Slashdot Mirror
Turning A FX5900 Into A FX5950 Ultra, Tool-Free
A reader writes "Some very interesting details coming from various tech sites such as ExplosiveLabs and 3DChips that shows it is possible to turn a GeForce FX5900 into a FX5950 Ultra (which is NVIDIA's top of the line video card chipset currently available) through simply using the FX5950 Ultra BIOS on the FX5900 video card."
Basically, if you do this, don't be surprised if your card becomes toast a shortwhile after. Umm, no. If you've ever overclocked, you'd understand that hardware starts getting errors when it's pushed too far. In a video cards' case, it will begin rendering things incorrectly.
The errors start happening LONG before hardware burns up, and is soon as the card is set to a slightly lower speed, the errors disappear.
Basically, if you get your card to where it gives no errors, and are able to keep it around the same temperature, it won't have any troubles.
hey!
FWIW the only reason I went with nVidia for the new computer I recently put together was for their Linux support. I dual boot the machine and I do game in Windows, but I wanted nVidia's proven track record of supporting their hardware with Linux drivers. The Twinview support for Linux is nice too, gotta have my dual monitors in Linux too.
I also wasn't looking for the top of the line vid card so since I wasn't dropping a wad of cash I did't care that I wasn't getting bleeding edge FPS performance.
anyone read the full forum posting and see that the q3a framerate dropped, along with some other things?
I think you're right, they seem to be basing the fact that it's an "Ultra" completely off the fact that the Windows control panel says "NVIDIA GeForce FX 5950 Ultra". The people on the message board post benchmarks, and they're all lower save for one - you can overclock the card slightly higher when it's running the Ultra's bios... w00t.
In many cases, yes. The last two or three generations of cards from NVIDIA and ATI have largely been made up of only two or three physically different cards per generation, per manurfacturer (with different bios's installed though ocasionally a resistor or two had to be soldered in a different location as well). In ATI's case, many of their recent "budget" cards could have extra pipelines unlocked by merely using a hacked driver.
"Revenue loss from slashdotters who value their time much less than their money: $ NOT MUCH"
None really, after all, how many slashdotters who might have bought an ATI card might buy the second most expensive card nvidia makes now? They wouldn't sell the 5900 if they didn't make money of it. They will make a profit on every fringe overclocker who jumps on the bandwagon.
I believe the SX was the 486 without the math coprocessor.
Exactly. On the AthlonXP there are jumpers that change the XP into a MP. There is another set that changes it to a Mobile Athlon. More jumpers change a 2500+ to a 3000+ (multipliers), etc, etc. Producting 10 different cores would make the processors cost about $1000 a piece. Selling 3200+s for more than they cost and 2500+s for less make AMD profitable (well, not really. but it's the right idea :)
Anyway, your 2500+ is only guarenteed to run at 1700MHz (or whatever). If it runs at 2200MHz, great. If not, tough shit. If you buy a 3200+, though, then it had better run at 2200MHz (200x11, right?). If not, then you can complain.
Selling underclocked 3200+s as 2500+s allows AMD to sell bad 3200+s instead of throwing them away. The reason that some overclock well is because AMD tests a few out of one batch, and if any are bad AMD brands them _ALL_ as 2500+s. So it's highly likely that you really have a 3200+, but, again, don't count on it.
My other car is first.
More likely, NVidia bins its chips like all other hardware manufacturers. Cards are manufactured to be the same, and are then stress-tested. Those that make the cut are shipped out at the highest speed, those that don't get underclocked. Sure, you can clock a 5900 back up, but the chances of failure are much higher than with a real 5950 Ultra.
It's hard to be religious when certain people are never incinerated by bolts of lightning.
They have a binary driver, that driver may or may not load. Nvidia has an installer, which will detect your kernel version and compile an interface for it regardless of what linux version you use.
The net result is ATI's drivers won't run on 70% of linux systems out there, whereas Nvidias will run on everything including the 2.6 series kernels before they even became stable.
On the other hand the actual performance of nvidia's drivers have been going downhill with each and every release.
I have a Chaintech GeForce FX5950, and I love it. I don't care if someone can flash their card to upgrade it. But I also realize that like all processors they are binned, and as such - my own card runs a bit cooler in the end.
This is mentioned in the article and is put down to relaxed memory timings decreasing performance. But the top overclock with the upgraded bios is higher, so it is worth it in the end.
It went both ways, actually. The 486SX was a DX with disabled/nonfunctional math coprocessor. The standalone math coprocessor you could buy for the SX, on the other hand, was a DX with disabled/nonfunctional CPU. Not a bad idea really.
Trust the Computer. The Computer is your friend.
That was exactly what the ExplosiveLabs post said.
According to Scott Mueller's Upgrading and Repairing PCs, the early SX's were DX's with the FPU mysteriously disabled. Later SX's really had no FPU at all.
And in another strange twist, a "487" was a full-blown 486DX with a different pin layout that disabled the main processor.
__CmdrTHAC0__
In Soviet Russia, Spanish Inquisition doesn't expect YOU!!
Before we get into the hack itself, we need to look at the chips and BIOSes involved. The 5900 cards use the NV35 chipset, and the 5950 uses the NV38 chipset. The two chips are very similar, but they are not exactly the same. This is not the Radeon 9500 to 9700 hack. In that situation, you had an R300 in both cards--here, you have to very similar chips. The differences between the NV35 and the NV38 are slight, at best, and as far as anyone knows, they have more to do with the cost of manufacturing than anything else (I've heard that 5900 cards are so cheap now simply because they are being dumped in lieux of 5950s).
So, where does that leave us? The BIOS hack. Essentially, it does three things to the best of anyone's knowledge:
So, the decrease in performance at the same clock speeds is due to the relaxed memory timings, but just like with anything else, you can get a higher overclock as a result.
HOWEVER--there is one potentially serious problem. Most people have reported that the 5950 BIOS flash has caused no change in the reported temperatures. Given what we know about the new BIOS and increased voltage, this makes no sense. I am, then, forced to wonder if the temperature diode becomes less accurate after the BIOS is flashed with the 5950 BIOS. No one has confirmed this, and since I don't have a 5900 to try it on, I can't either. However, it's something to keep in mind.
Finally, this is not newsworthy in the least. It's the same as people changing 9800 non-Pro BIOSes to those of 9800 Pros and getting better memory overclocks. It's nothing special or magical; you're not doubling the number of pipelines and the memory bus like you were with the 9500 to 9700 hack. However, it works (or seems to, at least), and it's pretty cool.
---
nV News
Actually, you got some things wrong. Firstly, the things that determine XP vs MP and Mobile vs Regular aren't jumpers; they're bridges. You have to connect them electronically by means of a pencil or rear window defroster kit.
/overclocker
And, the 2500+ runs default at 1833MHz.
Only affects overclocking. There is no GeForce FX to Quadro FX hack yet. Unwinder (the Russian genius who made just about EVERY GeForce and ATI hack of note and maintains RivaTuner) has said that it probably won't be possible, but you never know.
---
nV News
See what happens with chips is that every chip of a given type comes from the same fabrication process, same wafers. A given design of the P4 (like say the Northwood) ALL comes from the same place, regardless of speed. So, what happens? Does Intel just underclock lots of chips? No. They rate them.
Despite the amazing levels of controls, there are imperfections on silicon wafers, and imperfections in the etching process. Not every chip comes out the same. So when chips come off the wafer, they need to be tested and rated. Some fail outright, the just don't work at all. Those get tossed, or made into keychains or the like. Of the ones that DO work, they are tested for the maximum speed they'll reliably perform at and seperated into bins based on that. So off of a given wafer you can easily have chips that run anywhere from 1ghz to 2ghz and such.
Now, where underclocking comes in is a few cases:
1) Some companies tend to be conservative with their speeds. Intel is one of those. Generally speaking, their chips can really handle more than they claim. Intel is careful, though, and in the one case they weren't (certian 1ghz P3s) they got burned by chips that failed.
2) Sometimes, yeilds are just too good. Like you have a big demand for 1.6ghz chips, but most of what you are making runs at 2ghz or more. No problem, you take some from the 2ghz bin and underclock and mark them as 1.6ghz. They run slower just fine.
3) The chip runs at a higher speed, but has problems. Sometimes a chip will run faster, but parts of it fail to work prpoerly. So while 98% of the chip works fine at 2ghz, 1 unit just won't work past 1.5ghz. Can't really be selling chips that "mostly" work (remember how bad Intel got burne on the Pentiums with the FDIV bug) so it needs to be marked down.
Little real world example:
Back in the day of the Celeron A's, overclocking was real popular. Intel was having just great yeilds on their chips and most of their slow chips would really work much faster. So what you'd do is buy a cheap Celeron 300a, which was designed to run on a 66mhz bus, and run it on a 100mhz bus. This would bump the chip up to 450mhz. Basically, a system like this ran as fast or faster than a PII 450, and cost a hell of a lot less. Me and tons of friends did just this.
Well, the levels of success varied. My roomate at the time had a total and unqualified success. He dropped the chip in and it ran with no tweaking at all. As far as I know, he still has it in an anticillary system today. Basically, his chip was one from the 450mhz (or better) bin that had been marked down to meet demand.
I had less success. Mine I had to boost the voltage by about 20% to make it run stable at 450mhz. This I did and it worked fine... For about a year. Then my system started to have odd instabilities, crashing all over for no apparent reason. Went to the point of unusable in a very short time. The root of the problem was apparent when I had it calculate Pi and it got a slightly wrong answer. My chip was shot, and I had to get a new one. So while my chip could be made to run at 450mhz, it wasn't really capable fo taking it, and the stress eventually destroyed it.
Another friend simply never got it to work. Chip ran fine at 300mhz, but whenever he tried it at 450, the system just wouldn't POST. Tried cranking the voltage and all the tweaks he could think of, to no end. His chip was rated 300 for a reason, that's all it could do.
A similar situation existed with Intel's SX/DX chips. Basically, Intel found that a high number of chips had faulty math coprocessors. Thing was, the main unit worked fine, it was just the FP unit that was faulty. Well rather than throw the whole chip out, they'd just disable the math co and sell it as an SX.
So just because you can hack BIOS/microcode/whatever to make something run faster, doesn't mean it can handle it. Sometimes, it really is a faster chip underclocked, sometimes, it is clocked that speed for a reason. IT's a crapshoot. You also need to be careful since you CAN damage the chip doing it, like I did. No bigge for me, it was a Celeron that cost me like $80 and I got a year of use out of it. Be a much bigger deal if it was a $300 graphics card and you burned it out after a month.
Correct. The parent was ill informed.
No, I was well aware of the changes made. But the reason behind those changes was that those chips had flawed or damaged silicon. By disabing part of the features, they were able to resell them as "lower end" models instead of throwing them away.
Anyone in the fab business can tell you that more features on silicon == lower yield. Intel simply found a way to make a profit off of something that would otherwise have been worthless.
Javascript + Nintendo DSi = DSiCade
The parent post could well be correct. In software development, "time to market" is a huge factor. As a result a lot of features get developed without time for proper testing. It's not common to leave untested features in one release of a product which are disabled, test them later when time permits, and enable them as part of the next release. The obvious danger is that if the untested feature contains bugs, then.... doh!
He's talking about the GeForce 256 (or was it the TNT2?). It could be converted into a Quadro with a few CPU pin modifications.
The question is, how is their performance relative to the Windows drivers? NVIDIA's Linux drivers are 100% as fast as their Windows ones. Last time I checked ATI's drivers, they were half as fast as their Windows ones.
A deep unwavering belief is a sure sign you're missing something...
AMD's are only labeled like that because AMD has a more efficient processor. Intels can run the same data multiple times and not get the right output, but an AMD can nail it in one shot.
Karma: Good, or bust!
As a side note, my 2500+ Barton (overclocked) also runs 3 degrees Celsius cooler than my sister's 2400+ Thoroughbred (not overclocked) using the same copper heatsink and the same mobo, RAM, etc. to minimize variables.
the reason they are guarenteed at certain clock speeds, and yes, I know what im talking about, is because when a chip is cast it is cast on a wafer, the area of the wafer is not uniform as you go from the center to the edges of the wafre. When the process is being done the wafers have impurities which can exist in some of the chips cast on the wafer.
So what they do is they have machines that roll the new made chip through and test how high it is "safe" to clock it at a certain level. The ones that are more impure get sold as a lower ghz chip, the ones that are more pure get sold as higher ghz chips.
When you overclock, you are overclocking a chip that has failed its pre-test qualifications for the clock speeds you want to clock it to, which is why thye wont give you money on your warrentee if you tell them you fried your chip overclocking.
If you don't vote, you don't matter, so don't waste your time telling me your opinion
No but I sure remember those co-processor chips that DID magically make your 386 into a 486. I also remember those magic drills with which you could drill a hole in a single sided floppy and magically turn it into a double sided.
Those are good comparisons to the topic of this article, because both of those products had downsides similar to those of the 5900 mod.
- The 386 -> 486 chips gave you a faster processor, but not the other hardware to go along with it (e.g. a faster bus). The 5900 mod (as far as I can tell) gives you a 5950, but with only 128MB of RAM instead of 256MB.
- The floppy disk holes punches would sometimes let you use both sides of a disk (or use a double-density disk as high-density), but single-sided disks were only rated to be used on one side, and double-density disks had a different composition from high-density disks. Using a tool like this was asking to lose your data later on. I bought an old Roland sampler years ago whose previous owner had stored a bunch of data on HD disks (its drive didn't check for the hole), and they were all corrupted. The DD disks were all fine.
Products like these are really a bad way to spend your money. If you need an HD floppy, buy HD. If you really need a 5950, spend the extra $150 and get 256MB of RAM while keeping your warranty intact.
"...always new atoms but always doing the same dance, remembering what the dance was yesterday." -Richard Feynman
Yeah I actually tried that hack out - my friend from tech-junkie.com brought over his brand-new GF4 Ti4600 reference board that NVIDIA gave him for review, and I added a jumper to flip it between GF4 and Quadro. Yes that was me doing the soldering, and it took ages since the through-hole resistor leg was bigger than the surface-mount resistor pad ;) I'm sorry that the article is down but the site doesn't exist any more. Enough encouraging replies could get the article up on his private site though...
;)
Anyway the result was that the Windows drivers said we had a Quadro, but since my friend also had a Quadro reference board of whatever model is comparable to the GF4, we found that the real Quadro had extra OpenGL features that the fake one didn't. We tried BIOS swaps etc. and we never did get a Quadro... except for the one that NVIDIA already gave us
As an aside, the hacked GF4 is in the machine I use regularly at home and it's in front of me now. Still working perfectly, although I've never set it to Quadro since that would be a bit silly now wouldn't it... =)
Their website has drivers for every card they have ever made. I use a 6 year old Rage Pro in my firewall box, and an original Radeon DDR in my secondary machine I use for ripping DVD's and other time consuming activities. I installed the regular Catalyst drivers for the original Radeon I got in early 2001 that I installed on my 9700 Pro.
So in the Rage 128 days ATI had poor drivers, but that was years ago and that was not due to abandonment of a product. The company had a little trouble transitioning to 3D graphics cards, but so did lots of other companies. They eventually got their act together.
I am always amazed by these trollish fanboys. ATI used to be the only reliable company for drivers (1990-1998), and their image quality is and always has been top notch. Since the original Radeon came out, I think their support has been great.
I don't read or respond to AC posts