FirePro products, and the video cards themselves sometimes carry different display outputs or RAM loadouts -- but the GPUs at the heart of the card carry the same features as their consumer-class cousins. That doesn't mean professional GPUs are a bad purchase -- just be aware that they aren't built on fundamentally different technology.



Professional cards like the FirePro V3900 offer support for 10-bit color, up to five simultaneous displays (if you have monitors that support DisplayPort 1.2 and a hub), and accelerated rendering support for 3D applications like 3ds Max, Maya, Lightwave, and a number of other programs. This last feature is typically the reason why people invest in professional graphics cards rather than opting for consumer-class solutions.
Transparent
Understanding 3D Rendering, System Configuration
3D rendering can be confusing if you don't understand how workloads are split between the CPU and GPU. In most programs, final/production renders are still done on the CPU. There are various plugins that explicitly handle production rendering on the GPU, but CPUs are still the default option. The GPU is important because it's used to render what's known as the viewport.

Think of the viewport as a window into the work in progress; it can be configured to show either a simplified view of the total scene or to display particular aspects of the final render, such as camera positions, wireframes, and light sources. Both CPU and GPU performance matter, which is why graphics workstations typically ship with strong solutions in both categories.



When it comes to professional graphics, professional salaries and rendering times dwarf the cost of GPU hardware. Part of the reason professional cards are more expensive is the level of software support AMD, Nvidia, and software developers like AutoDesk provide. If a Radeon card doesn't work properly in 3ds Max or Maya, both hardware and software companies will most likely shrug. If a Quadro/FirePro card has a genuine problem, eyeballs promptly swivel.

Enough chatter -- let's check performance.

System Setup:
Our testbed for this review is an Intel Core i7-2600K CPU and Intel DP67BG motherboard. We configured the testbed with 16GB of DDR-1333 (the minimum recommended amount for the SPEC organization's 3ds Max Professional test) and an OCZ Vertex III. Windows 7 64-bit w/ SP1 was installed; all of the benchmarks were run using the 64-bit version of the tests. All benchmarks were run three times and then averaged.



Nvidia was unable to provide us with a low-end Quadro for testing and comparing the V3900 against a high-end Quadro that's 10 times more expensive, would've been a bit lopsided. We opted to benchmark a Radeon HD 6870 instead in order to test how a consumer card would fare in professional environments. Both the 6870 and V3900 use a GPU from AMD's Northern Islands family but the 6870 is significantly more powerful. It's also more expensive; Newegg lists it at $169.

We'll start with a familiar test to any HH reader -- Cinebench 11.5.

Transparent
Cinebench 11.5

Cinebench R11.5 64bit
Synthetic OpenGL Rendering Performance


Cinebench R11.5

Cinebench R11.5 is an OpenGL 3D rendering performance test based on Cinema 4D from Maxon. The benchmark goes through a series of tests that measures the performance of the graphics card under real world circumstances. Within Cinebench, graphics card testing makes use of a complex 3D scene depicting a car chase which measures the performance in OpenGL mode. Results are given in frames per second; the higher the number, the faster the graphics card.



We typically focus on Cinebench's CPU tests, but the application contains an OpenGL benchmark as well. Before the benchmark runs, viewport output is compared against a pre-calculated image -- if the viewport's version differs significantly from the pre-calculated version, the benchmark won't run. In this case, both the V3800 and V3900 returned a match rate of 99.3%, which is within acceptable variation.

There's no contest here -- the FirePro V3900 is 79% faster than the V3800, while the Radeon HD 6870 is 80% faster than the V3900.

Transparent
SpecViewPerf 11
SPECviewperf 11 is an industry standard, workstation-class, OpenGL performance benchmark. This software suite was released in June 2010 from the SPECgpc project group, and updated February 2011. It features a new GUI that accompanies the fully updated viewsets traced from newer versions of real world applications. The current iteration tests OpenGL performance in Lightwave, CATIA, EnSight, Maya, Pro/Engineer, SolidWorks, Siemens Teamcenter Visualization Mockup, and Siemens NX. Lightwave and Maya are both 3D rendering programs; the other software packages are used in CAD/CAM work.



SPECviewperf 11:
Computer Aided Three-dimensional Interactive Application

As you can see, the performance difference between the two cards varies depending on the application. Some tests, like the Pro/Engineer workload, show a 2% difference, while the V3900 is 47% faster than the V3800 in EnSight and 45% faster in Maya. The V3900 is an average of 21% faster than the V3800 in the entire SpecViewPerf 11 test set. The Radeon HD 6870 lags even the V3800 by a considerable margin, it's dead last in every benchmark save for EnSight-04, where it wins the test.

Now, let's consider performance in specific applications.

Transparent
Lightwave 3D 9.6
Newtek's Lightwave 3D is a  3D rendering program that supports both Macs and PCs. It was used to create the special effects for classic shows like Star Trek and Babylon 5, and films like the upcoming Iron Sky and Jimmy Neutron: Boy Genius.

Lightwave 3D 9.6:
3D Rendering and Animation
SPEC describes its benchmark as follows: "The benchmark...  contains 11 datasets -- ranging from 64,000 to 1.75 million polygons -- and representing such applications as 3D character animation, architectural review, and industrial design. Scores for individual workloads are composited under three categories: interactive, render and multitask."

Performance is normalized against a stock configuration; a score of 3.0 means that the testbed completed a task 3x faster than the original system.



We were surprised at how little difference we saw in Lightwave 9.6; especially since the Lightwave benchmark in SpecViewPerf showed a 13% performance improvement for the FirePro V3900 over its older cousin. This is evidence of how workload choice can impact whether the CPU or GPU is a limiting factor, even within the same program. The 6870 wins the Interactive test by a hair, the other two tests are essentially a tie.

Transparent
3DS Max 2011
Even gamers and general enthusiasts have likely heard of 3ds Max, it's the 800 lb gorilla on the 3D rendering block. We tested both cards using the newest edition of the SPECapc test for 3ds Max 2011. Unlike the other benchmarks in this review, which are available for free, the 3ds Max benchmark is available in two editions. The Personal version contains a subset of the Professional version's tests and sells for $20; the Professional Edition contains 58 tests, including large-scale benchmarks designed to stress even top-end workstations, and runs $495.

3ds Max 2011
3D Rendering and Animation
We tested the Professional Edition of the benchmark -- with one caveat. Both of our AMD cards threw errors if we ran the very first test (Test020 - Building City). Strangely, that particular test would run perfectly -- it was later tests that would fail and crash the benchmark. We contacted AMD and were told that the company is aware of the issue but can't offer a solution at this time. Our results, therefore, are included without the Large Scale CPU results that would normally be generated.



There's several interesting trends to remark on here. The V3900 roars ahead in the large model GPU and GPU shaders benchmark. The V3800's low memory bandwidth is the likely culprit; that GPU was choked on a 64-bit bus; the V3900 doubles the bit-rate to 128. Modeling performance increases by 45.5%, while Interactive graphics is a modest 12.2% faster. The arithmetic average performance improvement for the V3900 over the V3800 is 26%.

The Radeon HD 6870 beats the professional cards handily until it comes to manipulating large models via GPU. At that point, performance craters.

Transparent
Maya 2012
Our last benchmark is of Maya; we used the 2012 version of the program but combined it with the SPECapc test, which dates back to 2009.

Maya 2012
3D Rendering and Animation
Specapc's Maya test page states, "Nearly 40 individual tests are run on five workload models to reflect the processes used to model, animate and render scenes in Autodesk Maya 2009. A particular emphasis is placed on testing performance for large texture sizes and multiple viewports.



In Maya 2012, the V3900 is a hair over 10% faster in the graphics segment and 8% faster overall. Again, this is slightly out of synch with SPECViewPerf 11, which showed the V3900 leading the V3800 by 45%. Both figures are equally correct; they difference between them is a result of different workloads.

The Radeon HD 6870 can't keep up with the professional GPUs, it lags the V3900 by 40% in the graphics test and 28% overall.

Transparent
Conclusion
The FirePro V3900 is significantly faster than its V3800 predecessor. It costs ~$15 more than the V3800, but delivers an average of 20-25% better performance. Depending on your workload, that boost can be as high as 100%. While Nvidia wasn't able to provide us with a comparison card to test, the company told us that it expected the FirePro V3900's performance to rank somewhere between a Quadro 400 and a Quadro 600. The Quadro 400 is a definite low ball -- it's hampered by a 64-bit memory interface and just 512MB of RAM, similar to the V3800. The Quadro 600 fits the V3900's specs more accurately, with a 1GB frame buffer and a 128-bit memory interface, but is currently priced at $154.

We can't draw any performance conclusions from that, but the $40 gap between the FirePro V3900 and the Quadro 600 is a solid indicator of AMD's aggressive pricing. As we covered back when the FirePro V3900 launched, AMD has gone after the professional GPU business by attacking Nvidia's high margins any way it can. With an estimated 20% market share, AMD is still a minority player, but it has been gaining ground steadily the past five years.

Should You Go Pro? (Revisited)

If you're primarily a gamer who likes to dabble in a bit of 3D every now and then, a standard Radeon will suit you just fine. Anyone who works (or intends to work) with 3D modeling software on a regular basis should definitely opt for a FirePro card. The Radeon 6870's performance in workstation applications is too erratic and unpredictable to recommend it as a professional solution. AMD's FirePro V4900 is actually a much better match for the 6870, as it retails for $159 compared to the 6870's $169. The V4900 is clocked 23% higher than the V3900 and offers more than double the bandwidth. Combined, these two features would significantly reduce, if not eliminate, the performance delta between the V3900 and the Radeon 6870.



The AMD FirePro V3900 is a good card at an excellent price point. It's a solid step up from the V3800 and it more than holds its own in professional applications against a more expensive consumer solution. Its performance in leading 3D software programs refutes the idea that professional graphics cards must be expensive. If you're looking for a budget card that still delivers acceptable performance we'd definitely recommend it.


   
  • Strong performance for its price point
  • Significant improvement over the V3800
  • Advertised support for >2 displays depends on specific, specialized hardware


Content Property of HotHardware.com