3.5.7. InsideTNC FX5900XT Tuning – what is it?
This is a BIOS in the mvktech.net collection from InsideTNC videocard (See Unit 2.8.7). It seems to be a modification of the BIOS from a top-end card because the card itself has 2.2 ns chips. It often dramatically improves overclockability of the Hynix memory (up to a gain of +100MHz!), but probably through setting higher timings, because the performance increases not proportionally to the frequency gain. This BIOS is not recommended for the Samsung memory. You can download it here.
3.6. What’s the overclocking record for the 5900XT?
There’s no info yet about overclocking experiments on the 5900XT with extreme cooling, which would show “the limit of the card’s abilities”. By my personal estimates, such frequencies would be somewhere near 700/1050MHz.
The record for now is 650MHz GPU (Vgpu=2.0V) and about 1050MHz memory (Vmem=3.43V) without BIOS re-flashing. Actually this is my own result, as I held a world record for 5900XT performance in 3DMark 2001 SE (27229 points).
Without volt-modding, you can expect a maximum of about 530/975MHz (and up to 1050MHz by flashing BIOSes from other cards with relaxed timings).
4. Modifications
4.1. Does the card need additional cooling?
It’s desirable that you install additional fan(s) to blow at the card – this is the bare minimum for good overclocking. In fact, this helps the 5900XT card more than a replacement of the regular cooling system with a more efficient one (but without setting up the blowers).
4.2. What’s better: a single heatsink or individual heat-spreaders (on the graphics core and memory chips independently)?
Individual heatsinks are better as the core and memory have different temperature conditions.
4.3. Can I leave the memory naked, without heatsinks?
Yes. BGA-packaged DDR SDRAM is tolerant to temperature. If there’s no great overheat (you can touch the chip without burning your finger) – there’s no need to install heatsinks. I found memory heatsinks useful only after memory voltage mod – with some big volts on it chips became much hotter and its overheating was limiting factor in overclocking: only 980MHz with naked chips and 1040-1050MHz with copper heatsinks.
4.4. Is water-cooling of the core rewarding?
I carried out an experiment under ultimate conditions: the 600MHz core frequency, Vgpu = 1.9v. There was just a 10MHz difference between the regular cooler from Sparkle + additional blowers and Koolance EXOS water cooling (of course, watercooling won that test but clock gain isn’t too impressive).
4.5. What if I remove the protective cover from the core?
Although the graphics processor is cooled down more efficiently, you don’t usually get any overclocking advantages by removing the lid. At best, this gain will be something like 10-15MHz (in my personal experiment, it was 13MHz, from 615 to 627MHz). I don’t think you should do that with the standard cooler. In other cases, do it at your own risk. You won’t be worse off, but you lose your warranty (and unlike with the RADEON 9x00 you won’t have a chance to glue the lid back without losing the warranty-compliant looks of the product).





