Remember when ATI made crappy chipsets that no one bought, and all AMD systems were built with NVIDIA or even VIA chipsets? Yeah, that memory is fading for us too.

Today we get to officially unveil the 785G, the latest in integrated graphics chipsets by AMD for AMD. As the name implies, there is very little separating 785G from its predecessor (780G) and we’re quite ok with that. In fact, the only reason not to get a 780G is addressed by the 785: the new chipset supports 8-channel LPCM audio over HDMI. The lack of TrueHD and DTS-HD audio bit-streaming remains but neither is this feature supported on other chipsets. We also see a jump from HDMI 1.2 to 1.3 standards. Update: This story is developing in real time but AMD is telling us that 8-channel LPCM over HDMI is not supported in the final chipset. More info here.

The rest of the major improvements are strictly related to video playback duties. We jump from Universal Video Decoder (UVD) 1.0 on the 780G to UVD 2.0 capabilities on the 785G. The video decode engine supports decoding multiple HD streams (useful for picture-in-picture on a Blu-ray movie) and additional post processing effects if you are not a big video purist. Also new is the 785G's ability to perform detail enhancements on the fly.

The graphics side has not really changed that much. The new GPU is based off the RV620 core and is roughly the same size/complexity as the old one. In other words we get the same 55nm node process and almost the same amount of transistors, just a tad over 205 million.The 785G’s PowerPlay technology is improved with the core constantly adjusting clock speeds based on GPU utilization with a 60MHz target when idling compared to 500MHz at full load. AMD adds DirectX 10.1 support but stream processor count and clock speeds have not changed. Thus gaming performance remains mostly unchanged.

And in this corner: Intel

AMD is targeting Intel’s G41 with this 785G, and by simply by adding 8-channel LPCM and UVD 2.0 support it is no longer at a feature deficit for the HTPC audience. In fact, we cannot find one area where the 785G comes up short compared to the G41.

Intel’s G41 is a cost-reduced version of Intel’s G45 chipset. You lose PCIe 2.0 support (1.1 only) and integrated Gigabit Ethernet (only 10/100). Considering Intel recommends mating the ICH7 Southbridge to the G41 for cost purposes, you also lose RAID support, four USB ports, and two 3GB/s SATA ports compared to the AMD SB710.

  AMD 790GX AMD 785G AMD 780G Intel G45 Intel G41 NVIDIA GeForce 9300
CPU AMD Socket-AM2 AMD Socket-AM2 AMD Socket-AM2 Intel LGA-775 Intel LGA-775 Intel LGA-775
Manufacturing Process 55nm 55nm 55nm 65nm 65nm 65nm
FSB N/A N/A N/A 800 / 1066 / 1333MHz 800 / 1066 / 1333MHz 800 / 1066 / 1333MHz
Memory Controller N/A N/A N/A 2 x 64-bit DDR2/DDR3 channels 2 x 64-bit DDR2/DDR3 channels 2 x 64-bit DDR2/DDR3 channels
Memory Speeds Supported N/A N/A N/A DDR2-800/667
PCI Express 22 PCIe 2.0 lanes 22 PCIe 2.0 lanes 22 PCIe 2.0 lanes 16 PCIe 2.0 lanes 16 PCIe 1.1 lanes 20 PCIe 2.0 lanes
Graphics Radeon HD 3300 Radeon HD 4200 Radeon HD 3200 GMA X4500 GMA 4500 GeForce 9300 mGPU
Core Clock 700MHz 500MHz 500MHz 800MHz 800MHz 450MHz Core /
1.2GHz Shader
Shader Processors 8 (5-way) 8 (5-way) 8 (5-way) 10 10 16
Full H.264/VC-1/MPEG-2 HW Decode Yes Yes Yes Yes No Yes

The Radeon HD 4200 is where all the action is with this update but the more things change, the more they remain the same. We still have the traditional Northbridge/Southbridge layout with the NB supporting a 2GHz HyperTransport link compatible with the latest Socket AM3 processors. The 22 PCI Express 2.0 lanes are still around along with the four lane PCI Express 1.1 lanes supporting the A-Link Express connection between the two bridges.

The NB splits its 22 lanes between a single x16 link for graphics and six x1 links for expansion slots and onboard peripherals. Technically, the 785G chipset can't split its x16 graphics link into two x8 links for CrossFire X operation. The 785G supports hybrid CrossFire operation, but only with the older HD 3450/3470 video cards. We recommend skipping this feature, for now.

  AMD SB750 AMD SB700 Intel ICH10 Intel ICH7 NVIDIA GeForce 9300
Additional PCI Express None None 6 x1 PCIe 1.1 4 x1 PCIe 1.1 None
USB 12 ports 12 ports 12 ports 8 ports 12 ports
SATA (300MB/s) 6 ports 6 ports 6 ports 4 ports 6 ports
PATA 2 channels 2 channels None None 1 channel
RAID* RAID 0/1/5/10 RAID 0/1/10 RAID 0/1/5/10 None RAID 0/1/5/10
HD Audio Interface Yes Yes Yes Yes Yes
Ethernet Not Integrated Not Integrated Intel Gigabit LAN Intel 10/100 LAN (Gigabit through external controller) NVIDIA Gigabit LAN
Northbridge Interface 4 lane PCIe 1.1 4 lane PCIe 1.1 DMI 10Gb/s each direction, full duplex DMI 10Gb/s each direction, full duplex N/A, Single Chip Solution

The SB710 is essentially an updated SB700 with ACC (Advanced Clock Calibration) support, or looking at it a different way, a cost reduced SB750 without support for RAID 5 arrays. The SB710 has six 300MB/s Serial ATA ports with RAID 0, 1, and 0+1 support plus a single IDE port supporting two drives so JMicron and Marvell are held at bay on these systems. Twelve USB ports and a high-definition audio interface round out the SB710 as native GbE support is still missing.

It would have been nice to see next-gen USB and SATA support but that will have to wait for the SB8xxx updates next year. However, compared to the ICH7 mated with the G41, the SB710 is positively feature rich at the same price point.

We could continue on with subtle hints about why this chipset is a better alternative than the Intel G41 or its 780G/790GX siblings, but the bottom line is that AMD has taken an evolutionary step with this chipset, one for the better. That said, let's see how it performs against the other chipsets in its class.

Test Setup


View All Comments

  • Shadowmaster625 - Wednesday, August 5, 2009 - link

    Since when have intel integrated graphics been in the same ballpark as the 7xx? Before today, every benchmark I saw AMD beat Intel by about 50-200% (Without overclocking or sideport crap.) Now all the sudden many of these benchmarks are showing an edge of only 20%, and this is comparing the 785G vs the G41. What is going on here? I think you need to do a more detailed review and comparison vs Intel's topline model, GMA 4500MHD(?) and nvidia 9300 as well. Reply
  • TA152H - Tuesday, August 4, 2009 - link

    This could have been summed up in a simple paragraph, instead of page after page of nothing. Still, I guess you have to prove the points.

    It's pretty disappointing, really. The chipset doesn't represent a clear advantage over the 790GX, despite the author's best effort to distort facts and compare a overclocked 785 with a nominally clocked 790GX. It's always annoying when an author already has an idea of what he wants to present, and then finds way to do it. Better to go with an open mind and let readers make up their own mind.

    The same applies with the G41. Another lame attempt by the author to distort the article to make a preconceived point. Since you show the 790GX, shouldn't you show the G45? Guess not, it might make the pre-conceived purpose of the article less clear.

    AMD makes a crappy processor compared to Intel, not a pretty good one. Everything is relative. It's really a zero sum game. So, we have GREAT (Nehalem), VERY GOOD (Penryn), and PRETTY GOOD (AMD). Where's the bad? What's pretty good compared to? What's it better than. It's the worst of the three lines, even compared to Intel's last gen. In other words, AMD has the bad. Someone has to do it. Not everyone can be 99 percentile. Someone's got to be 1 percentile for the 99 percentile folk to exist. Not that Nehalem is 99, and AMD is 1. Probably more like 80 and 20. AMD processors are still usable, for sure, and I still think make good packages because of their superlative chipsets, but the processor on it's own merit has no reason to exist, except within this context and that of competition. It's worse than Nehalem at everything, by a lot, and is the same size. That's not pretty good, it's pretty bad.
  • bruce24 - Friday, August 7, 2009 - link

    re: "The same applies with the G41. Another lame attempt by the author to distort the article to make a preconceived point. Since you show the 790GX, shouldn't you show the G45? Guess not, it might make the pre-conceived purpose of the article less clear."

    I was also wondering why he would only show the G41. In the article he says "The direct price competitor", but if I go to, I can find multiple G45 boards in the same price range as the 785G.

  • Spoelie - Tuesday, August 4, 2009 - link

    Would just like to chime in on some test subjects that are left untouched

    *Is UVD decoding still limited to AVC profile L4.1 (the one used in bluray)? The competition (nvidia) fully supports profile L5.1, which ensures that they can accelerate *every* video. With ATi it's either hit or miss, there are videos out there that use it.

    *ATi has serious issues with their SATA implementation, mainly AHCI mode.. are they fixed in SB710? I'm thinking not. Refer to your colleagues at techreport..


    SB700: "We quite literally see more of the same in the SB700's Serial ATA controller. The port count here is up to six, but they're basically six of the same ports you get on the old SB600" ->">

    SB750: "Unfortunately, AMD's longstanding issues with AHCI Serial ATA controller configurations persist in the SB750, all but forcing users to run the south bridge in plain old IDE mode. That's not the end of the world, but IDE mode doesn't support Serial ATA perks like hot swapping and Native Command Queuing." -> benchmarks in following link are IDE mode ->">
  • cghebert - Tuesday, August 4, 2009 - link

    The comments in TR's 785g review can shed some light on your SB AHCI questions:

    Comment by Prototype:
    "I think it's a non-issue. The out-of-box Windows 7 and Vista AHCI drivers work just fine with the southbridge, it's just the vendor drivers from AMD (and bundled by the motherboard vendors) that cause subpar performance.

    Which you can avoid by, you know, just not installing them.

    It's not like they add any functionality you don't already have by using Microsoft's excellent driver. In my experience Microsoft's generic drivers tend to be more stable and less buggy than vendor drivers anyway, a result of the fact that hardware vendors couldn't write decent software to save their lives, not even Intel.

    The hidden issue is the CPU utilization of the USB drivers, really. Note how both SB710 boards use 4 times as much CPU time as the ICH7 USB driver.

    As far as ICH7 AHCI is concerned, Intel doesn't have support for AHCI in their ICH7 "Base" driver, but if the motherboard manufacturer exposes AHCI in the BIOS, Windows Vista and 7's generic AHCI driver by Microsoft can be used for the device. (And for Windows XP, the Intel AHCI driver's .inf can be modified to add the PCI product ID and loaded during the installation process.)"

    There is more information in the comments if you want to check into it further.
  • mmaenpaa - Tuesday, August 4, 2009 - link

    It seems that this feature is mostly forgotten. Even AMD/ATI is not talking too much about it. I do remember testing it maybe a year ago with X1250 chipset and there were too much problems (yeah, I did try to find a solution, but it propably would have taken more than 10 minutes, so I gave up :-).

    Now, just last week I tested with Gigabyte 780G mb and HD 4670 PCIE card and it simply worked. I had three monitors connected (XP PRO).

    I do believe this is quite a nice feature and if you are using ATI cards it is practically free.

  • HollyDOL - Tuesday, August 4, 2009 - link

    Hmm, no clue why, but despite what author says, both movie screenshots for Intel/AMD solutions look almost exactly same (difference being they are not the same frame). Is it just due to JPEG picture quality loss or the difference between AMD/Intel playback is practicaly uncomparable?
    Viewed on Eizo FlexScan panel, so there shouldn't be any quality reductions on my screen...
  • MrCommunistGen - Tuesday, August 4, 2009 - link

    I would have liked to have seen some SB700 (on the 780G) vs SB710 (on the 785G) benches on the USB/HD benchmarks. Reply
  • Kibbles - Tuesday, August 4, 2009 - link

    On page 6, the 4th graph is a duplicate of the 3rd. Reply
  • JarredWalton - Tuesday, August 4, 2009 - link

    Fixed, thanks. Reply

Log in

Don't have an account? Sign up now