Greekgol

Well, those crazy dudes at nVidia have certainly been busy. I know it's hard to believe, but this fall refresh marks the third product launch from them this year. What's that you say? How can that be? Okay folks, think back, way back, to the GeForce FX 5800 Ultra. Everyone remember that? It was delayed and delayed and launched this very year, in late January. Then there was the Spring launch of the mid-range 5600 and 5200 cards right alongside the announcement and release of the 5900 Ultra, the card that many thought is what the 5800 should have been. And here we are now, less than ten months later, with the 5950 Ultra and the 5700 Ultra.

WHQL-certified driver for GeForce FX 5-series, 6-series, 7-series, 8-series, and 9 series GPUs. Improved 3D performance and load times for GeForce 8 and 9 series GPUs in some DirectX 9 and OpenGL applications as a result an improved shader optimizer. Free download and instructions for installing the NVIDIA GeForce FX 5950 Ultra Video Card Driver for Windows 95, Windows 98, Windows ME, Windows 2000, Windows XP, Windows NT 4.0, Windows XP 64-bit, Windows Server 2003 64-bit, Windows Vista, Windows Vista 64-bit.

What's new about the 5950? Not a lot, really. It's running at higher clock speeds than the 5900 Ultra, but that's about it. Both cards sit in a 2D mode clocked at 300MHz core. The 3D Mode core speed of the 5900 is 450MHz whereas the 5950 is set to 475MHz - just a tiny bit higher. A larger difference is in the memory speed; 425MHz on the 5900 Ultra and 475MHz on 5950 Ultra. This allows for just over 30GB/s of available memory bandwidth, which is (for those uncertain of what that means) an awful lot of bandwidth.
Oh, the card looks a bit different, too. It's heavier thanks to a bulkier cooling solution and still requires a 4-pin power connection, big surprise. The heatsink design takes a page from ABIT's 'OTES' heatsink/fan combo, which gave you a card that took up two PCI slots, with the backplate on one acting solely for exhausting the air from the card. nVidia's changed it around a bit for the reference board and it's actually pulling air in from outside of the case with the second PCI backplate-slot-thing. However, as with the reference 5900 Ultra, expect many manufacturers to come up with their own designs for cooling.
With that big metal and plastic construct over the card I couldn't get a shot with sufficient lighting to actually be able to see it. Also, I'd like to note that the back of the card that I have has a plain black heatsink on it and not a cool green one with the nVidia swirl on it. Meh.
It's been about two months since our look at the MSI 5900 Ultra card and since then several new driver revisions have become available; in addition, several new games have been released that make for excellent testing. I've even done some minor tweaking to the test system:
Test Setup:

• AMD Athlon XP 2700+ @ 2.24GHz
  
• 1024MB Crucial PC2700 DDR @ 172MHz
  
• Asus A7N8X Deluxe (nForce 2 chipset), nForce reference drivers 2.45, AGP @ 72MHz
  
• ATi All-in-Wonder Radeon 9800 Pro 128MB, Catalyst 3.8
  
• nVidia GeForce FX 5950 Ultra 256MB, Detonator 52.16
  
• Windows XP Service Pack 1, DirectX 9.0b

The first thing you might notice is some minor overclocking. This might be new to some readers, and to others it might make you think, 'well, I could have clocked it higher than that.' I was just going for a minor speed increase and didn't feel like hunting down a really high but stable overclock. 6MHz on everything seemed fine to me. There's new driver revisions as well, but that's to be expected.
Unfortunately, because of the combination of the changes to the system (and improvements in driver releases) and the unavailability of the 5900 Ultra reference board I'm not able to do a direct comparison between the two. I have decided to test the 5950 Ultra against a 9800 Pro - in this case represented by the All-In-Wonder board - but expect a review soon on ATi's new 9800XT so you can see how the new parts stack up.
For the most part, the tests used here utilize the updated methods outlined in the A7N8X Deluxe Review. Obviously I'm no longer testing in 800x600 and I've added a few new tests as well.
As always, 'High Quality' (HQ) settings are plain 4X Anti-Aliasing and 8X Anisotropic Filtering. For the nVidia cards, in addition, 'texture sharpening' is enabled; for the ATi cards, the 'quality' Anisotropic Filtering mode is enabled. No other driver settings were changed and no video cards were harmed during the writing of this review.
Yes, I know, I said I wasn't going to be including the full results from 3DMark03 a ways back, but at this point I figure all of that 'driver optimizations' and 'IQ shortcuts' stuff is behind us. You can see that the 5950's lead is rather small.
One thing to note here - and this was the only strangeness I experienced while testing the 5950 - is that the card wouldn't run GLXS in 1024x768 'normal' mode. On the second test the test system would quite literally freak out: the GPU fan would start spinning like crazy, the hard drives would stutter, it was very odd. At first I figured this might be a result of the overclocking, but setting default speeds didn't fix it. My next guess was that I lacked a sufficient power supply for the testing, but disconnecting all extra peripherals still caused the problem.
No where else did such a problem appear, and as you can see, the card ran all the way through XS Mark with HQ settings enabled. I haven't yet heard back from nVidia on this, and on the chance that it was the power supply I've got a new one on the way. If that affects the problem then I'll be sure to post an update here.
Single-texturing lead still belongs to the 9800 Pro, by a small margin. With HQ settings enabled the positions reverse. The 5950 has a pretty decisive advantage with multi-texturing enabled, though.
The 5950 holds a very slight lead in its pixel shader score here, increasing to a larger amount with HQ settings enabled, likely due to the higher core and memory speeds of that card. Vertex shader scores are pretty much even, with scores, as always, virtually unaffected by Anti-Aliasing and Anisotropic Filtering.

Nvidia Geforce 740m Driver Windows 10

This is a somewhat bigger lead over the 9800 Pro than the 5900 reference board had, but it's still less than 1MPolys/sec. The 6FPS increase is fairly sizeable when you're sub-60fps, though.
No real surprises here, although it was interesting to see how close the two cards came in Chameleon Mark with the HQ settings.

Asylum Geforce Fx 5950 Ultra

I was just thinking to myself that since both Matrox and nVidia have some kind of test program like this available, maybe ATi should release one. They could call it Monkeymark and it could test how quickly your video card could render a banana. Hmm…
A few changes and additions have been made to the gaming tests.
Nothing has changed about the Jedi Outcast test except that I think it will soon be time to replace it with another OpenGL game. Doom 3 is practically 'right around the corner' so that would be excellent timing. I was surprised here to see such a large drop in performance in HQ mode for the 5950, whereas the Radeon is barely effected in comparison.
I picked up Jedi Academy a few weeks ago and it is a most excellent game. Unfortunately it did not come with a demo, as Jedi Outcast did (at least, not one that I could find in any of the pk3 files) so I recorded my own. I'll try to find a place to post it that can handle a lot of downloads, because I really don't want to wang my own website that way.
At any rate, you can see here that Jedi Academy is a bit tougher on your system than Jedi Outcast. Well, that's not entirely fair, a better description would be that this particular demo is tougher than the one used in JK2. But the pretty colors and special effects in JK3 make for a great test. Here we have the 5950 holding a definite advantage over the 9800 Pro, due most likely to the higher clock speeds.
I tested this newer Wolfensteine game (which is available for free on the official website, for those of you who don't know) using the same demo used by the Tech Report, used with permission of course. (Thanks Scott!)
Again the 5950 comes out on top, in this case with a pretty sizeable lead in all resolutions and quality settings used.
Only a very small lead held by the 5950 here; these results are much lower than some 5900 Ultra scores with earlier drivers. Interesting, but I couldn't begin to comment on that. Something has changed over the last few detonator driver revisions, but basically the cards are neck-and-neck (or whatever the video card equivalent of a neck is) here.
A pretty big gap in 1600x1200 HQ here, with the FX card coming in 33% - 10FPS faster. Once Half-Life 2 is released I'll probably stop using Comanche 4 as a test.
The 'rolling demo' of X2: The Threat that was recently released has an option in its main menu to be run as a benchmark. When enabled, the demo runs through 25 scenes and comes up with an average frame rate for the total test. It's a beautiful demo that really pushes the card with all kinds of lighting effects and some very nice high-poly models. I'm really looking forward to this one, almost as much as I'm still hopelessly hoping for a Freespace 3.
Due to what were essentially time constraints, I chose to only run this test in 1024x768, as it's already a fairly taxing test I figure this would serve well enough. There's a small but noticeable difference between the two here, again with the 5950 coming out on top. It's not a huge lead, though, which makes me think it's mostly due to the higher clock speeds on the 5950.
Much like earlier tests with the 5900, the 5950 is ahead in normal quality settings, but the positions are reversed in HQ modes. It seems that even with all of the extra memory bandwidth, the 5950 is simply taking a bigger performance hit here for filtering and anti-aliasing.
Though the numbers are close in normal quality for the Facing Worlds flyby, enabling HQ settings puts the 5950 way, way at the top. Higher core speed, more memory bandwidth and twice the amount of on-board RAM make a huge difference here. The Radeon 9800XT has 256MB of onboard memory and slightly higher clock speeds; look for that review soon and remember to watch for how much of a difference it can make.
Ah, Halo. X-Box owners had the pleasure of playing this game before the rest of us, but the PC release was definitely worth the wait. With the 1.02 patch some problems were fixed with the benchmarking mode, and it serves as an excellent test of performance for some games not yet released. Halo was not tested in HQ mode. If you read through the technical and performance FAQ for the game (available from the downloads section of the official website) you'll see that those settings don't really improve the looks of the game at all; rather, they just serve to slow it down.
What you see here is that even though the 5950 has twice the memory and higher clock speeds, the lead it has is under 3FPS in all three tested resolutions. Yes, this is an amazing-looking game, but don't be surprised if it brings your system to its knees when the firefights get crazy (and they do get very, very crazy).
Considering the frequency with which new video cards and GPUs are released these days, it's kind of hard to say that something is as fast as it gets. In some situations, particularly ones where available memory and bandwidth really come into play, having so much of both is great. In other cases the 5950 Ultra is either marginally faster or somewhat slower than a 9800 Pro. That makes the near $200 price difference a bit difficult to swallow; some people consider even $300 for a video card to be too much.
Which I think serves as an excellent example for this point: it just isn't worth it for most people to keep up with 6-month product cycles. If you just spent $500 on a 5900 Ultra or $400 on a 9800 Pro over the summer, you don't need a 5950. The performance increase, small as it is, just doesn't justify the cost. You could have gotten a Radeon 9700 Pro a year ago (yes, they're more than a year old) for under $400 and even the performance increase over one of those just isn't worth $500.
That being the case, if you've bought a high-end DX9 card already, you don't need another one. For the rest of you still in the dark ages of DirectX 8, though, you've got to consider the price/performance ratio: the key factor in determining which card to buy with your hard-earned money. Sure, $500 would get you a pretty big boost in power, but if $400 would get you a comparable increase, well, wouldn't it just make sense to save $100?
I'm losing track of myself. Price aside, the 5950 serves as an excellent refresh to the 5900 released this past May. The new detonator drivers have helped to improve performance (in most cases), which it seems can be safely expected from nVidia right around launch time. With the exception of that one error in GL Excess, I haven't experienced any compatibility issues, either.
The only question that remains in my mind is this: how long can a card like this last? If you look at the performance in some of the newer tests -- X2, Halo, and Jedi Academy -- you'll see that we're not hitting those 100+ framerates that we've come to expect from cards. These are still excellent scores for games with such high levels of graphical quality, and as they use the newer features of DirectX 9 you're going to see that going forward. Don't be confused. Your $400 card doesn't suck. Your games have just started looking a whole lot better.

Nvidia Geforce 840m Driver

I'd be hard pressed to suggest one $500 video card over another at this point, and I expect even after I finish with the 9800 XT review that won't have changed. In that case, if you've got $500 to spend on a card, choose what you will. But if you decide to pick up something less extravagant and have some cash left over, feel free to send it my way.
From NVIDIA:

What do you need to know about free software?