Radeon HD 4850 Vs. GeForce GTS 250: Non-Reference Battle

Overclocking The Asus EAH 4850 Matrix Using iTracker

When overclocking a garden-variety Radeon HD 4850, the ATI Overdrive utility will cap out at 700 MHz GPU speed. Fortunately, the BIOS on the Asus 4850 Matrix ups this to 800 MHz, allowing us to push our GPU past the stock 750 MHz speed of a Radeon HD 4870. And push it we did.

The ability to adjust the voltages on the GPU and memory will really break the chains usually imposed on most overclocking attempts. Increasing the voltage is a fundamental overclocking technique, and we probably wouldn’t have gotten past 700 MHz on the GPU core without it.

After doing a little research, we found that a stock Radeon 4870 has a GPU voltage somewhere in the neighborhood of 1.263 V. We thought this would be a good place to start for the GPU, but it wasn’t quite enough as we took it to 1.33 V for our maximum stable overclock.

After many attempts and settings, we cranked the Asus EAH4850 MT from a stock speed of 625 MHz GPU/1,986 MHz memory to 764 MHz GPU and 2,013 MHz memory. This GPU overclock is very good for a Radeon HD 4850--an increase of 139 MHz over stock and surpasses even the Radeon HD 4870 stock clock speed by 14 MHz. Unfortunately, the memory wasn’t nearly as accommodating, with a slight overclock of 19 MHz DDR at 2.03 V.

We actually got much higher GPU clocks out of the card, but after repeated testing, it was clear we had to back off a bit to reduce temperatures and prevent visual artifacts. But the final 764 MHz GPU and 2,013 MHz memory clocks were rock solid throughout all of our benchmarks and stress tests.

One of the great things about the iTracker utility is not only its overclocking capabilities, but its underclocking features as well. The utility allows us to set an extremely low clock speed and voltage for 2D use in Windows, and a high clock speed and voltage for 3D use in the same profile.

However, we did come across one slight complication when underclocking: after we had some crashing problems launching Fallout 3, we did a ton of testing to diagnose the issue. It turns out, if the 2D clock is set too low--say at 300 MHz--some applications will falter when ramping up to 3D clock speeds. Both Fallout 3 and Far Cry 2 demonstrated this behavior. Once we boosted the speed to 400 MHz, the problem no longer occurred. 

After all is said and done, we can’t say enough about the ability to modify voltages in the iTracker utility for the sake of both overclocking and underclocking. It’s a great tool with which to work, and after we got over some of the poor interface design decisions, we truly wish everybody made a tool like this for their products.

Create a new thread in the US Reviews comments forum about this subject
This thread is closed for comments
58 comments
    Your comment
    Top Comments
  • Time to ban DemonHorde 665, the abuse of the English language is making all dead spelling teachers spin in their graves.
    17
  • Other Comments
  • In the second picture of the 4850, the card can be seen bent due to the weight.
    0
  • The Gigabyte would be more effective with 2 fans.
    0
  • rags_20In the second picture of the 4850, the card can be seen bent due to the weight.


    Hi rags_20 -

    Actually, the appearance of the card in that picture is caused by barrel or pincushion distortion of the lens used to take the photo. The card itself isn't bent.

    / Tuan
    3
  • demonhorde665... try not to triple post.
    looks bad... and eratic. and makes the forums/coments system
    more clutered than need be.

    ps. your not running the same bench markes as Toms so your not really comparable.
    yes, same game and engine, but for example in crysis, the frame rates are completely different from the start, through to the snowey bit at the end.

    pps. are you comparing your card to there card at the same resolution?
    5
  • Hi,

    I've been looking for a comparison like this for several weeks. Thank you although it didn't help me too much in my decision. I also missed some comments regarding the Physix, Cuda, DirectX 10 or 10.1 and Havok discussion.

    I would be very happy to read a review for the Gainward HD4850 Golden Sample "Goes Like Hell" with the faster GDDR5 memory. If it then CLEARLY takes the lead over the GTS 250 and gets even closer to the HD4870 then my decision will be easy. Less heat, less consumption and almost same performance than a stock 4870. Enough for me.

    btw. Resolutions I'm most interested in: 1440x900 and 1650x1080 for 20" monitor.

    Thank you
    2
  • Under the test setup section the cpu is listed as core 2 duo q6600, should it not be listed as a quad? Feel free to delete this comment if it is wrong or when you fix the erratum.
    -1
  • Why a Q6600/750i setup? That is certainly less than ideal. A Q9550/P45 or 920/X58 would have been a better choice in my opinion (and may have exhibited a greater difference between the cards).
    4
  • zipzoomflyhighand no the Q6600 is classified as a C2D. Its two E6600's crammed on one die.


    No, its classified as a C2Q. E6600 is classified as C2D.
    6
  • ZZFhigh,

    Directly from the article on page 11:
    Quote:
    Game Benchmarks: Left 4 Dead Let’s move on to a game where we can crank up the eye candy, even at 1920x1200. At maximum detail, can we see any advantage to either card? Nothing to see here, though given the results in our original GeForce GTS 250 review, this is likely a result of our Core 2 Quad processor holding back performance.

    Clearly this is not an ideal setup to eliminate the processor from affecting benchmark results of the two cards. Most games are not multithreaded, so the 2.4Ghz clock of the Q6600 will undoubtedly hold back a lot of games since they will not be able to utilize all 4 cores.

    To all,

    Stop triple posting!
    3
  • Quote:
    The default clock speeds for the Gigabyte GV-N250ZL-1GI are 738 MHz on the GPU, 1,836 MHz on the shaders, and 2,200 MHz on the memory. Once again, these are exactly the same as the reference GeForce GTS 250 speeds.


    Later in the article you write,
    Quote:
    or the sake of argument, let’s say most cards can make it to 800 MHz, which is a 62 MHz overclock. So, for Gigabyte’s claim of a 10% overclocking increase, we’ll say that most GV-N250ZL-1GI cards should be able to get to at least 806.2 MHz on the GPU. Hey, let’s round it up to 807 MHz to keep things clean. Did the GV-N250ZL-1GI beat the spread? It sure did. With absolutely no modifications except to raw clock speeds, our sample GV-N250ZL-1GI made it to 815 MHz rock-solid stable. That’s a 20% increase over an "expected" overclock according to our unscientific calculation.


    Your math is wrong. A claim of 20% over clock on the GV-N250ZL-1GI would equal 885.6 MHz. 10% of 738MHz = 73.8 MHz. So a 10% overclock would equal 811.8 MHz. 815 MHz is nowhere near 20%. In fact, according to your numbers, the GV-N250ZL-1GI barely lives up to its 10% minimal capability.
    3
  • This whole article is completely invalid and the results are skewed because, as was documented on tweaktown, Catalyst 9.3 performance is much lower compared to 9.2. Catalyst 9.4 reclaims some of those performance losses, but 9.2 is still a bit better, if you compare the two analyses. Redo these tests with 9.2 drivers.
    1
  • weakerthans4Later in the article you write,Your math is wrong. A claim of 20% over clock on the GV-N250ZL-1GI would equal 885.6 MHz. 10% of 738MHz = 73.8 MHz. So a 10% overclock would equal 811.8 MHz. 815 MHz is nowhere near 20%. In fact, according to your numbers, the GV-N250ZL-1GI barely lives up to its 10% minimal capability.


    No what he is saying is this- Gigabyte claims that the extra copper in the PCB will allow for a 10%-30% further increase compared to how much a standard cards speed can be raised by overclocking. So saying that a standard card oc's to 800MHz which is a 62MHz increase, Gigabyte is claiming a 6.2 (10%) to 18.6 (30%) MHz further increase on top of that. So "technically" a 20% increase would have put it at 816.4 MHz, only 1.4MHz more than the 815MHz he acheived.
    1
  • Time to ban DemonHorde 665, the abuse of the English language is making all dead spelling teachers spin in their graves.
    17
  • personally I think it's the Zalman accounting for a bulk of the 20% extra and not the couple ounces of copper. That cooler rocks.
    3
  • To the reviewer: Good article, but you forgot two things:

    The GTS 250 is a 9800GTX+ is a 9800GTX is -also- an 8800GTS 512. So this...3 year old card is still running strong.

    Also, Gigabyte's Ultra Durable is for two functions, overclocking and obviously, durability. Yes, it will overclock better. But it also will probably never stop functioning.

    From someone who's gone through numerous motherboards and graphics cards with minimal overclocking on either, that means a lot more than performance.
    -5
  • it is known that nvidia cards tax the cpu less. So if a title is cpu bound than the nvidia card will usually come out on top. Thats why you see them performing similarly when resolutions increase and when you move away from cpu dependency
    -6
  • KyleSTLWhy a Q6600/750i setup? That is certainly less than ideal. A Q9550/P45 or 920/X58 would have been a better choice in my opinion (and may have exhibited a greater difference between the cards).


    It's in the specs but I should have stressed the point: I overclocked the Q6600 to 2.7 GHz, it was plenty quick for these cards.
    0
  • RamarTo the reviewer: Good article, but you forgot two things: The GTS 250 is a 9800GTX+ is a 9800GTX is -also- an 8800GTS 512.


    Not exactly. The 8800 GTS at least sported diffrent clockspeeds. I also believe it was on a larger die, if memory serves.
    0
  • tacoslaveit is known that nvidia cards tax the cpu less.


    Is it? If so, please provide some proof of that statement as I haven't seen evidence of that.
    1
  • weakerthans4Later in the article you write,Your math is wrong. A claim of 20% over clock on the GV-N250ZL-1GI would equal 885.6 MHz. 10% of 738MHz = 73.8 MHz. So a 10% overclock would equal 811.8 MHz. 815 MHz is nowhere near 20%. In fact, according to your numbers, the GV-N250ZL-1GI barely lives up to its 10% minimal capability.


    You misunderstand Gigabyte's claim. As universalremonster points out, they're alaiming a 10% increase in overclocks over other GTS 250's, not claiming that all of their cards will overclock 10% over stock clocks.
    0