Gigabyte's Radeon HD 4650: Are AGP Graphics Still Good Enough?

Test Settings And Benchmarks

We're bringing our old ASRock Dual939-SATA2 out of retirement for these tests. This motherboard is ideal for AGP comparisons because it sports both a full-speed PCIe x16 slot as well as a full-speed AGP 8x slot. This will allow us to compare the AGP Radeon HD 4650 against its PCIe counterpart to see if the aging bus is causing any performance loss.

In addition to the PCIe Radeon HD 4650, we'll test the long-standing AGP flagship, ATI's Radeon HD 3850. Also, for those of you thinking about upgrading an older card, we're throwing in a Radeon X700 Pro for reference. The Radeon X700 Pro's performance is roughly comparable to that of the Radeon 9800 Pro, Radeon X1300 Pro, and GeForce 6600 GT, so users of all these cards will have a good idea what they have to gain with an upgrade.

Our CPU of choice this time around will be AMD's Athlon 64 X2 3800+, one of the first dual-core CPUs available. We already know that most modern games benefit from at least a dual-core CPU, so we thought the Athlon 64 X2 3800+ would be a good baseline.

We were also planning to test an Athlon X2 5600+ in order to gauge which games were CPU- and GPU-limited. Unfortunately, our Athlon 64 X2 5600+ CPU didn't agree with the ASRock Dual939-SATA2 CPU expansion card, so we'll have to explore what a faster processor can do in the aforementioned follow-up.

As for the operating system, we've chosen good-old Windows XP 32-bit, which we think will be the typical operating system for an older box sporting AGP connectivity. In Part 2, we may explore what Vista can do. Let us know your thoughts on which you'd prefer to see tested in the comments section.

Note the Frankenstien-esque collection of RAM -- we considered cleaning this up, but it's probably a good representation of a lot of older rigs that have been upgraded over time. While the system sports a full 2 gigabytes of the stuff, it all has to run at the speed of the lowest common denominator, and that's DDR 333. Even so, this isn't likely to cause a large performance hit compared to running 400 MHz memory.

The only setup issue of note was the driver situation. The most recent Catalyst drivers don't seem to work with the new Gigabyte GV-R465D2-1GI AGP, so we used the ones that came bundled with the card. They worked fine with no issues, but buyers should know it might be a while (if at all) before the official Catalyst package supports this new graphics card. Other than that, we should mention that the antiquated Radeon X700 card isn't supported anymore, so we had to use older Catalyst 9.3 drivers.

Swipe to scroll horizontally
Test System Configuration
CPUAthlon 64 X2 3800+Socket 939, 2.0 GHz, 2 x 512 KB Cache
MotherboardsASRock 9393dual-SATA2ULI 1695 chipset, BIOS v2.3
RAMPatriot EP PC3500 (CL2.0-3-2-5)  (1 x 1 GB) DDR-400 at DDR-333 CAS 2.5-3-3-7Kingston KVR PC3500 (CL3.0-3-3-8) (2 x 512 MB) DDR-400 at DDR-333 CAS 2.5-3-3-7Samsung PC2700 (CL2.5-3-3-7) (1 x 512 MB) DDR-333 at DDR-333 CAS 2.5-3-3-7
GraphicsRadeon HD 4650 AGP - 600 MHz GPU, 400 MHz Memory, 1 GB DDR2Radeon HD 4650 PCIe - 600 MHz GPU, 400 MHz Memory, 512 MB DDR2Radeon HD 3850 AGP - 668 MHz GPU, 829 MHz Memory, 512 MB DDR3Radeon X700 AGP - 425 MHz GPU, 432 MHz Memory, 256 MB DDR2
Hard DriveWestern Digital WD1200JB 120 GB, 7,200 RPM, 8 MB cache, UltraATA/100
SoundIntegrated Audio
NetworkIntegrated 100 Mb Networking
PowerNextherm PSU460 460W, ATX 12V, EPS 12V
OSMicrosoft Windows XP Professional 32-bit SP2
GraphicsRadeon HD 4850 PCIe, Radeon HD 3850 AGP: Catalyst 9.6Radeon X700: Catalyst 9.3Radeon HD 4650 AGP: 8.62-090423a080042EAti2mtag / XP
Swipe to scroll horizontally
Benchmark Configuration
3D Games
Left 4 DeadVersion 1014, Custom THG Benchmark Test Set 1: Medium Settings, High paged memory Test Set 2: Medium Settings, High paged memory, 4x MSAA, 8x AF
CrysisPatch 1.2.1, DirectX 9, 32-bit executable, benchmark tool Test Set 1: Low Quality, Medium Textures, Shadows, Shaders, Water, Sound Test Set 2: Low Quality, Medium Textures, Shadows, Shaders, Water, Sound, Physics
Tom Clancy's H.A.W.X.In-game benchmark Test Set 1: Lowest Settings, DirectX 9 Test Set 2: Highest Settings, DirectX 9
World in ConflictPatch 1009, DirectX 10, timedemo Test 1: High Details, No AA / No AF Test 2: Very High Details 4x AA / 16x AF
Fallout 3Custom THG Benchmark Test Set 1: Medium Settings Test Set 2: High Settings, 4x MSAA, 8x AF
Far Cry 2DirectX 9, 32-bit executable, benchmark tool Test Set 1: Low Quality Test Set 2: Medium Quality, HDR Enabled
  • amdfangirl
    Don't they have an AGP HD 4670?
  • drealar
    *Raise up hand*
    I'm still using an AGP HD 2600XT :D.
    Good to see that Gigabyte and actually Sapphire too coming up with another card for AGP. But frankly, we (most AGP users in my neighborhood ) are keeping our dollars for now since upgrading to a new system is the most likely option in 2010.

    Another thing is, from the benchmarks, it got me worried if I ever upgrade my GPU since most of the result shows CPU bottleneck. If these are the result for a dual core Athlon, then it must be worse for my single core P4 HT :(

    Oh BTW, are you reading the power consumption correctly?
    "Even the Radeon HD 3850 and its GDDR3 memory peak at a mere 7 W more than the PCIe Radeon HD 4650 under load."
    194 - 178 = 16W. So either the graph is wrong or you mis calculated.
  • zinabas
    amdfangirlDon't they have an AGP HD 4670?
    a quick search for "agp 4670" on newegg gave me and (Core clock: 750 mhz, 1gb ddr3)

    I own a pcie version of it and I really think before part 2 of this article goes up you should acquire one.
  • rambo117
    amdfangirlDon't they have an AGP HD 4670?
    yah, i just saw that on the egg today. Must have just come out.
    i liked the article btw. ouch at the performance numbers though xD
  • zinabas

  • zinabas

    links aren't working
  • rambo117
    yah, this is the one i saw:

    looks pretty nifty. iceq FTW!
  • eddieroolz
    AGP is like Windows XP, it just never dies :D

    Anyways, it's interesting to read something about legacy ports for a change. I never even figured out how to work with computer hardware when AGP was around, so I never got to mess around with it.
  • one-shot
    Most of these benchmarks seem to show a CPU limitation across the board. I owned that same CPU, but used a PCI-E 7900GS with it at the time. I think if the CPU limitation was eliminated, we would see a better overall comparison. Also, if there is an AGP 4670, it would be beneficial to older machines to see the benefits of that GPU. It might show the two cards being equal due to the CPU limitation we're encountering.
  • masterjaw
    I don't think there are lots of people out there who would still hold on to their AGP slots when PCI-e cards are becoming so cheaper and the performance is way better. Even doing a low-end system upgrade would still kick older system's ass all the way to the north pole. But still, another option for those who can't get over the AGPs. They must've been clearing up some old stocks.