Slashdot Mirror
Nvidia and AMD Hug It Out, SLI Coming To AMD Mobos
MojoKid writes "In a rather surprising turn of events, NVIDIA has just gone on record that, starting with AMD's 990 series chipset, you'll be able to run multiple NVIDIA graphics cards in SLI on AMD-based motherboards, a feature previously only available on Intel or NVIDIA-based motherboards. Nvidia didn't go into many specifics about the license, such as how long it's good for, but did say the license covers 'upcoming motherboards featuring AMD's 990FX, 990X, and 970 chipsets.'"
You already can use an Nvidia card as dedicated physx with an AMD card, but in order to not create a bottleneck and actually experience a performance loss, you need an Nvidia card that is more or less on par with your AMD card. So if you have, say, an AMD 6970, you would need like an Nvidia 460 at the very least to get good enough performance boost for it to even be worth the extra cash instead of just going crossfire.
Since all the exclusion did was hurt nVidia in sales for people who stay loyal to AMD and refuse to go intel just for SLi. Allowing SLi on AMD boards will boost nVidias sales a bit.
Do you really think so? We've been down this road before and while it's sometimes a nice ride, it always leads to a rather anticlimactic dead-end.
(Notable examples are VLB, EISA, PCI and AGP, plus some very similar variations on each of these.)
Maybe. I've only somewhat-recently found myself occasionally wanting more than 512MB on a graphics card; perhaps I am just insufficiently hardcore (I can live with that).
That said: If 512MB is adequate for my not-so-special wants and needs, and 2GB is "just too small" for some other folks' needs, then a target of 8GB seems to be rather near-sighted.
HTX, which is mostly just Hypertransport wrapped around a familiar card-edge connector, has been around for a good while. HTX3 added a decent speed bump to the format in '08. AFAICT, nobody makes graphics cards for such a bus, and no consumer-oriented systems have ever included it. It's still there, though...
This. If there is genuinely a need for substantially bigger chunks of RAM to be available to a GPU, then I'd rather see it nearer to the GPU itself. History indicates that this will happen eventually anyway (no matter how well-intentioned the new-fangled bus might be), so it might make sense to just cut to the chase...
Kid-proof tablet..
nVidia and AMD got along great before AMD bought ATi. nVidia really helped keep them floating back when AMD couldn't make a decent motherboard chipset to save their life. nForce was all the rage for AMD heads.
Well it is in the best interests of both companies to play nice, particularly if Bulldozer ends up being any good (either in terms of being high performance, or good performance for the money). In nVidia's case it would be shooting themselves in the foot to not support AMD boards if those start taking off with enthusiasts. In AMD's case their processor market has badly eroded and they don't need any barriers to wooing people back from Intel.
My hope is this also signals that Bulldozer is good. That nVidia had a look and said "Ya, this has the kind of performance that enthusiasts will want and we want to be on that platform."
While I'm an Intel fan myself I've no illusions that the reason they are as cheap as they are and try as hard as they do is because they've got to fight AMD. Well AMD has been badly floundering in the CPU arena. Their products are not near Intel's performance level and not really very good price/performance wise. Intel keeps forging ahead with better and better CPUs (the Sandy Bridge CPUs are just excellent) and AMD keeps rehashing, and it has shown in their sales figures.
I hope Bulldozer is a great product and revitalizes AMD, which means Intel has to try even harder to compete, and so on.
Intel and AMD are the only games in town for current desktop chipsets, as all the other competitors have dropped out. VIA, gone, SiS, gone (and good riddance), Acer Labs, gone, Nvidia, gone.
It makes plenty of sense not to cut out half the companies in the market, even if that same company also competes with you.
upon the advice of my lawyer, i have no sig at this time
Having built my last two gaming rigs to utilize SLI, my opinion is that it's more trouble than it's worth.
It seems like a great idea: buy the graphics card at the sweet spot in the price / power curve, peg it for all its worth until two years later when games start to push it to its limit. Then buy a second card, which is now very affordable, throw it in SLI and bump your rig back up to a top end performer.
The reality is less perfect. Want to go dual monitor? Expect to buy a third graphics card to run that second display. Apparently this has been fixed in Vista / Windows 7, but I'm still using XP and it's a massive pain. I'm relegated to using a single monitor in Windows, which is basically fine since I only use it to game, and booting back into Linux for two-display goodness.
Rare graphics bugs that only affect SLI users are common. I recently bought The Witcher on Steam for $5, this game is a few years old and has been updated many times. However if you're running SLI, expect to be able to see ALL LIGHT SOURCES, ALL THE TIME, THROUGH EVERY SURFACE. Only affects SLI users, so apparently it's a "will not fix". The workaround doesn't work.
When Borderlands first came out, crashed regularly for about the first two months. The culprit? A bug that only affected SLI users.
Then there's the heat issue! Having two graphics cards going at full tear will heat up your case extremely quickly. Expect to shell out for an after-market cooling solution unless you want your cards to idle at 80C and easily hit 95C during operation. The lifetime of your cards will be drastically shortened.
This is my experience with SLI anyway. I'm a hardcore gamer who has always built his own rigs, and this is the last machine I will build with SLI, end of story.
You realize the limiting factor in system RAM access is the PCIe bus, right? It isn't as though that can magically be made faster. I suppose they could start doing 32x slots, that is technically allowed by the spec but that would mean more cost both for motherboards and graphics cards, with no real benefit except to people like you that want massive amounts of RAM.
In terms of increasing the bandwidth of the bus without increasing the width, well Intel is on that. PCIe 3.0 was finalized in November 2010 and both Intel and AMD are working on implementing it in next gen chipsets. It doubles per lane bandwidth over 2.0/2.1 by increasing the clock rate, and using more efficient (but much more complex) signaling. That would give 16GB/sec of bandwidth which is on par with what you see from DDR3 1333MHz system memory.
However even if you do that, it isn't really that useful, it'll still be slow. See graphics cards have WAY higher memory bandwidth requirements CPUs. That's why they use GDDR5 instead of DDR3. While GDDR5 is based on DDR3 it is much higher speed and bandwidth. With their huge memory controllers you can see cards with 200GB/sec or more of bandwidth. You just aren't going to get that out of system RAM, even if you had a bus that could transfer it (which you don't have).
Never mind that you then have to contend with the CPU which needs to use it too.
There's no magic to be had here to be able to grab system RAM and use it efficiently. Cards can already use it, it is part of the PCIe spec. Things just slow to a crawl when it gets used since there are extreme bandwidth limitations from the graphics card's perspective.
I can run PhysX fine on my machine and it has an both AMD proc and graphic card (HD 3800).
It might be that the name of the game is "Let's all gang up on Intel"... given that Intel has squeezed Nvidia put of motherboards, and AMD has integrated graphics all to herself for now, getting closer is sensible, because the dominating player has a)signaled that it wants to enter your arena, and b) has probably reached, as Microsoft has reached, a performance plateau after which further technology advances are not as valuable to consumers .
Having said that, the SLI market is, and will remain, marginal in term of number of system installations. Given the pace of advance, we may be reaching in graphics what we have reached in processors; the basic market needs are more than adequately satisfied by entry-level systems.
"If a boss demands loyalty, give him integrity. But if he demands integrity, give him loyalty." (John Boyd, 1927-1997)
Hard to say(until bulldozer drops) whether NVIDIA thinks that they are really good, or good enough. Since NVIDIA no longer has an intel chipset business, tying SLI to Intel platforms no longer serves to move more product; but to restrict the size of their potential market(since anybody who wants the, often quite aggressive, price/performance of an AMD part won't be buying more than one NVIDIA card, at most).
So long as they are confident that AMD's CPUs will be good enough not to bottleneck SLI configurations, trying to sell multiple cards to people who purchase AMD CPUs seems only reasonable. If, of course, they think that the CPUs will be even better than good enough, the approach is even more reasonable, so it doesn't tell us too much about which it is.
Parent is incorrect and it's therefore no suprise that he provided no evidence or even supportive argument for his assertions.
'physx' is a marketing term and an API currently only hardware accelerated through nVidia cards. Adding more AMD cards, as the parent suggests, doesn't do squat if what you want is 'physx' on a hardware path. Games typicall only have two paths, software or 'physx', so the load either lands on the main-CPU (you only have AMD card(s)) or on the GPU (you have nVidia card with physx enabled).
He's also incorrect about "bottlenecks" (what are those? Surely not PCIe lanes) and really, anyone who modded that informative did so without knowledge of the issues. Quite the opposite is true, even a lower-end card dedicated to this task will provide much better performance than having to go through the software path.
Of course, this would all be pretty moot if AMD could provide an API-conforming hardware path for THEIR cards, but for some reason that isn't happening.
Other posts talk about the old dedicated PhysX cards. Those are as relevant as a S3 Virge in this discussion. Forget about them.
And yes, nVidia explicitly disables physx when it detects a non-nVidia card is installed. You can use hacked drivers. That's not the point.
As I said. Asshats.
You took my practical argument and made it theoretical, but I'll play. ;)
I never had an EGA adapter. I did have CGA, and the next step was a Diamond Speedstar 24x, with all kinds of (well, one kind of) 24-bit color that would put your Tseng ET3000 (ET4000?) to shame. And, in any event, it was clearly better than CGA, EGA, VGA, or (bog-standard IBM) XGA.
The 24x was both awesome (pretty!) and lousy (mostly do to its proprietarity nature and lack of software support) at the time. I still keep it in a drawer -- it's the only color ISA video card I still have. (I believe there is also still a monochrome Hercules card kicking around in there somewhere, which I keep because its weird "high-res" mode has infrequently been well-supported by anything else.)
Anyway...porn was never better than when I was a kid with a 24-bit video card, able to view JPEGs without dithering.
But what I'd like to express to you is that it's all incremental. There was no magic leap between your EGA card and your Tseng SVGA -- you just skipped some steps.
And there was no magic leap between your 4MB card (whatever it was) and your 32MB Riva TNT2: I also made a similar progression to a TNT2.
And, yeah: Around that time, model numbers got blurry. Instead of making one chipset at one speed (TNT2), manufacturers started bin-sorting and making producing a variety of speeds from the same part (Voodoo3 2000, 3000, 3500TV, all with the same GPU).
And also around that time, drivers (between OpenGL and DirectX) became consistent, adding to the blur.
I still have a Voodoo3 3500TV, though I don't have a system that can use it. But I assure you that I would much rather play games (and pay the power bill) with my nVidia 9800GT than that old hunk of (ouch! HOT!) 3dfx metal.
Fast forward a bunch and recently, I've been playing both Rift and Portal 2. The 9800GT is showing its age, especially with Rift, and it's becoming time to look for an upgrade.
But, really, neither of these games would be worth the time of day on my laptop's ATI x300. This old Dell probably would've played the first Portal OK, but the second one...meh. And the x300 is (IIRC) listed as Rift's minimum spec, but the game loses its prettiness in a hurry when the quality settings are turned down.
But, you know: I might just install Rift on this 7-year-old x300 laptop, just to see how it works. Just so I can have the same "wow" factor I had when I first installed a Voodoo3 2000, when I play the same game on my desktop with a 3-year-old, not-so-special-at-this-pint 9800GT.
The steps seem smaller, these days, but progress marches on. You'll have absolute lifelike perfection eventually, but it'll take some doing to get there.
Kid-proof tablet..
And what exactly was that snork for? The "bang for the buck" has been firmly in the AMD side of the aisle for quite some time, as you can see here on this handy chart. Thanks to the Intel socket roulette it has gotten to the point that depending on the socket one can build an AMD quad complete for less than a dual Intel and motherboard alone, that is how much extra you are paying for all those ads (and the kickbacks to OEMs of course).
Since it came out that Intel was rigging their compiler and bribing OEMs (where the hell is the antitrust bust anyway?) I've been selling AMD exclusively and my customers couldn't be happier. The lower price of AM3 boards and CPUs means they can get better features and more options for less money and having those extra cores helps to future proof a system. I mean how can you complain when you can get an AMD triple OEM PLUS a good ASRock motherboard for under $100, or that I can deliver a fully loaded AMD quad WITH a Geforce 210 AND a wireless keyboard/mouse combo and Win 7 HP X64, all for $550 while still making a decent profit? I certainly can't and my customers couldn't be happier with the performance. Not everyone is only worried about ePeen benchmarks you know.
As for TFA this isn't really surprising and the only shock is that Nvidia didn't do this sooner. Intel in their usual douchebag manner killed the Nvidia chipset business (again WTF? Where the hell is the antitrust already?) by refusing them the right to build on anything past LGA775, thus forcing everyone on Intel boards to have a craptastic Intel IGP whether you wanted it or not, so working out a deal to have SLI on AMD seems only natural.
To me the bigger question is why the hell don't they sell a 1x GPU board designed for PhysX. There are many of us I'm sure that don't have dual x16 boards that would be happy to buy an Nvidia chip just to add PhysX but screwing us over by disabling PhysX support if it detects an AMD GPU is just stupid, especially since that pretty much kills any point of having Nvidia on AMD with nearly all AMD boards having Radeon IGPs now. If they want to sell more GPUs a 1x board dedicated to PhysX would probably sell like hotcakes for those of us on AMD platforms.
Personally until they quit screwing us with the drivers I'll keep using Nvidia for strictly HTPCs (you can get a Geforce 210 for like $20 after MIR) and for those that want to game on the cheap sticking with the HD4850 (can't beat 256 bit pipeline for $60 refurb, and makes a cheap crossfire monster) or the HD5770.
ACs don't waste your time replying, your posts are never seen by me.
It's anecdotal, to be sure... All I can tell you is that when I have tons of windows open on Win7, then switching to old ones takes a while to repaint (and it's quite noticeable). With few windows open, it's effectively instantaneous (i.e presumably within a few VSYNCs). And, no offense taken, but I absolutely do want high-performance tab/window switching in my desktop applications. If I don't have to wait for contents to be repainted, then I don't want to.
And yes, I'm quite well aware that transfers from system memory to the GPU (or any other device) over PCIe are plenty fast for desktop normal operations. That's why I suspect the framebuffer size, rather than bandwidth to/from system memory, is the limiting factor. It's the only resource that would be approaching its limit with a huge number of windows open.
Other anecdotal point, WinXP doesn't show this same behavior with similar numbers of windows. Haven't had a chance to play with compiz or OSX, so I cannot comment on those. My *nix usage is generally limited to remote connections over NX, VNC, or just a remote xterm.
"onward!" cried the copper man, little knowing brass corrupts...
In fairness to Intel, if you need pure bang, particularly if it is necessary that said bang be delivered to a single-threaded application, AMD has nothing on them.
However, as you say, AMD has plenty of more-than-fast-enough offerings that are dirt cheap, and tend to be supported by slightly cheaper motherboards. Given that many modern games tend to be GPU bound much of the time, gamers on budgets are generally pretty well served by cheaping out a bit on the CPU and going up a model or two on the GPU side. Since NVIDIA sells no CPUs(barring Tegra, which is irrelevant here) and no longer has a chipset business worthy of note, they'd be fools not to try to scoop up the "good enough AMD rig and enough cash left over for a slightly ludicrous video card" demographic.
Now, given Intel's strength, they aren't about to cancel SLI support on intel, despite being fucked over on the chipset side; but ignoring AMD doesn't make much sense.