It’s been many months since we last saw an Nvidia Quadro, firing through the door in rapid succession were ATI's FireGL and FirePro cards, which have taken an unprecedented lead within the professional graphic card arena. So when we learned a few months back that the Quadro FX4800 was on its way, our interested eyes snapped open as we asked ourselves the question - is Nvidia going to regain the elusive diamond studded crown?
On paper, this looks to be one mighty beast, bulging with 1.5GB of memory and a very cleverly clocked GPU and DDR3 memory. Most importantly, it supports the new 55nm core process or for some the GT200GL supporting some 1.4 billion transistors, so this should in actual theory see some massive improvements upon its predecessor the Quadro FX4600 with the clever implementation of 192 CUDA processing cores.
Nvidia's PR claims this for the Quadro FX reference CUDA: "All Quadro FX GPU’s include a substantial portion of die area dedicated to processing, unlike CPUs where a majority of die area is dedicated to onboard cache memory. Rough estimates show 20% of the transistors of a CPU are dedicated to computation, compared to 80% of GPU transistors. GPU processing is centred on computation and throughput, where CPUs focus heavily on reducing latency and keeping their pipelines busy (high cache hit rates and efficient branch prediction)."
So the obvious target markets for this ultra high end professional graphics card are geophysicists, designers, scientists, engineers, medical imaging and technical professionals demanding nothing but the best from their workstations.
Before we move any further forward, let's have a swift look at what the card actually has to offer the demanding end user.
| |
Nvidia Quadro FX 4800 |
| CUDA processing units |
192 |
| Memory Configuration |
1.5GB DDR3 |
| Memory Controller Interface |
384 bit |
| Memory Bandwidth (GB/s) |
76.8GB/sec |
| Shader Model Support |
4.0 |
| Direct X10 Support |
Yes |
| Open GL Support |
2.1 |
| High Colour Fidelity Output |
10 bit |
| Display Output Connectors |
1xDVI-I, 2X Display Port |
| Dual Link DVI Support |
Yes |
| Multicard/SLI Frame Rendering |
Yes |
| G-Sync |
Yes (optional) |
| SDI |
Yes (optional) |
| Power Consumption |
146 watts |
Nvidia Quadro G-Sync
One of the more radical steps is the optional support for both G-SYNC and SDI daughter cards. This is a good move forward as you can see below.

Workstations using the Quadro FX 5800 and FX4800 graphics boards combined with the Nvidia Quadro G-Sync option board can be Frame Synced together to synchronize their display outputs to each other, this capability allows you to create power walls, caves, and large venue display systems driven by multiple workstations. Additionally the Genlock feature allows the graphics output to be synchronized to an external source, typically for film and broadcast video applications.
Nvidia Quadro SDI
Nvidia has provided a brief demonstration of what this feature can do.

Nvidia's explanation for this graphic is that Serial Digital Interface (SDI) is the SMPTE broadcast video standard for transfer of uncompressed broadcast-quality video throughout a production environment, including cameras, recorders, and displays. With the addition of an SDI interface card both the Quadro FX 5800 and FX 4800 are intended for digital broadcast professionals who use various applications — such as virtual sets, sports, and weather news systems — to composite live video footage onto virtual backgrounds and send the result to live video for TV broadcast. Quadro SDI also allows film production, post-production, and finishing professionals to preview the results of 3D compositing, editing, and colour grading in real time on high-definition (HD) broadcast monitors. This graphics-to-video-out function delivers uncompressed 8-bit, 10-bit, or 12-bit SDI from programmable graphics, enabling a direct connection to broadcast monitors, switchers, tape decks or SDI projectors.

What cannot be seen here are the Power and SLI connectors. The card runs from only 1 X 6 pin PCI Express power source, so there's no need for worries on power supplies. End users who have older power supplies that only came prefitted with 2 X 6 pin PCI Express power plugs - you are in luck as hidden away from the streamlined cooling mount is the power point in its normal place.
Maintaining streamlining of the card is a rubber cap which fits over the SLI connector and so on, that's a good way to maintain optimal airflow around both the card and within the system. So some nice thoughtful pieces of engineering have been built in.
Test Systems
Our "in house" test systems for this are courtesy of Intel and Supermicro. The Intel Desktop Mainboard DX58SO (Smackover) with Intel 3.2GHz Intel Core i7 965 Extreme Processor. This is a superb resilient platform that just keeps pushing out the data without errors.

The Skulltrail has been around for sometime now and still continues to amaze many with its raw processing power. Many have seen this configuration working flawlessly within the workstation environment. We took it one stage further to show the SoHo users that this mainboard configuration can be used successfully with the Nvidia Quadro FX4800 in the workstation environment.
We aimed to show how powerful the set-up really is that we have today. Although these are expensive parts of equipment the facilities that the board has to offer are huge. Although Skulltrail is only PCI Ex 1.1 compliant, the results we have collated will astonish you all.
The last system comes by means of Supermicro - we named it "The Cruncher Machine" due to its superb flexibility within the professional workstation arena. Today, to "check sum" some previously issued Nvidia Quadro figures we have configured it to clarify the guideline figures.
| Component |
Test system 1 |
Test system 2 |
Test system 3 |
| Motherboard |
Intel DX58SO
(Smackover) |
Intel desktop
D5400XS |
Supermicro
X70WA-N |
| CPU |
1X32.GHz
Core i7 965
Extreme |
2X3.2GHz
Core 2
Extreme QX
9775, 1600MHz
FSB, 12MB L2 |
2X3GHz Xeon
E5472
Harpertown,
12MB L2,
1600FSB |
| CPU cooling |
Akasa 966
Socket 1366
Nero HSF |
Akasa AK-965
Intel HSF |
2XAkasa 384
Low Profile |
| Memory |
3X2GB Crucial
DDR3 1333MHz
Nehalem |
4X1GB Crucial
800FSB DDR2
1.8V FBDIMMs |
4X2GB Kingston DDR2 800MHz
FBDIMMs |
| Hard Drive |
2X300GB WD
Velociraptor,
16MB cache
S-ATA 3GB/s
Raid 0 through
ICH 10 chipset |
1X300GB WD
Velociraptor 16MB cache,
S-ATA 3GB/s |
1X300GB WD
2.5-inch
Velociraptor 16MB cache,
S-ATA 3GB/s |
| Video Card |
Nvidia Quadro FX4800
|
| Driver Rel |
178.46 WHQL
|
| Chassis |
Akasa Infiniti
ZOR |
Akasa Omega
eATX |
Akasa Omega
eATX |
| PSU |
Akasa
Powermax
1000W |
Akasa
Powermax
1200W |
Akasa
Powermax
1000W |
Benchmarks and Software Used 32-bit and 64-bit Mode
To maintain our target audience within the professional market place, we opted for a complete set of professional applications benchmarks.
Test System 1 - 32-bit Mode
Microsoft Windows XP Professional SP3
Autodesk 3ds Max 9 SP2 Scenes
SPECapc for SolidWorks 2007™
SPECviewperf 10
Test Systems 2 and 3 - 64-bit Mode
Microsoft Windows XP Professional 64-bit and SP2
MAXON CINEBENCH 10 64-bit Test
SPECviewperf 10 64-bit Test
Each set of tests has been applied on the clean system hard drives shown above to ensure that no residue drivers were left installed with all updates/patches applied. A test/render has been completed many times over different periods of the system uptime. Maintaining the fair play rules of SPECviewperf and SPECapc for SolidWorks 2007 we did not manipulate any of this mainboards functionality and left it at its default settings in which it was supplied. Tests have been conducted at 1280 X 1024 @ 60Hz in 32 bit colour. Results that have been shown within this review are from the application/benchmarks first run in accordance with the SPECviewperf fair play rules.
The results speak very clearly for themselves.
Autodesk 3ds Max9 SP2 Scene
| Scene |
Time to finish |
| CBALLS2 Video Post |
1min, 25 seconds |
| Radiosity Render full 101 Frames |
9 min, 41 seconds |
| Underwater, full 301 frames, 5 frame pass |
1h, 4min, 40 secs |
SPECapc for Solidworks 2007 32 bit

SPECviewperf 10 32 bit

Maxon Cinebench 10 64 bit test

SPECviewperf 10 64 bit test

Conclusions
Let’s firstly start by looking at the all important SPECviewperf 10 results that so many hinge upon.
These are good and here they are in a consolidated format.
| |
Test system 1 |
Test system 2 |
Test system 3 |
| 3dsmax-03 |
53.04
|
49.81
|
47.32
|
| catia-02 |
64.42
|
59.48
|
57.92
|
| ensight-03 |
53.41
|
24.55
|
53.84
|
| maya-02 |
234.34
|
225.94
|
222.81
|
| proe-04 |
63.73
|
49.73
|
49.24
|
| sw-01 |
141.08
|
99.95
|
99.73
|
| tcvis-01 |
28.78
|
28.78
|
28.78
|
| ugnx-01 |
33.39
|
33.48
|
33.40
|
Taking into consideration that this is the first release of this driver, we must ask ourselves how much more can be gained from it. Interestingly it has been reported that a new driver is around the corner within the next few months and this will address the shortfalls seen within TCVIS and UGNX. So we will and redo these tests when the time is right.
As we've said before, Supermicro is a beast worth seeing in action. Taking into consideration this was running on a 3GHz Xeon in many places it comes very close to the other results shown. It would be interesting to see just how close, if evenly matched with two 3.2GHz Xeons running in at Raid 0. As we've said, while the Skulltrail platform only supports PCI Ex 1.1 compliancy, it still romps home with some very respectable results indeed. So those of you out there with this system and planning the upgrade it could just be worth it just now with the Quadro FX4800 until the more modern Dual Nehalem platforms arrive.
Within the OpenGL results these are what we expected, with the exception of the Intel i7 3.2GHz results. These are remarkably fast. Though the 32-bit OpenGL results are most certainly optimised we see for Intel's i7 processor, which is a good sign as buyers looking its latest offering will know that they should not see any unforeseen problems. As shown earlier, the mainboard supports both ECC and Unbuffered Non ECC memory.
The results from SPECapc for SolidWorks 2007™ has blown us away. To date we have never seen such high scoring. To check sum it was no freak result we completed a fresh run iaw the fairplay rules and the results came in extremely close to the first run. Some very happy smiling faces will be quietly walking about today within Intel's HQ at Satan Clara.
What of the cost and availability of this new wondrous bit of equipment fresh from the pedigree stable of Nvidia? At the time of going to press, the current pricing of the Nvidia Quadro FX4800 wings in at the starters gate at $1,999 USD or €1350 Euros. The boards will ship in the next couple of weeks.
Notwithstanding, the utilisation of CUDA, developer and performance tools available to the proficient end user, this card has much to offer. Therefore Nvidia's professional driver team has looked very carefully at its new contribution to the professional end user. There has been much focus on getting the final WHQL drivers out on time with the product.
It's early days to make comments and predictions on the way ahead, though what has been observed today has been exceptional. How will the market react? This is a very difficult question as its been some time since a product of this magnitude has hit the streets.
Ultimately, this is when things get very difficult indeed, in one breath the Nvidia Quadro FX4800 has performed remarkably well indeed. These are very early days for a product of this nature and with the forthcoming new drivers improvements will be seen. For those looking for a swift upgrade at a snip of the cost of the previous Quadros then this is the way ahead as the price for such good value will make so much of a difference. X
|