Gomeler.com

Review: Sapphire HD3850 512MB CrossFire

03.05.08 - 05:19pm

When one card won’t do, logic says two should do the trick. This is the logic that applies to AMD’s CrossFire and CrossFireX platforms. Typical computers have a single graphical processor that handles all graphical computation, multi-GPU setups though split the workload between the GPUs in hopes of wrangling some additional performance from the system. Today I have two Sapphire HD3850 video cards and a Gigabyte X38-DS4 to perform my first foray into multi-GPU setups.

CrossFire Requirements

CrossFire has a rather loose set of requirements especially when compared to SLI. In theory CrossFire can work on any motherboard with atleast two physical PCIe slots but due to driver issues it is currently restricted to Intel and AMD chipsets. My Gigabyte X38-DS4 has two PCIe 2.0 slots with x16 electrical connections providing the most bandwidth capable in order to help max out these cards performance. In addition to the motherboard you will need atleast one proprietary CrossFire connector which forms a bridge between the cards to to allow them to communicate without clogging the PCIe bus.

This setup is surprisingly quiet compared to the performance it delivers. It is possible to run this with just the CPU fan as it blows over the GPU heatsinks but I prefer to keep things as cool as possible.

Installation

Since my cards have Artic Cooling Accelero S1s in place of their stock heatsinks I was forced to only use a single CrossFire bridge. I’ve been told this shouldn’t impact performance which is good as the heatsinks really let the cards scale very high. For this test I’m simply using a RAID0 setup with Windows and my 3DMark Suite and Crysis.

• CPU: Intel QX9650 3.0GHz 12MB L2 @ 400FSB x 10
• MoBo: Gigabyte X38-DS4
• RAM: Buffalo Technology 2×1GB PC2-8000 Firestix
• PSU: OCZ 700w GameXStream
• VGA: 2x Sapphire 512MB HD3850

On the backside of each cards are the voltage modifications. The red wire is attached to the feedback pin and the white wire is attached to a screw for my mosfet cooling and attached to the ground plane of the card.

Test Method

While some individuals may disagree with this statement, I find the Futuremark 3DMark series to be very influential on what cards I purchase. The benchmark results don’t directly compare with gaming power but they are a standard benchmark we can all compare with. The Crysis benchmark has become a sort of standard game test and while it isn’t indicitive of all games, it gives you a general idea of how a setup performs. With these tests I performed a series of runs to fully explore the aspects of CrossFire. There are stock runs with one and two cards, an easily achievable overclock with one and two cards, and then my “stable” benching overclocks with two cards. The easy overclock is performed with completely bone-stock cards and possible on stock-coolers with acceptable thermal levels. The benching overclocks are only possible only with a voltage modification and substantial cooling but I’d still consider them very low in terms of the possibilities with these cards.

3DMark Results

3DMark 2001 and 3DMark 2003 scale better with CPU speed than graphical power while 3DMark 2003 and 3DMark 2006 scale incredibly well with CrossFire and GPU speed increases. 3D03 and 3D06 really show how CrossFire flexes it’s muscles with 65% to 70% performance boosts. With so much potential left in these cards and their cost this is very impressive and bodes well for some future benchmarking. Something you can’t see in these results was the 3D06 CPU score. Due to CrossFire’s CPU overhead, the CPU score dropped from 6292 with the single-card runs to 6120 with the CrossFire runs. Compared to the GPU boosts, the CPU droop is negligible but still a fact of life and should be noted.

Crysis Results

The Crysis benchmark was the standard Island GPU run with a 1680×1050 resolution and all effects set to Medium via the pull-down menu. Contrary to popular thought, Crysis looks great at this setting and with 30+ frames per second averages and minimums really shows how powerful these cards are. Jumping to two cards really brings out the big guns with average frames per second jumping nearly 50%.

Conclusion

CrossFire is a very unusual concept to consider. Economically it rarely makes much sense as you will never see a 100% scaling by adding another card. For typical video-gaming on monitors under 24 inches there also is little need for CrossFire as you would have trouble straining a single card. CrossFire really only makes sense for high-resolution gaming and benchmarking, and with both options responsible for a sliver of video card sales, it is no surprise that I have hesitations towards recommending CrossFire for the average consumer. Currently two HD3850 512MB cards will cost you $340 which sits between the 8800GTS 512MB, 8800GTX, and HD3870X2. With an overclock I’d have to say the HD3850’s make sense compared to the 8800GTX and 3870X2 but the 8800GTS 512MB complicates matters as it can easily overclock to surpass the 8800GTX and compete within 75% of CrossFire HD3850 performance. In all other cases I’d have to say simplify your life with an 8800GTS or 8800GT 512MB. Now on the other hand, as unprofessional as this is, CrossFire is damn good fun. I crack a smirk whenever I look at my testbed with the monstrous heatsinks, twin video cards, and desk fan keeping the entire ensemble cool. It’s a complicated situation and for that I won’t be giving this setup a final score.

Pros

• Incredible Performance
• High Fun Factor
• Maximizes Platform Potential

Cons

• Financially Inefficient
• Power Hungry

Final Score: N/A

Notice the rubberband connecting the heatsink to the right HD3850. I use this to keep the cards from leaning due to the weight of their heatsinks. Modifications are in order for my test-bed with a bracket for cards.

Category: Articles, Hardware, Review