Anand's retail Radeon 9500 Pro review - much faster!!!....

[Bambers]

Formatting gives a nice fast clean OS. Pity it only lasts for 2weeks before windows fills itself up with random crap again

 

[John Reynolds]

I don't believe reformatting is ever a necessity (unless there's data corruption somewhere), and I'm certainly going to blame ATI if it's causing problems on their video card.

Let me tell you a little story. Kept hearing how fast the GF2s were, so even though I was quite happy with my V5 at the time (fall of '00), I bought a GF2 Ultra for $200 from a Nvidia fan site webmaster. Plopped it into my ME gaming rig after carefully removing the 3dfx drivers and doing a fairly thorough registry cleanup. Game after game puked. Formatted, reinstalled, and everything then worked fine. Eventually put the V5 back in for another 9-10 months due to the Ultra's (which was a review sample from Nvidia) poor 2D, AA, and texturing, but that's another story (but, boy, it suh did run Q3 fast).

Anyways, my only conclusion must be that 3dfx's drivers were pristine and Nvidia's suck.

 

[Tahir2]

Lets not conclude anything other than it is the nature of the OS.

Then we can all be friends and stop pointing fingers one way and then the other

 

[Tahir2]

Oh and another ting I just formatted and only... only after testing for 48 hrs without one lockup did I install the SiS AGP drivers... and what do you know.. in 24 hrs I have had one freeze and 3 BSOD's.

Anyone know a way to remove the SiS AGP driver without formatting..

 

[MuFu]

Actually, I heard that the NV31 was going to be a performance (~$200), while the NV34 was going be the mainstream part.

Yup, that's right.

The clockspeed is not the only difference from the NV30 - otherwise I'm sure they wouldn't bother with the specific nomenclature. It is no doubt closely related but there is much functionality they will have stripped from the core to bring the cost down. For example, I very much doubt NV31 will support FP16 and FP32 natively.

MuFu.

 

[KimB]

Yup, that's right.

The clockspeed is not the only difference from the NV30 - otherwise I'm sure they wouldn't bother with the specific nomenclature. It is no doubt closely related but there is much functionality they will have stripped from the core to bring the cost down. For example, I very much doubt NV31 will support FP16 and FP32 natively.

MuFu.

On the support for FP16 and FP32, the NV30 doesn't really support FP32 natively. That is, its performance is half that of FP16 performance with FP32 enabled. That means, essentially, two FP16 pipes combine for one FP32 pipe.

I don't really think that the NV30 is capable of any less than one FP32 madd per pixel pipeline per clock, though.

Personally, this is what I see:

1. The GeForce FX's primary problem right now is heat. Reducing the frequency considerably should vastly reduce heat consumption, removing the need for the "FX Flow" unit, and possibly the external power connector. These should reduce the cost by a significant amount.

2. The changes may not be superficial. For example, it may just be a move to the low-K dielectrics, to further tackle the heat/power consumption problem.

3. Possibly different packaging. The current NV30 packaging is rumored to have a ton of pins, possibly enough for a 256-bit memory interface. The NV31 may be repackaged to remove the excess pins.

Anyway, I really hope that programming functionality is not reduced any. Performance, fine. But reducing programming functionality just holds games back (read: Radeon VE, GeForce4 MX).

 

[Hyp-X]

The last time I reformatted was because I was repartitioning the drive for Linux.

I've just recently installed Linux which involved repartitioning my drive - yet no reinstall was neccessary.

Actually my current Win2000 installation survived a motherboard change and it had the following video cards in it: GF3 classic, GF4Ti4200, Parhelia prototype, Parhelia final, R8500LE, R9700Pro (current one). And I swapped these a lot.

The only issue I had was after the Parhelia driver install OpenGL didn't work on nVidia cards - could be resolved by a driver reinstall.

If I reinstalled the OS with every card change I'd have almost zero time left to do work.

Compatibility issues are normally caused by installing crap programs like codec packs (number one problem source), adware programs (dl managers, file swapper programs), desktop curtomization programs / screensavers, and a lot of other junk people insist on installing.

The primary reason people experiencing less problems after reinstall is that they do not reinstall these junks as soon they retest the thing that didn't work before. So they become contributors to the reinstall myth.

 

[darkblu]

Yup, that's right.

The clockspeed is not the only difference from the NV30 - otherwise I'm sure they wouldn't bother with the specific nomenclature. It is no doubt closely related but there is much functionality they will have stripped from the core to bring the cost down. For example, I very much doubt NV31 will support FP16 and FP32 natively.

MuFu.

On the support for FP16 and FP32, the NV30 doesn't really support FP32 natively. That is, its performance is half that of FP16 performance with FP32 enabled. That means, essentially, two FP16 pipes combine for one FP32 pipe.

I don't really think that the NV30 is capable of any less than one FP32 madd per pixel pipeline per clock, though.

no, that only means gfx's implementation of fp32 is twice slower than fp16 on the same chip. it does not imply anything about the 'combining of narrower pipelines into wider ones' whatsoever.

 

[Dio]

Compatibility issues are normally caused by installing crap programs like codec packs (number one problem source), adware programs (dl managers, file swapper programs), desktop curtomization programs / screensavers, and a lot of other junk people insist on installing.

What he said, double.

The thing I really love about modern versions of Windows is that once you've installed a bunch of cards, if you swap them around it recognises them instantly - not even needing the drivers again. There have been a few exceptions - most common is that a particular driver set won't stick to a particular card / machine combo and I've had to use an old driver - but these have been really rare in the last couple of years.

I'm very much a minimalist-installer though - for security reasons as much as anything else. I think this hypothesis is therefore a good one...

 

[KimB]

no, that only means gfx's implementation of fp32 is twice slower than fp16 on the same chip. it does not imply anything about the 'combining of narrower pipelines into wider ones' whatsoever.

Of course it does. This is essentially what nVidia reps have stated. When NV30's are available to test, though, all will be made clear.

 

[Pete]

I'm not sure how to make a Ghost of Windows with standard MB drivers (so you can just install your MB's drivers after re-Ghosting with the clean image), but you can certainly create a Ghost image with Standard VGA drivers for easily swapping video cards. I'd imagine more people swap video cards then motherboards.

I've also read that it's pretty much required to reinstall Windows after a MB change, so I'm surprised that Chalnoth's system is still stable. Impressive.

 

[Dio]

I've also read that it's pretty much required to reinstall Windows after a MB change, so I'm surprised that Chalnoth's system is still stable. Impressive.

It depends on the OS and the change. In my experience, if the MB chipset hasn't changed it will always work fine, and if the MB chipset vendor and CPU type hasn't changed it usually works. With more radical changes it is a bit roll-the-dice.

For win9x, you could get it to work every time by going into the device manager and removing all the motherboard devices - most importantly, the PnP BIOS. When you change the HD you then run the manual hardware detect, it detects the new BIOS, then picks up the new MB bits.

I haven't upgraded a Win2k machine without also upgrading the harddrive in a long while - the 2k machines tend to be closed-box and the XP machines haven't got to upgrade point yet. I can't remember exactly how I got that working, but I suspect I did it the same way.

 

[darkblu]

no, that only means gfx's implementation of fp32 is twice slower than fp16 on the same chip. it does not imply anything about the 'combining of narrower pipelines into wider ones' whatsoever.

Of course it does. This is essentially what nVidia reps have stated. When NV30's are available to test, though, all will be made clear.

you apparently missed the crux of my post - getting twice higher performance at fp16 does not imply that fp32 is done through combining two fp16 pipes.

 

[T2k]

I've also read that it's pretty much required to reinstall Windows after a MB change, so I'm surprised that Chalnoth's system is still stable. Impressive.

It depends on the OS and the change. In my experience, if the MB chipset hasn't changed it will always work fine, and if the MB chipset vendor and CPU type hasn't changed it usually works. With more radical changes it is a bit roll-the-dice.

For win9x, you could get it to work every time by going into the device manager and removing all the motherboard devices - most importantly, the PnP BIOS. When you change the HD you then run the manual hardware detect, it detects the new BIOS, then picks up the new MB bits.

Agreed.

BTW, try to change the boot-up SCSI controller to a different one...

 

[Nagorak]

Oh and another ting I just formatted and only... only after testing for 48 hrs without one lockup did I install the SiS AGP drivers... and what do you know.. in 24 hrs I have had one freeze and 3 BSOD's.

Anyone know a way to remove the SiS AGP driver without formatting..

Try using scanreg in DOS to go back to an earlier registry version.

The newest SiS AGP drivers suck, try the ones that came on the CD with your motherboard.

 

[Althornin]

I've also read that it's pretty much required to reinstall Windows after a MB change, so I'm surprised that Chalnoth's system is still stable. Impressive.

It depends on the OS and the change. In my experience, if the MB chipset hasn't changed it will always work fine, and if the MB chipset vendor and CPU type hasn't changed it usually works. With more radical changes it is a bit roll-the-dice.

For win9x, you could get it to work every time by going into the device manager and removing all the motherboard devices - most importantly, the PnP BIOS. When you change the HD you then run the manual hardware detect, it detects the new BIOS, then picks up the new MB bits.

I haven't upgraded a Win2k machine without also upgrading the harddrive in a long while - the 2k machines tend to be closed-box and the XP machines haven't got to upgrade point yet. I can't remember exactly how I got that working, but I suspect I did it the same way.

Yeah, under win9x the best thing to do is simply go intot he registrty and delete the "Enum" key. removes all installed devices then reboot, and it will redetect everything. And i mean EVERYTHING.
The problem with winNT/2k/xp is usually IDE controller drivers, not chipset ones. you can usually work around this by taking the machine and removing the IDE controllers and installing "standard ide controller" drivers from microsoft (slower performance) and then switch mobos, but there is no nice and simple enum key removal. NT/2k/XP is a lot pickier about device drivers tahn win9x is. At work, we make images all the time (Ghost) and we wish there was a simple fool proof method to ensure win2k/xp images transfer over to new hardware like there was with 9x. But there isnt, that we have found.