News
 

Bookmark and Share

(0) 
Our colleagues from the on-line store theFastest.ru told us that the modification of Sapphire RADEON 9500 with 128MB graphics memory into RADEON 9700 sometimes turned out a failure. Having carried out this experiment with quite a few graphics cards, they discovered that in 30% of all cases the cards worked unstably after the hardware modifications. Moreover, almost nothing depended on the BIOS version (Sapphire or Hercules RADEON 9700) and the working frequencies: these 30% of the cards worked incorrectly in any case with 8 pipelines enabled. The “incorrectness” implies some kind of “rippling effect” in 3D applications made by the falling out pixels. Each graphics card showed its own picture, and sometimes even the whole sets of pixels were missing on the screen.

Now let’s try to study the case carefully, to find out what’s happening. Judging by numerous messages in the Forums of well-known web-sites (such as Rage3D) and by the feedback we received by e-mail, many of you have already undertaken the hardware modification of their RADEON 9500 into RADEON 9700 despite the risks.

The results of these modifications were not always that great: in some cases there appeared artifacts in 3DMark2001, DX9-demos, etc. In most cases they took the form of chess fields built by the blocks of missing pixels or separate pixels. Here are some screenshots with the missing pixels (click to enlarge):

   

What does it all mean? The forum people had two suppositions about the origin of this phenomenon.

Version 1. The soldering damages the chip and the wafer. However, those of you who modified the cards with the help of a soldering-iron and then discovered some artifacts, could get rid of them by returning everything to the initial condition, i.e. by returning the resistor to its place and reflashing the chip’s own BIOS.

The other way to modify your card with the help of conductive glue also had its advantages and disadvantages. Again, if there appear any artifacts after you modify the card and reflash the RADEON 9700 BIOS, they can be eliminated by washing off the glue and reflashing the original BIOS. Certainly, the performance of the card gets lower after that as there are only 4 pipelines working.

If the hardware modification with the soldering or glue technique could damage the chip so that some of its blocks appeared non-functional any more, the whole thing could be impossible to restore. As a result all artifacts remained even if you returned to the initial 4 pipelines.

For example, the experiment carried out by pclab.pl site was a success, despite the awful quality of the soldering:

So, it looks as if artifacts are not caused by the physical interference or chip damage, but by enabling 8 pipelines.

In fact, we should describe in greater detail how the modification carried out by soldering or with the conductive glue enabled all 8 pipelines. Then things will be clearer.

All chips based on R300 architecture feature different DeviceID. DeviceID is a kind of “passport”, which is unique for the chips versions used. By R300 based chips this DeviceID is not locked in the chip but is written into it from the BIOS every time the PC starts and initializes the graphics card. And the BIOS of RADEON 9700 Pro, RADEON 9700, RADEON 9500 Pro and RADEON 9500 carry different DeviceIDs. The drivers read the DeviceID from the chip registers and then get the information about the number of working pipelines. For instance, when the driver sees the DeviceID for RADEON 9500, it enables 4 pipelines. And if it sees the DeviceID for RADEON 9700 – 8 pipelines.

So, what does the resistor actually do then? When the resistor stands in its initial position on RADEON 9500, it prevents a part of DeviceID from reprogramming. And even if you take your RADEON 9500 based graphics card and reflash the BIOS for RADEON 9700, the important part of the BIOS, which makes two DeviceIDs different from one another, will not get into the chip, so that the driver will still see a RADEON 9500. The hardware modification removes the protection, so that the new RADEON 9700 DeviceID could push the driver to enable all 8 pipelines.

So, all operations with the resistor are not connected with the pipelines or any other parts of the chip. They simply remove the write protection for DeviceID into the chip registers. This once again proves that the second version makes much more sense...

Version 2. To tell the truth, it is a very unpleasant one. Some RADEON 9500 chips are none other but the “defective” RADEON 9700 ones. During the chips testing, they check whether all its blocks are functioning properly and if some of them do not work, the chip is marked as defective. However, if the defective blocks are inside the pixel pipelines and there are 4 pipelines at the most that do not work well, then these 4 non-functional pipelines can be simply disabled and the chip can become a RADEON 9500.

How could we then explain the fact, that far not all the RADEON 9500 modification attempts end up in a failure?

Well, we think that there are not too many defective chips, first of all. The chips are either OK, or have such problems, which could not be hidden behind 4 disabled pipelines. In this case to maintain the RADEON 9500 sales on a certain level, they have to turn a number of fully-functional chips into RADEON 9500, especially since the production cost of both is the same. As for our experiments with RADEON 9500 128MB and RADEON 9500 64MB graphics cards, which ended up successfully, we seem to have been just lucky to have got these fully-fledged chips.

In fact, the current situation with RADEON 9500 reminds us of the RADEON and RADEON LE case. If you remember, these chips featured a disabled HyperZ unit and were clocked at lower frequencies. The cards based on RADEON LE cost less than the solutions on the fully-fledged RADEON. Most RADEON LE owners who managed to enable the HyperZ unit got a significant performance increase and didn’t have any problems with the graphics quality. However, after enabling HyperZ by some chips, we received different artifacts, which indicated the defectiveness of the HyperZ unit.

So, it is really sad but not all RADEON 9500 based cards appeared modifications-friendly.

Now we look forward to the software solution, which is currently being developed by Unwinder, and which will allow to enable all 8 pipelines of the RADEON 9500 via software. The software way will be much easier to use, so we expect the whole bunch of experiments to be reported about. Hopefully, we will be able to collect detailed stats about the successful and failed modifications after that and compare it with the chips marking, production date, etc.

Discussion

Comments currently: 0

Add your Comment




Latest News

Monday, July 28, 2014

6:02 pm | Microsoft’s Mobile Strategy Seem to Fail: Sales of Lumia and Surface Remain Low. Microsoft Still Cannot Make Windows a Popular Mobile Platform

12:11 pm | Intel Core i7-5960X “Haswell-E” De-Lidded: Twelve Cores and Alloy-Based Thermal Interface. Intel Core i7-5960X Uses “Haswell-EP” Die, Promises Good Overclocking Potential

Tuesday, July 22, 2014

10:40 pm | ARM Preps Second-Generation “Artemis” and “Maya” 64-Bit ARMv8-A Offerings. ARM Readies 64-Bit Cores for Non-Traditional Applications

7:38 pm | AMD Vows to Introduce 20nm Products Next Year. AMD’s 20nm APUs, GPUs and Embedded Chips to Arrive in 2015

4:08 am | Microsoft to Unify All Windows Operating Systems for Client PCs. One Windows OS will Power PCs, Tablets and Smartphones