Radeon HD 4830: High-Speed, Cheap CrossFire

Test System Setup and Benchmarks

First off, we’d like to once again note that we did not use a reference 8800/9800 GT for our tests and instead chose the factory overclocked Gigabyte 8800 GT TurboForce edition. With a 100 MHz graphics processor overclock to 700 MHz and a 20 MHz memory overclock to 920 MHz, Gigabyte’s 8800 GT sports one of the highest factory overclocks available for an 8800/9800 GT.

Because we’re testing both SLI and CrossFire dual-card configurations, we can’t use the same platform for both. The only platform that is both SLI- and CrossFire-certified is Intel’s new X58 Core i7 chipset. Instead of using that one, though, we used a DFI Lanparty X38 for CrossFire testing and an MSI P7N Platinum for SLI testing.

This presents us with a bit of a challenge in that Intel’s X38 chipset provides true 16x PCIe bandwidth to both graphics card slots when used in CrossFire mode. Unfortunately, Nvidia’s 750i chipset only allows for 16x PCIe bandwidth for a single card, and when two cards are used in an SLI configuration, both graphics slots are allotted 8x PCIe bandwidth each.

However, there’s not really any problem at all. The 750i chipset offers PCIe 2. 0 signaling, and two 8x PCIe slots should deliver plenty of bandwidth for a couple of 8800 GTs in SLI configuration. Thomas Soderstrom’s recent article PCI Express and CrossFire: Scaling Explored demonstrated that even two powerful Radeon HD 4870 cards were not significantly hampered when each card had 8x PCIe 2. 0 bandwidth compared to 16x PCIe 2. 0 bandwidth for each card. It appears that if both motherboards are up to the PCIe 2. 0 standard, there won’t be a notable performance difference.

Furthermore, the Nvidia 750i chipset makes much more sense to compare to an X38 motherboard when price is concerned. They both can be found for about $160. When you consider the cheapest dual-slot 16x PCIe SLI solution is the 780i chipset priced between $200 and $300, the 750i is a much more realistic comparison within a budget.

With this in mind, the Radeon HD 4830 cards were tested on the DFI Lanparty X38 motherboard, while the GeForce 8800 GT cards were tested on the MSI P7N Platinum 750i motherboard. Both motherboards were fitted with the exact same E8500 processor and PNY DDR2 memory. We swapped a single 8800 GT and 4830 between the motherboards and ran a few benchmarks to ensure that the results were similar on either platform.

Both the PowerColor Radeon HD 4830 and the Sapphire Radeon HD 4830 sport the exact same clock speeds and GPU, and it quickly became apparent in our testing that they would also produce the same results. Therefore, to minimize confusion in the benchmark charts, we have only included one representation of Radeon HD 4830 performance in both single-card and CrossFire configurations. The exception to this is, of course, where overclocking and noise levels were tested.

As an added bonus, we included the fastest video card currently available, the Radeon HD 4870 X2, in our benchmarks to see where the performance of these cheaper solutions fit in relation.

Now, let’s examine the test systems:

Component CrossFire System SLI System
CPU Intel Core 2 Duo E8500
3. 16 GHz, FSB-1333, 6 MB Cache
CPU Cooler Xigmatek HDT-S1283 Unchanged
Motherboard DFI Lanparty X38
Intel X38, BIOS 6. 00 (10/03/2008)
Northbridge Voltage: 1. 30 V
MSI P7N Platinum
Nvidia 750i, BIOS 1. 2 (7/11/2008)
2x 2, 048 MB, DDR2-800, CL 4-4-4-12 at 1. 80 V
Graphics Radeon 4830
RV770LE GPU , 575 MHz
512 MB GDDR3 900 MHz
Gigabyte GeForce 8800 GT
Nvidia G92 GPU, 700 MHz
512 MB GDDR3 920 MHz
Hard Drives Western Digital Caviar black 640GD
640 GB, 7, 2000 RPM, 32 MB Cache
Sound Onboard Unchanged
Network Integrated Gigabit Networking Unchanged
Power Antec Neo 650W
ATX 12 V/EPS 12 V, 57A between three 12 V Rails
Optical LITE-ON 20X DVD±R SATA Model iHAS120-04 Unchanged

Software and Drivers
Operating System Windows Vista Ultimate 32-bit, SP1 Unchanged
Graphics Driver Catalyst 8. 10 32-bit Edition Nvidia Forceware 180. 48

Now that you have an idea about the system settings we used, have a quick gander at the games and benchmark settings we applied before digging into the nitty-gritty of the benchmarking charts:

Benchmarks and Settings
Version: 1. 2. 1
Video Quality: Very High Details
Demo: CPU-Benchmark + Tom’s Hardware Tool
Far Cry 2:
Version : 1. 0. 0
Video Quality: Ultra High
Demo: FC2 benchmark Tool
Supreme Commander
Forged Alliance
Version : 1. 5. 3599
Video Quality: Highest Settings
Demo: WallaceTX_006_006
Benchmark: Fraps 2. 9. 4–Build 7037
World in Conflict
Version : 1. 0. 0. 9
Video Quality: Very High details
Demo : Game-Benchmark
Race Driver GRID
Version : 1. 0
Highest Details
Gameplay: Detroit Pro-tuned race, rear starting position
3DMark Vantage
Version : 1. 02
Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
Comment from the forums
    Your comment
  • badge
    Great look into the 4830. Makes me want to buy a Xfire setup using these.
  • Doltron
    If you couldn't get Tray Tools to work with the Sapphire card why not use another program? Instead of giving up and coming up with a lame conclusion.

    Also 993*2 doesn't equal 1885 and the 4870 is clocked at 750 not 780.
    Your sound and temp charts have FPS on their X axis.
  • hannibal
    It's nice to see good cards from both companies, ATI and NVidia!
    And the price is not bad at all. The competition is so good!
  • neiroatopelcc
    The chart on page two sais 4850 runs at 625 - but stock is 600, and 4870 at 780 - which is 750 stock ... so is the 4830 speed correct?
  • cangelini
    Numbers and charts are corrected.
  • cangelini
    Actually stock clocks on the 4850 *are* 625. :P
  • V3NOM
    I'm sure i saw that "4850 - smarter by design" article at anandtech first. or somewhere else... the name anyway not necessarily the article >.>
  • giovanni86
    I knew that the 8800GT wasn't that fast, but those benchmarks ahve to be wrong... Sorry Nvidia fan boy here. Bye.
  • V3NOM
    Um actually, the MSI runs at x16/x8 in SLI mode. If you instead got an evga 750i FTW motherboard, you would find it runs at x16/x16 in sli, thanks to its unlocked NF200 chip. the 750i FTW is not a reference nvidia board as the MSI is.
  • V3NOM
    LOL at 1680x1050, the 4870X2 IMPROVES when 4xAA is added? i smell a rat...
  • cangelini
    Which test are you talking about, Venom? is added? i smell a rat...
  • daskrabbe
    Far cry 2
  • cangelini
    daskrabbeFar cry 2

    In Far Cry 2 there is a .8 frame difference, and shifting to 1920x1200 costs 2.1 frames at 4xAA. This is a processor bottleneck. In other words, performance is similar with and without anti-aliasing applied because the graphics card is nowhere near taxed at that resolution or the one above it.
  • kelfen
    Nice article ATI has really been on the move in all price ranges in creating compitition I would think the next gen cards are going to be a die strink if you look at how they got 4xxx. 3xxx die shrink and 4xxx beef up in power for competitive cards/price.
  • neiroatopelcc
    cangeliniActually stock clocks on the 4850 *are* 625.

    Oh my bad. I mixed up the numbers with some on g92 chips (just bought 28 9600gt's yesterday)

    Anyhow - the 4870 is 750, not 780 - at least they were when I bought mine.

    "On a side note, we will mention that GRID is one of those games that really does require AA for the best visuals. Happily, the game engine seems very easy on the video cards and even the single-card configurations were able to provide 4xAA with playable frame rates."
    I want to add that this is only true for current generation cards. My dad's p4 with a 7600gs can only run it with grahpics at very low @ 800x600 - though he runs suppreme commander just fine at 1024 ...
  • cangelini
    neiroatopelccOh my bad. I mixed up the numbers with some on g92 chips (just bought 28 9600gt's yesterday)Anyhow - the 4870 is 750, not 780 - at least they were when I bought mine.

    Yup, you're right--the chart was originally incorrect, but I went back and corrected that spec, along with the memory frequency mentioned by Doltron.

    Curious to hear how your dad's system runs SC no sweat at 1024. This is one of those ones that consistently drops test platforms to their knees. He actually gets playable frame rates on a P4?
  • neiroatopelcc
    yes he does. Mind you it's not with aa on or anything set at max res. But he plays it just fine. He doesn't have the expansion though - doesn't play it all that much. Dunno if the expansion makes any difference.
    His rig (2.4 northwood, 2gb pc3200, 7600gs on a cantherwood chipset) plays test drive, age of empires 3 and supreme commander at playable levels, but doesn't do grid playable. I suppose he'd have a chance at grid if we'd oc the cpu, but last time we ran 3,2 I ended up breaking their c:\windows\system32\config\system file ... and he didn't like that.
    Would have loved to see this with the new Cat 8.12's, as theyre getting much better performance than the 8.10's.
  • cleeve
    jaydeejohnWould have loved to see this with the new Cat 8.12's, as theyre getting much better performance than the 8.10's.

    Yeah, unfortunately the 8.12s just came out and this article has been a long time in the making. :)
  • cleeve
    DoltronIf you couldn't get Tray Tools to work with the Sapphire card why not use another program? Instead of giving up and coming up with a lame conclusion.

    Mostly because the card didn't seem to be able to get past 690 MHz core without problems in the Catalyst Control Center, so there didn't seem to be much point in persuing overclocking much further.

    But for the sake of completeness I can give Rivatuner a shot this evening and see if anything changes. I'll let you know. :)