Meet The Radeon 4770

With 826 Million transistors, the RV740 GPU that powers the Radeon HD 4770 features a native 640 SP (128 five-wide vector units arranged in 8 SIMD cores) as opposed to the 640 cut-down-from-800 SP 4830. Among the other differences is the fact that the 4770 hooks into GDDR5 over a 128-bit memory bus at almost the same clock speed (producing just a little bit less bandwidth at half the pinout).

AMD reports average TDP to be about 80W, so despite the fact that this is a 40nm part that pulls a little less power for the same job than its older brothers, the Radeon HD 4770 still requires a 6-pin PCIe power connector. This isn't a huge amount of power, and AMD has single slot boards that fall in to this range. Of course, it likely gets a little more complicated at 40nm when you have less surface area to dedicate to heat transfer. Thus this is a dual slot part rather than a single slot part. Such is life.

So, rather than a totally killer single slot card with no power connector at $99, we've got a dual slot card with a power connector at $110. Not ideal, but we can work with that. Rather than the 40nm process, form factor or targeted design being the selling point, the real issue is going to be the competition.

We will be comparing the Radeon HD 4770 to the GeForce GTS 250 512MB (aka the 9800 GTX+) and the GeForce 9800 GT. These two cards sort of sandwich the Radeon HD 4770 in terms of price with the 9800 GT coming in at $100 and the GTS 250 512MB at slightly more than $120. So the question will continually be: does the extra +/- $10 make a difference.

This part essentially improves upon and usurps the position of the Radeon HD 4830. Word from AMD was that we should see the 4830 start to fall by the wayside. For our analysis we are including the Radeon HD 4830 and the Radeon HD 4850. Here's a breakdown of how the AMD hardware stacks up:

ATI Radeon HD 4770 ATI Radeon HD 4850 ATI Radeon HD 4830
Stream Processors 640 800 640
Texture Units 32 40 32
ROPs 16 16 16
Core Clock 750MHz 625MHz 575MHz+
Memory Clock 800MHz (3200MHz data rate) GDDR5 993MHz (1986MHz data rate) GDDR3 900MHz (1800MHz data rate) GDDR3
Memory Bus Width 128-bit 256-bit 256-bit
Frame Buffer 512MB 512MB 512MB
Transistor Count 826M 956M 956M
Manufacturing Process TSMC 40nm TSMC 55nm TSMC 55nm
Price Point $110 $130 $100

It's worth noting that the bandwidths of the 4770 and the 4830 are 51.2GB/s and 57.6GB/s respectively.

We have also tweaked a couple of our tests to better target the ~$100 segment. The biggest change was with our Crysis test where we dropped everything down by one quality level ending up with all mainstream settings except for gamer shaders. The other was just a small tweak: not pushing things beyond the high quality default settings in Age of Conan (though we did enable 4xAA).

In the middle of testing, we accidentally let our copy of Left 4 Dead update itself rendering our benchmark un-timedemo-able. Thus we have to leave Left 4 Dead performance out of this article, but we can say that at the highest quality settings the 4770 is capable of playable framerates at up to 1680x1050.

Our test setup is still the Intel platform with a top of the line CPU in order to remove any other bottlenecks from the system. These performance numbers show the potential the graphics card has to offer. If the rest of a system is unable to achieve performance levels along the lines of what we show here, then it doesn't matter what graphics card we plug in at this price: it will end up performing pretty much the same as any other option (at the system bottleneck level). These tests show the potential of a graphics card when the potential of the graphics card makes a difference. That said, most Phenom II, Core 2, and Core i7 systems will be very close to these numbers at the common resolution of 1680x1050 with the tested hardware; the fast system/CPU generally only becomes a factor at lower resolutions or with multiple GPUs.

Test Setup
CPU Intel Core i7-965 3.2GHz
Motherboard ASUS Rampage II Extreme X58
Video Cards ATI Radeon HD 4770
ATI Radeon HD 4830
ATI Radeon HD 4850
NVIDIA GeForce GTS 250
NVIDIA GeForce 9800 GT
Video Drivers 9.4, 9.4 Beta for 4770
ForceWare 185.68
Hard Drive Intel X25-M 80GB SSD
RAM 6 x 1GB DDR3-1066 7-7-7-20
Operating System Windows Vista Ultimate 64-bit SP1
PSU PC Power & Cooling Turbo Cool 1200W

Without further ado, here's the performance numbers.

Index Age of Conan Performance
Comments Locked


View All Comments

  • Luminair - Tuesday, April 28, 2009 - link

    whose power readings are correct, anandtech or hardocp
  • tomoyo - Tuesday, April 28, 2009 - link

    Ya the power readings are all over the map here. Xbitlabs has the 4770 with MUCH lower idle power than the 4830. While Anandtech shows the 4770 consuming more at idle. Something is strange with the power measurements of this graphics card.
  • RagingDragon - Wednesday, May 13, 2009 - link

    Everbody is measuring system power consumption at the wall socket instead of GPU power consumption; therefore, power comsumption varies since each site uses different CPU's, motherboards, powersupplies, etc.
  • kmmatney - Tuesday, April 28, 2009 - link

    I haven;t read the XBitLab article yet, but they normally use a technique where they directly measure the power draw of the card, rather than the entire system. I believe they are more accurate. I'll read their article now, since it probably has OVERCLOCKING!!! How could AT miss that - at least OC and run one benchmark so we can see the percentage gain.
  • tomoyo - Tuesday, April 28, 2009 - link

    Looking at HARDocp's power results, they don't make sense. Supposedly their system without video card is 46 watts from the wall and system with 4770 is 53 watts? There's no way the idle is 7 watts for the video card including psu inefficiencies. This contradicts greatly with xbitlabs and anandtech...which already both contradict each other regarding power usage. Basically none of the power results can be trusted which is really annoying.
  • Korr - Tuesday, April 28, 2009 - link

    What a terrible piece. This is the exact reason why I rarely visit AT anymore. One resolution per game tested? No overclocking? One page of text to whine about the name?

    Please never review computer hardware again.
  • aeternitas - Wednesday, June 10, 2009 - link

    I guess simple graphs are still too complex for some people to understand. lol
  • JarredWalton - Tuesday, April 28, 2009 - link

    Don't know about you, but I see resolution scaling charts showing 1280x1024, 1680x1050, and 1920x1200.
  • 7Enigma - Wednesday, April 29, 2009 - link

    B...B...But the other resolutions don't have pretty bar graphs! Seriously don't feed these trolls. If they can't even READ the review before blasting it for things included, they don't deserve to get answered.
  • AtenRa - Tuesday, April 28, 2009 - link

    Personally I don’t mind about the naming of the card.

    Although this card is the first 0.40nm and it’s the first time the GPUs are manufactured using a smaller process than CPUs, I don’t see the reviewers to be exited for this product. I mean this review feels like it was made in 15 minutes just for having it in the site. No O/C and no CrossFire results. Have you ever wandered if 2 x 4770 are faster than a 4890?? how about 3 x 4770 ??

Log in

Don't have an account? Sign up now