Many 3850 Options for Your Scratch
Other 3850 Options

Not all 3850 cards are created equal.  When buying one, users must consider some of the differences between models from manufacturers.  For example the Asus 3850 is often quite a bit more expensive for the same card, but it includes a full copy of Company of Heroes.  Most of the “stock” 3850s are crowded around the $169 price point, and that usually means minimal extras and no bundled software.

The next tier up features the 512 MB cards from a variety of manufacturers.  These boards start at around $185 and up, and at this level they are pretty bare in terms of overclocking, extended bundle, etc.  The extra framebuffer is pretty important here.  Many new games are starting to utilize more than 256 MB of onboard memory between texture and shader data.  When users enable anti-aliasing, the memory needs grow even larger.  If a user is not terribly keen on spending extra money then the 256 MB versions do perfectly fine in most games that I have tested to date.  If a user is looking for future compatibility, then the larger frame buffer should be foremost in their mind.

VisionTek Radeon HD 3850 Review - Graphics Cards 10
The card in question.  The VisionTek HD 3850 is based on the AMD reference design, and is eerily similar to most other HD 3850s out there.

The final tier is the 512 MB overclocked boards.  VisionTek is also offering one of these numbers for around $189, and it includes a dual slot cooler that should keep the OC 3850 purring along nice and quiet.  It is honestly one of the better deals around as the core is clocked around 725 MHz with the memory running in the same region as the standard 3850.  The bigger cooler means quieter operation and lower temperatures, but users do lose the extra slot on their motherboard.

VisionTek is offering two really nice products in the sub-$200 market with their stock 3850 at $169 and their overclocked version starting at $189.  The extra performance and the larger frame buffer make the extra $20 worth it to most users.  If a user is looking to save the money and have only a single slot solution, then the standard 3850 is still a great deal.

Test Setup

I decided to test the HD 3850 against the Gigabyte GeForce 8600 GTS.  The GTS in the passive form from Gigabyte is about $20 cheaper than the stock 3850 at this time, so the price difference is not all that great.  I also decided to use my nice, new Phenom test bed to see how these budget cards performed on what is essentially a budget quad core platform.  I use a good mix of DX9 and DX10 gaming, from older titles to brand new.

I tested at both 1280 x 1024 and 1600 x 1200 resolutions.  These boards are not exactly equipped to run much higher than that with acceptable smoothness.  This is primarily due to the 256 MB frame buffer and the use of AA throughout the tests.

AMD Phenom 9900 (2.6 GHz)
Asus M3A32-MVP Deluxe
2 x 1 GB Corsair DDR-2 @ 1066 5:5:5:15
Seagate 7200.10 320 GB HD
Lite-On DVD-R/RW
Thermaltake PurePower 600 Watt Power Supply
Windows Vista Ultimate 64 bit
AMD Catalyst 8.1
NVIDIA Forceware 169.29

VisionTek Radeon HD 3850 Review - Graphics Cards 11
Sure enough, the top card (VisionTek) is essentially identical to the manufacturing sample I received from AMD (bottom).  This really is not a bad thing because the reference design is quite good.


3D Mark 2005

This old and trusty benchmark is still rather vertex limited on older cards, but with the unified architectures that we are running the benchmark now it seems to spread its wings a bit more.  Standard settings are used.

3D Mark 2005

HD 3850

GeForce 8600 GTS

3D Marks



Game Test 1

60.2 fps

51.1 fps

Game Test 2

49.7 fps

35.9 fps

Game Test 3

94.1 fps

56.3 fps


11.0 fps

11.9 fps


11.1 fps

12.2 fps

We see a couple of interesting results here.  First off the HD 3850 is faster in each of the GPU tests.  When we hit Game Test 3, which is the most complex of the tests, the 3850 almost doubles performance over the 8600 GTS.  The 8600 features 32 stream processors at 1.4 GHz, while the 3850 is pushing 320 stream processors (though of course architecturally the stream processors are very different) at 668 MHz.  The one exception is the CPU tests.  At first glance we can perhaps see that the driver overhead on the NVIDIA part is much smaller, and allows for greater CPU performance.  This observation is primarily conjecture based on one test.

3D Mark 2006

The updated benchmark really pushes DX9 parts with more complex shaders and scenes, though they are all based on the previous 3D05 tests.  This is a much more balanced benchmark than the previous version, and it can really hammer even top of the line cards.  The standard settings are used.

3D Mark 2006

HD 3850

GeForce 8600 GTS

3D Marks



Game Test 1

29.730 fps

18.830 fps

Game Test 2

35.302 fps

20.812 fps


38.242 fps

19.528 fps


45.538 fps

24.530 fps


1.239 fps

1.200 fps


1.761 fps

1.783 fps

We see much the same situation with this benchmark as the past.  The 3850 is just basically running roughshod over the older 8600 GTS.  We can see exactly what that extra $20 will buy us.  That is, if you run 3D Mark 2006 all the time.  Note as well that the CPU bound results are almost identical, which does not say much about my above mentioned supposition about driver efficiency.

« PreviousNext »