The performance problem of fresh video drivers for AGP video cards based on Radeon HD 46xx series chips

Here is a short story from life - but at the same time it is a story about the “blue screen of death”.

Based on considerations of reasonable frugality, I still use the computer purchased in the first (and not the second) half of the two thousandth, and therefore equipped with the AGP bus, not PCI-E. For the time being, I was happy with the NVidia GeForce 7600 GS vidyuha, all the more so since they never produced anything better on NVidia chips for sticking in AGP: the eight-thousandth series was already on PCI-E. However, in mid-May, faced with the need for hardware acceleration of a video player, which, when playing a number of high-quality H264-encoded 720p-videosI started to stutter unacceptably, I involuntarily recalled: let NVidia CUDA technology on AGP be unavailable, but its ersatz alternative DXVA 1 is supported (for example, in CoreAVC ) on all ATi Radeon HD chips - two thousandth, three thousandth, four thousandth, five thousandth and newer series.

It was then that I drew attention to the vidyuha SAPPHIRE HD 4650 AGP , which is based on the Radeon chip of the same name. I was further delighted by the fact that (according to Wikipedia) the NVidia GeForce video cards of the seven thousandth series were released in 2005 and 2006 and only supported DirectX 9 with the shader 3.0 model - as for the more recent four thousandth series Radeon HD chips, then they are shown there as having been released much later (in 2008) and supporting DirectX 10.1 and the shader model 4.1.

A little thought, I decided to pay ≈tri hundred rubles more to buy a little more than a fancy chip of the same 4600-th generation and a part of the PowerColor HD4670 1GB of DDR3 the AGP . The difference between 4650 and 4670 is not particularly large, but still the latter is slightly steeper.

_{I almost didn’t punish myself with this toughness, because the PowerColor HD4670 1GB DDR3 only works on AGP 8x, and the SAPPHIRE HD 4650 also works on AGP 4x, respectively, if I couldn’t remember correctly that my home computer supports AGP 8x, then money would be wasted (or, for example, there would be a need to negotiate a moneyback). Fortunately, my computer is not so old not to understand AGP 8x.}

Of course, the first thing after installing the video card should have thought about updating the drivers. If only because the hardware acceleration in Firefox 4 requires drivers of at least version 8.741 . Alas, on the PowerColor website, this driver offers vintage drivers version 8.712, no more!

The logical way out of the deadlock seemed to be the ability to download and install the latest AMD drivers. Unfortunately, this opportunity turned out to be even more horrifying and hopeless, because the new AMD drivers, when installed on this vidyuha, immediately cause  BSoD ("blue screen of death") during the initial boot of the system. It turns out this problem (i.e., incompatibility of AGP versionsRadeon with new AMD drivers) has been a hell of a long time (since last year), but no one was really able to fix it. I myself could not even suspect such nastiness in advance, so before I accidentally looked for a story about it on the AMD user forum, I managed to try a little less than a dozen different versions of drivers that have been released over the past six months “unsuccessfully, of course; and transported without any use until one in the morning. The situation is complicated by the fact that AMD does not at all provide the AGP version of its previous drivers (but only the AGP version of the latest driver, not a damn good).

Of the desired driver (for October 2010), involuntarily had to search in the archive on the Sapphire website: As you recall, Sapphire also produces AGP video cards based on Radeon chips. True, this archive turned out to be organized perverted: if a file does not appear on one of the proposed “mirrors”, then it is not suggested to choose another “mirror”; and since the rest of the page is already blocked by the "modal window" of the dialog, all that remains is to click the "Back" button and search again for what you want, re-hammer the model of video and operating system.

Having achieved what he wanted, he first adjusted the brightness and contrast, and then he started to compare the performance of the “new” video card (PowerColor HD4670 1GB DDR3 AGP) with the old one (NVidia GeForce 7600 GS) using a browser performance test with HTML5 Canvas .

This test yielded 11 fps or 12 fps (frames per second) in Firefox 4 on top of Direct3D 9 on top of the NVidia GeForce 7600 GS. And now it has produced 16 fps or 17 fps in Firefox 4.0.1 over Direct3D 9 over PowerColor HD4670 1GB DDR3 AGP. Consequently, video performance has increased ≈1.5 times, if only this is not the effect of the difference between Firefox 4 and Firefox 4.0.1 (which I allow myself to doubt). The video player also started using DXVA.

It would seem to me good. However, this story does not have a satisfactory happy ending. See for yourself how bad she looks. A large international corporation continues to provide on its site only those new versions of video drivers that immerse AGP optionsin the "blue screen of death", and this is known only informally. Only last year's drivers from alternative sources are suitable for use on AGP. At the same time, AMD doesn’t care, because AGP-options are produced by third companies in the third world, and officially “Radeons” come out only in the form of PCI-E.

Could this go on forever?

No, of course not. Sooner or later, something new will appear in the new video drivers (for example, support for 3D video on the Internet, which NVidia already has ), and then all buyers of AGP-video (and I along with them) will be left out of the bright future. An alternative would be to upgrade to PCI-E -purchase of a new motherboard, new processor, new RAM, new video card, new cooler, finally.

Forced upgrade.