Slashdot Mirror
NVIDIA GTX 295 Brings the Pain and Performance
Vigile writes "Dual-GPU graphics cards are all the rage and it was a pair of RV770 cores that AMD had to use in order to best the likes of NVIDIA's GeForce GTX 280. This time NVIDIA has the goal of taking back the performance crown and the GeForce GTX 295 essentially takes two of the GT200 GPUs used on the GTX 280, shrinks them from 65nm to 55nm, and puts them on a single card. The results are pretty impressive and the GTX 295 dominates in the benchmarks with a price tag of $499."
Just in time for holiday shopping!
When I read about the evils of drinking, I gave up... reading.-Henny Youngman
this is like the razor wars (double blade! triple blade! quad blade! pento blade!). With OpenCL and DirectGPU (or whatever MS is calling it this week), this should be good for anyone trying to build a cheap superGPU cluster.
Do you even lift?
These aren't the 'roids you're looking for.
I might consider upgrading from my 2MB VGA after seeing it in action... :)
Have you tried GTA4? I get a nice 3fps with a Crysis killing machine. I think I'll need two of these in SLI.
Belief? Hope? Preference?The Existential Vortex
how do you people buying this stuff get over the noise of the fans? and how often do you have to exchange the fans? i buy only hardware with speed-to-noise ratio near infinity.
A helathy does of competition in this market is going to drive the efficiency and performance up and at the same time keep the prices down to a reasonable degree. Now if only AMD would man up and make a core i7 killer then the pain will have been brought.
2 years later, and we will be able to saddle a graphics card and fly with it in skies.
Read radical news here
I'm glad that people are out there buying graphics cards that can render the latest games in QuadHD resolution at 120 frames per second... it makes the integrated graphics in eee class PCs that much better when the tech trickles down 5 years later.
As a somewhat mystified recent purchaser of a GTK 260 from eVGA, I was amazed to discover that NVIDIA has such problems with their linux drivers. I owned one of their older cards before and built a new computer and thought it was a no-brainer to pick NVIDIA for linux (freedom issues are notwithstanding, but I decided to go with the pragmatic choice). Only after I ran afoul of the powermizer slow switching crap, or other weird issues such as the misreporting of the screen refresh frequency, did I start digging and realized how many problems there are. As it is, I've got the beta 180.16 driver installed and it's better but I still had to do some tricks to shut off the powermizer feature. Just this morning had some other weird problem with screen corruption that's never happened before with my old hardware but more or less the same software on top of it.
For me personally, I could care less if the card hardware is great if the drivers suck. NVIDIA, fix your linux drivers please. Next time I'll give a much harder look at amd.
Gentlemen! You can't fight in here, this is the war room!
It's not just about the graphics. GPUs are being called upon to do much more, from AI to Physics, to folding@home. Even encoding and decoding audio and video formats.
Shai Schticks:"You don't make peace with friends, you make peace with enemies"
Comment removed based on user account deletion
Comment removed based on user account deletion
GTA4 kills proc and not GPU so much - I have a GTX260 Core 216 and a Quad-core Q9550, and I get an easy 40 FPS with medium textures enabled. I hear it's supposed to go higher when Nvidia releases their new drivers. In any case, Steam Forums was having a shootout in many threads when GTA4 was released and there's a pretty clear correlation between processor power and higher framerates. It's also supposed to be 64-bit optimized so I might be getting a boost from running 64-bit Vista.
If I have not seen as far as others, it was because giants were standing on my shoulders. --Hal Abelson
GTA4 is processor dependent, not GPU dependent, cos it's the crappiest console port we've seen in years.
NVIDIA, fix your linux drivers please.
NVIDIA, open your linux drivers please.
factor 966971: 966971
I wonder if this card will suffer from microstutter. The 9800GX2 benchmarked very well but real world performance was lacking because the card essentially varied between very high fps and very low fps, so it still lagged even though it got decent average fps.
With these dual cards it's best to look at their low fps rating. An average fps is often misleading.
Ten years ago the video card wars were in full swing. Each generation brought amazing new advance and impressive technology.
But nVidia and ATI haven't realized that we passed the point of diminishing returns years ago. Mobility and battery life are what matter. And I know there are hardcore PC gamers out there, but there are only a handful of companies even bothering to push the high-end graphics, so you buy a $500 video card and there are exactly ZERO games that take full advantage of it. Wait a year or so, and you may find that one or two of the few high-end PC game makers decide to throw you a bone and add support. And as a bonus, you get SIGNIFICANTLY increased power consumption, and the video card addicts are just wasting resources so they can all whack-off to Shader 30.0 soft shadows on eyelashes.
It's a weird, captive, completely pointless market unless you're doing 3D rendering for a living (for movies, for commercials, for product design, etc.).
You get 25fps in GTA4 and 40fps in Crysis
:D
Now, which one (given those numbers) would you expect to look better?
Which one actually looks (a lot) better?
Theres your problem.
Also it has a spectacular memory leak that sees it using up all available physical memory after a while, grinding to a halt and refusing to load any new textures, which is actually pretty funny the first time. Driving around in a flying car avoiding invisible buildings
A while back, AMDTi said that they were not competing at the high-end anymore: "There were also very specific admissions that AMD/ ATI isnâ(TM)t competing at the high end with Nvidia, nor do they intend to match up to the GTX 280 with a release of their own uber-chip." source. So to say "ATI had to combine two cards to be on top!" kind of completely misses the point. (emphasis added.)
For the interested, there's a great article at anandtech talking about how the R770 came to be pretty awesome... Really, though, it's not a super-high-end part.
the DRM in GTA 4 may be slowing it down (If you are using legit version, that is).
Definitely. Runs pretty shitty here, with a 3.0ghz Core 2 Duo, GTX 260, 4GB DDR2-800. I hear the quad-core people have it good.
WHO NEEDS SHIFT WHEN YOU HAVE CAPSLOCK/ DAMN1
I've been working on a game review for GTA4 (shameless plug: game-over.net). My machine meets literally the minimum system requirements and I'm running it at a steady 50fps in 1024x768. Many, many people are reporting problems with GTA4, and there seems to be no rhyme nor reason to it. Smoking fast machines can't run it, my machine can. Rockstar really pushed this game to market well before it was ready.
Jealously hoarding mod points since 2007.
Nvidia never lost the performance crown. AMD did not even bothered to compete with Nvidia for performance at the high end.
Read this excellent article.
What AMD did with the RV770 series was to totally pwn everything below the super high end.
When the 4870 was released at $299, it was generally worse than GTX280, but it easily beat the GTX260 which was priced at $399.
When the 4850 was released at $199, it easily matched the 9800GTX which was priced at $249
Unfortunately until its released all these "Special previews" are just mouthpieces for nvidia, notice all the special previews use those 4 or 5 games nvidia has been telling them to use.
Of course they come out on top when you control the tests. Im not a fanboi either way, it just annoys me when people make decisions based soley upon reviews that are clearly biased.
not quite. physics calculations are very well suited to vector processing, which is why most Physics Processing Units are vector processors, just like GPUs, array processors, DSPs, stream processors, etc.
a GPU may be more specialized to handle 2D/3D graphics rendering, but they can also perform physics calculations quite well because of their SIMD architecture. GPGPU stream processors in particular are of great interest to the scientific community because of their superb performance in scientific modeling applications which require crunching large sets of numbers.
of course, there are some vector processors that are better suited to physics calculations while others may have architectures & instruction sets specialized towards graphics or digital signal processing/video encoding, but sometimes the distinction is purely marketing-based. so if you had to choose between using a general-purpose scalar CPU or a vector GPU on a graphics card to perform physics calculations, it would be far wiser to use the GPU.
It's been recommended on several OC forums to disabled clip recording in-game--apparently this is offers a substantial performance increase... YMMV, but it's free to try. Unlike this new NVIDIA beast.
The real question is: Did Nvidia simply make enough of these (~30 - 200) certainly very carefully hand-picked chips to homologate them as a valid offering in claiming the top spot on the performance charts...
...or are they actually producing them in quantity such that anyone who wants one can buy one at the stated price?
Personally I'm betting on the former being far more likely than the latter.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
it makes the integrated graphics in eee class PCs that much better when the tech trickles down 5 years later.
This is also good news right now for the "sweet spot" gaming PC builders. Each time these new bleeding-edge-$500-200W-XXTREME cards come out, the previous two generations of graphics cards tend to suddenly drop drastically in price.
When I built my current middle-of-the-road gaming computer, I put in an ATI HD3850 for $150, with the expectation of adding a second on Crossfire once the price drop occurred. Looking at Newegg.com, the 3850's have hit ~$55.00 this week. My computer looks to be due for an upgrade after Christmas, it seems!
>> Standing on head makes smile of frown, but rest of face also upside down.
Please cite that.
I'm running L4D on my (very) old computer, a 1.6 Ghz AMD single core with a 7600 GS and 1.5GB ram. The game runs fine at medium settings (despite the fact that I am way, way under the minimum system requirements), and when I briefly swapped out the 7600 for a 7900, I was able to turn a few of the settings from medium to high (1024x768, textures low, medium effects -> 1280x1024, textures medium, high effects) and still get a stable 20-25 average frame rate.
Not quality benchmarks, I know, but the engine hasn't changed drastically since HL2 except for graphical improvements (=GPU limited), so claims about it being CPU limited haven't been true since the first public version of the Source engine, and that's assuming they were even true back in 2004.
Obligatory Soundbite Catchphrase
Just look at one of the hundreds of graphs available that directly tell you which one is better. You have to do this perhaps once every two or three years. Sure, the naming scheme is stupid but not buying them because of it is ridiculous.
And really, it's not that common. AMD CPU: higher number is faster. AMD GPU: Higher number is faster. Intel CPU: Higher number is faster. It's just Nvidia with their anal marketing.
You guys have got it all wrong... ATI cards are all the rage... Or, I guess they were until the Radeon...
Bow-ties are cool.
I have a GTX 280, and I think it's suspicious that it's absent from their benchmarks, to say the least. Why would you include GTX 260 and an ATI card, but not GTX 280?
Though TF2, HL2, and the like may be CPU limited, the bottleneck is so low (think: early P4) that any decent gaming rig from the past few years can produce much better graphics quality and framerates on these Source titles than they can with other games out at the same time, such as Oblivion.
>> Standing on head makes smile of frown, but rest of face also upside down.
People need to understand that sometimes, those detail sliders aren't meant to be cranked all the way up on current hardware. They're there to "future-proof" the game, so that it can still look pretty 2 or 5 years down the road. Wing Commander 4 did it for example.
Of course, it's not a huge incentive for developers to future-proof a game when all they get for it is a forum-bashing like "omg the game sux i can't get 50 fps on my 1337 rig".
Don't whistle while you're pissing.
I don't know very many people who play the latest games in 1024x768 anymore. Heck, I was playing Q3 that high in late '99. Today it's all about 1680x1050, 1920x1200... maybe a bunch of peeps on 1280x1024 laptops or CRTs.
Maybe I should try running Crysis in 320x240 for shits and giggles.
-Billco, Fnarg.com
They do, ever since NVidia implemented the PhysX API in their GPU drivers (for GeForce 8 and up).
On my PC, the PhysX control panel lets me choose between the GeForce or an AGEIA card. They rolled this out a few months ago, works on XP and Vista as far as I know.
Screenshot
-Billco, Fnarg.com
The DRM isn't the problem, I think. It's just that they rushed it out for the Christmas shopping period knowing people would buy it. There isn't a demo so you can't even try it first to see if you're in the apparent majority that has severe problems with the game.
Actually, I don't think Rockstar 'get' the PC platform anymore. They actively prevented people from changing the game files so mods are impossible without hacks. They put mechanisms in place to prevent trainers from working so you can't have fun with silly cheats. The game requires you to get a Games For Windows Live account, and a Rockstar Social Club account. The atrociously poor performance, poor performance scaling and awful configuration options are just the icing on the cake.
Unlikely, if they're focused on results and not marketing. From what I've seen, adding more GPUs results in rapidly diminishing returns in the majority of cases - have a look at some benchmarks of Nvidia's tri-SLI for example.
According to the Steam Hardware Survey last updated October this year, about a quarter of it's user base have a primary resolution of 1024 x 768. Surely there are some that set certain games to higher resolutions, but I would expect the percentage of those people in the 1024 x 768 bracket to be similar to the percentages in the other brackets. Except those at 1920 x 1200, naturally.
More Twoson than Cupertino
I call BS I am beyond minimum requirements, I don't get near that on 800x600.
What are your specs?
Belief? Hope? Preference?The Existential Vortex
I am beyond minimum requirements.
...but does your system meet the "Recommended" requirements?
Don't tell me to get a life. I'm a gamer; I have LOTS of lives!
The same rule holds true for nVidia cards. They shifted direction in their numbering scheme after they got to the 9000 series, it's true, but ATI did the same thing when they need a successor to their 9xxx cards.
"16MB (fuck off, MiB fascists)" - The Mighty Buzzard
A fight to the death? No! A fight to the pain!
Misread that for a second as "A fight to the Palin"...
But then the image of Michael Palin slapping Sarah Palin made it all worthwhile...
Bow-ties are cool.
I don't see why this is modded flamebait. The GeForce 8 and some of the GeForce 9 series have this exact issue of failing because of faulty die packaging, it's been around for almost a year, and it still hasn't been fixed.
I hope that these cards don't die. Mine died within three weeks and forced me to send my laptop in for repair.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
Low end graphics have not advanced much at all.
After all, isn't that one of the big things regarding the Vista Ready debacle?
When on board graphics cannot even display desktop eye candy I think it needs to be admitted that there is no trickle down for low end or on board graphic cards.
The low end and on board graphics cards are just as weak now as they were 5 years ago, I'm not positive, but I think they might even be worse than the old intel 810's.
Don't know something? Look it up. Still don't know? Then ask.
HL2 based games only use a single core on my system and frequently don't even max that out. Hardly massive demand. TF2 runs well even on my 4 year old notebook.
"The Adobe Updater must update itself before it can check for updates. Would you like to update the Adobe Updater now?"
The 32-bit Linux kernel cannot efficiently handle more than 1 GB RAM. Linus Torvalds himself is outspoken on this subject and recommends using a 64-bit kernel with more than 1 GB (alas, I can't find the reference right now, but he has written many times on the subject in RealWorledTech.com's forums).
Here, I was real curious about this issue, so I looked it up:
http://linux-mm.org/HighMemory
Basically, when a system has more than 1GB of memory, the kernel can't directly map all physical RAM into its own address space (because 3GB of the virtual address space is reserved for applications...) - which makes certain operations more complex than they need to be. The kernel needs to take some extra steps when dealing with more than 1GB of physical RAM when exchanging memory addresses between different modules, for instance... The problem gets worse as you go beyond 4GB RAM and have to deal with PAE, etc...
You can get around that by changing the amount of virtual address space reserved for the kernel vs. user space - for instance a 2GB kernel/2GB user split of virtual address space on a system with 2GB RAM, or 3GB kernel/1GB user split on a system with 3GB RAM - but of course that's not ideal as it limits how much virtual memory (RAM + swap, etc.) is available to user processes...
On the other hand, using 64-bit in userspace is pointless for most users. Most applications don't need more than 3GB of RAM. Also, there are binary plugins, etc, which only work in 32-bit.
I don't know that I agree with that. Among other things, in x86-64 there are twice as many general-purpose registers. This makes it a lot easier for GCC to optimize compiled code effectively. Additionally the other features present on modern processes can be safely taken as assumptions by the compiler and standard libraries. The binary-only problem isn't as bad as it used to be, either - video drivers are available in 64-bit, flash is available in 64-bit, Java is available in 64-bit... I just recently built a new system and I decided to go fully 64-bit - I think it's worth it, and so far there's been only minimal difficulty. I guess the main disadvantage of a 64-bit userspace is the increased size of pointers and std::size_t, etc. using more RAM. To me the tradeoff would be worth it.
64-bit kernel and 32-bit userspace is the most efficient and practical choice for the vast majority of users.
Bow-ties are cool.
There is really two things going on here. They are trying to build manycore GPUs and trying to produce commercially usable video cards at the same time. If they can do both successfully, it will be good for all of us, but solid hardware, and good drivers all across the board will be required. The new C1080 card has 240 cores and plugs into a PCI Express slot. Contemporary NVIDIA Video cards have less. This is the new game. A frame with four of these boards is now a supercomputer. I just wish I could get one of these into my Mac Pro (I must be dreaming).
not quite. physics calculations are very well suited to vector processing
Correction, *some* physics calculations are well suited to vector processing, but not all.
Simple particle physics is trivial to throw at a vector processor. This is pretty much limited to adding a bit of prettyness to a scene. HavokFX is a really good example, and a lot of the hardware accelerated parts of PhysX implements this stuff (eg fluids, cloth etc). Since these are mainly visual effects, it makes a hell of a lot more sense to put this stuff onto a GPU rather than a PPU.
Basic primitive collision gets a bit of a speed kick in PhysX. There are however two problems that i've seen with this though. Firstly, take a look at the ageia hanger of doom demo to see how adding lots more physical boxes to a game, doesn't translate to better gameplay.
The second problem is far more severe. Attempt any kind of feedback in your simulation (character control for example), and you'll find that disabling hardware acceleration gives you significantly better performance.
So, if particle physics is better suited to the GPU; simulating loads of boxes isn't needed for better gameplay; and the kinds of physics that people want to see in games, is better off without a PPU; One could fairly easily argue that the ageia physX card is the first and last PPU we'll ever see....
I prefer 800x600 because my CRT has much higher refresh rates at that resolution. Even if I get solidly high FPS at 1680x1050, running the game at 60-75hz seems so jerky compared to the 160hz that my eyes are used to. That being said, I'm fully aware that I'm in the minority.
The reason they don't allow modding is that whole "hot coffee" thing you might have heard about, you know the lawsuit against the company based on a third party mod that almost shut them down.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
it has absolutely nothing to do with whether something is a "visual effect" or not. the GPU does not care whether its calculations are for generating 3D graphics or a table of numbers. and physics problems definitely aren't limited to gaming.
the fact of the matter is, physics calculations generally involve applying the same operation to a large set of numbers, thus greatly benefit from data parallelism. that is why SIMD processors are perfect for physics calculations.
and the reason why Ageia PhysX chips are the only dedicated PPUs on the market is precisely because there's very little practical difference between a PPU and a GPU, which is why most PPUs are GPUs (and vice-versa). there just isn't a market for a dedicated PPU when your GPU's stream processors will do the exact same thing. the Cell processor's SPU stream processors are used for both graphics and physics calculations. likewise, the NVidia GPUs in next-gen graphics cards will likely all have PhysX support (even current gen graphics cards without the PhysX chip will soon have PhysX support through firmware upgrades).
and i really don't see what gameplay has to do with anything. whether a game is well designed has nothing to do with whether or not consumers have dedicated PPUs or not.
Your eyes can tell the difference between 160hz and 100hz refresh rates? I thought the human eye topped out around 60hz-75hz. Your super human.
this statement....it was a pair of RV770 cores that AMD had to use in order to best the likes of NVIDIA's GeForce GTX 280....is false. im sorry to break it to yous but one 4870X2 mops the floor of even two GTX280's in sli. and dont try to say im just a whiny fan boy. this is not the case as i am running a 9600GT. i buy whats worth the money for the performance im getting. With that being said, ATI has been raping the market when it comes to price for performance for the last 2-3 years non stop. NVIDIA needs to get into the game and stop selling their inferior cards at such rediculous prices. If they make something that owns ATI's cards, well then they should at least make the price feasible. not dropping a 600+ price tag on it. thats not the way to sell a video card when u can get a 4870X2 for under 500 bucks. Come on NVIDIA, we're waiting for a real performance/bargain.
http://www.google.com/search?q=tf2+cpu+limited&ie=utf-8&oe=utf-8&aq=t
Basically if you had increased your processor from 1.6 to say 2.0ghz you'd notice a big difference. Performance scales with video card, but also significantly with CPU.
Most other games are limited by GPU much more and don't show the same scaling with CPU.
I KUT J00 M4NG!!!
Just because it is playable doesn't mean that it doesn't scale with CPU power. Try running their games on a P4, then swap out only the processor and watch as your framerates go up significantly. Most other games are so GPU limited that they do not show this same scaling framerates in response to CPU change.
Seriously google it. Or just try it yourself.
I KUT J00 M4NG!!!
My super human indeed...
I actually didn't state that I can tell the difference between 160hz and 100hz, although I generally can. I always hear people say that human eyes can only see 30-60 frames per second, but they forget that there is a lot more going on for a computer than just the screen refresh rate. If the game is in a hectic area that requires more computation, FPS will be lower in the game, and if your FPS drops below 60 while your refresh rate is at 60, it will cause some frames to be shown more than once per refresh. The higher the refresh rate, the less likely you are to notice such events.
Also, if your refresh rate is higher and you aren't using "wait for vertical sync", the your mouse inputs will make a difference on the screen quicker. I also happen to run my mouse at 1000hz...
Once again, I'm fully aware that I'm in the minority.