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.
This came up a week ago at nV News here, and it's spread really fast (whether or not it came up somewhere else first, I'm not really sure--might have been Futuremark or something, but we had it before the Korean site that supposedly started it). The thread has a ton of feedback, by the way, so it's something to consider. (oh, and/. mods, links are nice too.:) )
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:
Increases the voltage to the core slightly.
Loosens the memory timings on the DDR (yes, video cards have memory timings just like motherboards and system RAM).
Sets the default clocks to 5950 levels (no RivaTuner or Coolbits necessary).
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.
Windows XP still uses the 2D-only GDI, as does Longhorn (the successor to XP to be launched sometime in 2004, if memory serves). Originally, Longhorn was going to contain a 3D GDI (using DX, I assume), but according to my friend Element at Winbeta, the development of the new GDI was taking too long to be implemented in Longhorn. So, a 3D GDI in Windows has been pushed back until Blackcomb (to be launched between 2006 and 2008, and almost no information whatsoever is available about it except the 3D GDI bit).
Another question though, which I will ask my 3D coder buddies--what impact will pixel shaders and vertex shaders have on these 3D windowing systems? It seems to me that it could either add significant amounts of detail or significantly increase speed by taking advantage of long shaders (assuming the shader implementation is good.. *cough NV30 would be bad for this cough*). Still though, it could explain why PS/VS are being pushed so heavily in DirectX.
Damnit, I hate it when I forget to log in (yes, that AC comment is mine).
JC made this comment back at the NV20 launch. The game was designed to maximize the featureset of the NV20 and early in the development was being tested on a GF3. He says: "
GF4-MX will still run Doom properly, but it will be using the NV10 codepath with only two texture units and no vertex shaders. A GF3 or 8500 will be much better performers." No mention of the GF4's performance, as this was at the time of the GF4 launch (if memory serves again).
But seriously, if you needed a GF4 to really run Doom 3 with every possible detail, why would ATI be switching to a 2 year cycle between major chipset revisions? You do not need all the speed that you can get right now EXCEPT if you are using significant amounts of antialiasing and anisotropic filtering. JC is smart enough not to design a game engine that will run well and look better than the previous generation of engines on only less than 5% of computers out there.
Considering the Xbox is just a higher-clocked GeForce 3 and a Pentium 3 733... it'll run Doom3 fine.
And the alpha... Christ, it was version ZERO POINT ZERO TWO. The next person to say, "Well the alpha ran _______________ on my computer" gets slapped.
"FSAA, BTW, is tremendously difficult for video-cards to do (because they're actually rendering at 2x or greater resolutions): There is no current video card that could dream of doing even Urban Terror (a Q3 mod) at 1600x1200 with FSAA at acceptable frame-rates."
No. This is bogus. The Kryo2 chip, along with the GF1/2, had a form of FSAA where it basically rendered the image at a much higher res than it was going to be displayed, but these don't exist anymore (at least not in the R300/NV30/NV25/R200). I BELIEVE this is called supersampling--don't quote me on this, I'm not a coder and don't care too much about FSAA modes (I have a GF3. I can't use AA in ANYTHING but the oldest games.). Supersampling takes a much larger performance hit, but a lot of people regard it as looking better than the newer method. This newer method is called multisampling--it actually renders the image multiple times, offsetting it each time. This is why color compression has become so important. 4x MSAA COULD take up to four times the memory bandwidth of normal rendering, but with adequate color compression, you could get it down to two times or 1.5 times the bandwidth. This is part of the reason why nVidia went with a 128-bit bus on the GFFX--it thought it had good enough color compression.
Anyway, moving right along, there are two forms of MSAA (multisampling antialiasing)--ordered grid and rotated grid (once again, do not quote me on this).
So basically, FSAA ain't as simple as rendering at 3200x2400 and reducing that to 1280x960 anymore.
I can't use the first PCI slot even with my GeForce 3. You can't use it with a GeForce 1, either, if I recall correctly--the heatsink (just the HEATSINK) is too big. Now, if you had some insanely shaped card, it would fit, but the whole PCI slot cooler thing is making something over nothing.
Now the noise... THAT is the problem. But, apparently, a closer-to-retail revision of the card that [H]ardOCP got had a much softer cooler than the original reference card. Plus, if Gainward makes its 7dB GFFX cooler, that's not going to be a problem at all. Still, it won't be on a 5800 Ultra after all, which is kinda too bad.
Yes, 5800 Ultra is gone. 5800 will be for sale, but at 400/800, it's not going to win any speed awards (unless drivers manage to improve its performance by 20%, which won't happen by the time R350 (successor to the Radeon 9700) comes out). The chips used for the Ultra will still be used, however, in the QuadroFX 2000. ATI's R300-based FireGL cards are still poor performers, and even with relatively poor drivers from nVidia, the FX 2000 will beat it handedly (the Quadro4 is beating the FireGL cards as well, according to Tom's Hardware).
But, it's not just a rumor anymore. When it first came to [H], everyone regarded it as BS. It was a rumor posted on a board that spread incredibly rapidly. But, apparently it's been confirmed by either OEMs or nVidia itself to those with good contacts. BFG has stopped taking preorders, AFAIK, because...
"According to an e-mail John Malley sent out a couple of days ago, BFG is concerned that pre-sales may exceed their allocation of units."
So, yes, the 5800Ultra is gone. Oh well. NV35 in June, according to some.
Slashdot Mirror
they CAN'T, because of S3TC support. have to have a license for it, and that implies that it can't be open-sourced.
is here, courtesy of Andy Mecham, NVIDIA's Linux driver guru.
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.
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.
Windows XP still uses the 2D-only GDI, as does Longhorn (the successor to XP to be launched sometime in 2004, if memory serves). Originally, Longhorn was going to contain a 3D GDI (using DX, I assume), but according to my friend Element at Winbeta, the development of the new GDI was taking too long to be implemented in Longhorn. So, a 3D GDI in Windows has been pushed back until Blackcomb (to be launched between 2006 and 2008, and almost no information whatsoever is available about it except the 3D GDI bit).
Another question though, which I will ask my 3D coder buddies--what impact will pixel shaders and vertex shaders have on these 3D windowing systems? It seems to me that it could either add significant amounts of detail or significantly increase speed by taking advantage of long shaders (assuming the shader implementation is good.. *cough NV30 would be bad for this cough*). Still though, it could explain why PS/VS are being pushed so heavily in DirectX.
Damnit, I hate it when I forget to log in (yes, that AC comment is mine). JC made this comment back at the NV20 launch. The game was designed to maximize the featureset of the NV20 and early in the development was being tested on a GF3. He says: " GF4-MX will still run Doom properly, but it will be using the NV10 codepath with only two texture units and no vertex shaders. A GF3 or 8500 will be much better performers." No mention of the GF4's performance, as this was at the time of the GF4 launch (if memory serves again). But seriously, if you needed a GF4 to really run Doom 3 with every possible detail, why would ATI be switching to a 2 year cycle between major chipset revisions? You do not need all the speed that you can get right now EXCEPT if you are using significant amounts of antialiasing and anisotropic filtering. JC is smart enough not to design a game engine that will run well and look better than the previous generation of engines on only less than 5% of computers out there.
Considering the Xbox is just a higher-clocked GeForce 3 and a Pentium 3 733... it'll run Doom3 fine. And the alpha... Christ, it was version ZERO POINT ZERO TWO. The next person to say, "Well the alpha ran _______________ on my computer" gets slapped.
No. This is bogus. The Kryo2 chip, along with the GF1/2, had a form of FSAA where it basically rendered the image at a much higher res than it was going to be displayed, but these don't exist anymore (at least not in the R300/NV30/NV25/R200). I BELIEVE this is called supersampling--don't quote me on this, I'm not a coder and don't care too much about FSAA modes (I have a GF3. I can't use AA in ANYTHING but the oldest games.). Supersampling takes a much larger performance hit, but a lot of people regard it as looking better than the newer method. This newer method is called multisampling--it actually renders the image multiple times, offsetting it each time. This is why color compression has become so important. 4x MSAA COULD take up to four times the memory bandwidth of normal rendering, but with adequate color compression, you could get it down to two times or 1.5 times the bandwidth. This is part of the reason why nVidia went with a 128-bit bus on the GFFX--it thought it had good enough color compression.
Anyway, moving right along, there are two forms of MSAA (multisampling antialiasing)--ordered grid and rotated grid (once again, do not quote me on this).
So basically, FSAA ain't as simple as rendering at 3200x2400 and reducing that to 1280x960 anymore.
Now the noise... THAT is the problem. But, apparently, a closer-to-retail revision of the card that [H]ardOCP got had a much softer cooler than the original reference card. Plus, if Gainward makes its 7dB GFFX cooler, that's not going to be a problem at all. Still, it won't be on a 5800 Ultra after all, which is kinda too bad.
But, it's not just a rumor anymore. When it first came to [H], everyone regarded it as BS. It was a rumor posted on a board that spread incredibly rapidly. But, apparently it's been confirmed by either OEMs or nVidia itself to those with good contacts. BFG has stopped taking preorders, AFAIK, because...
"According to an e-mail John Malley sent out a couple of days ago, BFG is concerned that pre-sales may exceed their allocation of units."
So, yes, the 5800Ultra is gone. Oh well. NV35 in June, according to some.