Slashdot Mirror
SLI On Life Support For the AMD Platform
JoshMST writes "For years AMD and Nvidia were like peas and carrots, and their SNAP partnership proved to be quite successful for both companies. Things changed dramatically when AMD bought up ATI, and now it seems like Nvidia is pulling the plug on SLI support for the AMD platform. While the chipset division at AMD may be a bitter rival to Nvidia, the CPU guys there have had a long and prosperous relationship with the Green Machine. While declining chipset margins on the AMD side was attributed to AMD's lackluster processor offerings for the past several years, the Phenom II chips have reawakened interest in the platform and they have found a place in enthusiasts' hearts again. Unfortunately for Nvidia, they are seemingly missing out on a significant revenue stream by not offering new chipsets to go with these processors. They have also curtailed SLI adoption on the AMD platform as well, which couldn't be happening at a worse time."
This is pure conjecture, but to me it seemed as if when AMD and ATI became one team and Nvidia and Intel became the other, that it would make sense for each one to offer incentives (read: threats) so that their partner would not bend over for the competition. So its not like its completely up to Nvidia to start improving their standing with AMD because of pressure from Intel. If that made any sense, then I'll drink a couple more beers before posting next time. Out
"It's ok, I'm completely secure as long as my iron is off"
Beginning of the end?
Why on earth if you're NVIDIA do you make it harder to find mainboards to leverage your tech? I'd have expected this move by AMD first, you'd think NVIDIA would be wanting to have their tech available everyplace possible.
Looks like no more NVIDIA for me, time to research what ATI has available. I like my AMD chips.
It is very important to always drink more beers before posting here. Otherwise, there is no chance of a +5 Insightful mod.
NVIDIA tries to jinx AMD, but ends up jinxing themselves. This has been tried throughout the ages and often ends up at the same result./>
Move on, nothing to see here.
Hydraulic pizza oven!! Guided missile! Herring sandwich! Styrofoam! Jayne Mansfield! Aluminum siding! Borax!
Dual GPU solutions are so pointless, a waste of money for little performance gain, that doesn't even work in some games.
The fact is that a very marginally small portion of people actually use more than one video card. And why should anyone really, when modern day consoles cost about the same amount as one would spend on a moderately high end processor + video card, why the hell would most people want to spend an extra 300 bucks or so to have an extra video card at only 25% or less extra benefit in framerate? Only the hardcore ones with the extra wallet is who. As for me, I'm more than happy with my $1000 system with ONE video card, and I know its going to last me at least and extra year or two anyway.
Anyway all I'm saying is AMD has the ability to tie in their own processor + GPU combo, plus let the consumer buy a separate GPU, thus getting their own "SLI". If they play their card right, they can just give the finger to NVIDIA and provide some real competition that this market really needs to prevent us all from paying $200-300 for a decent GPU these days.
Asus has jumped in bed with Microsoft as of late. With AMD's purchase of ATI and promise of open source drivers and Nvidia's failure to move forward in open source, Nvidia and Asus has seen the last dollar of mine.
Slashdot could have linked to the article the story submitter wrote for PC Perspectives: SLI on Life Support on the AMD Platform: Oh SNAP!.
Until, that is, millions of their mobile GPU chips keel over from heat death due to improper package bump and underfill construction.
And their single GPU chips are so big that they're impossible to manufacture cost effectively.
And that they need expensive PCB's because 512-bit wide memory is necessary when DDR3 has go up against ATI's more advanced DDR5 boards with half the required memory bus width for near equivalent memory performance.
And when two small, cheap, easy to manufacture chips beat out the biggest chip every time.
And when you're trying to get DirectX 11 running for the first time while making a radical architecture shift all while going to a new chip making process against a rival who is already shipping 40nm chips and has essentially had DX11 running in their past three generations of chips.
Yeah, I'm not sure Nvidia has nearly all the goods right at this moment.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
Here's a Forbes Magazine interview with the CEO of Nvidia: Nvidia's Plan For Beating Moore's Law: Chief Jen-Hsun Huang on how GPUs could get ahead of CPUs. But read the comments. Readers are not impressed.
There is a general impression now, apparently correct, that Nvidia is not honest and cannot be trusted. HP bought Nvidia graphics chips, and when they were found defective, neither company was completely honest about fixing the defects, articles say.
An Inquirer article, Nvidia cuts out reviewers for the GTS250, says "IT IS ALWAYS funny when an unethical company turns on its own supporters as Nvidia did with the latest 'all new' GT250 cards. This time however, their PR stunts cross the line from unethical to purposely false, and hilarity ensues."
Another quote from the Inquirer story: "This time however, they crossed the line from plausible deniability to flat out deception. In the middle of last week we heard what Nvidia was up to this time around, but just couldn't believe they would be THAT sleazy."
Now that Intel is integrating faster GPUs into its chipsets, there is a perception that eventually there will be little room for Nvidia.
Larrabee doesn't change a damned thing. A beowulf cluster of shitty Intel GPUs doesn't magically remove the stench of failure. It's just a whole lotta more suckage on one die.
-Billco, Fnarg.com
Really the article makes it sound like Nvidia is abandoning AMD chipsets but it's just SLI support. When they started making this decision it looked like AMD was totally dead in the enthusiast market. Even die-hards were switching to Intel chips. It seemed for a while there that the market for dual graphics cards on AMD was nearly dead. Now that AMD has a good chip again Nvidia will probably be scrambling to get a new chipset out for enthusiasts.
As I understand it, you don't really double your performance by putting two cards in. How many people seriously drop the coin to do this? Everything I've read says you'll get better bang for the buck by buying one good card, saving the money you would have spent on the second and then buying an equivalent card in three year's time that will kick the arse of the first card.
Kwisatz Haderach
Sell the spice to CHOAM
This Mahdi took Shaddam's Throne
There are single-slot dual GPU offerings from both camps. If you actually need/want SLI/CrossFire, what's the point of running 2 cards when you can have 1?
AMD = Value.
SLI = Not Value.
AMD has consistently shown that they want to put a computer at every set of hands on the planet. Geode, PIC, OLPC. Now it would be nice if those computers had fast 3D graphics or GPU parralel processing, but that really seems like an easy way to waste the real power of computers.
I have loved many Nvidia products in the past, but stepping away from AMD seems like a poor choice on Nvidia's part.
Impressively, ATI's drivers still suck. You'd think they'd've learned by now. I don't game. I just want a damn graphics controller - a slow and steady, efficient and cool graphics controller - that has drivers that work properly. Easy to get with nVidia. Not so easy to get with ATI. And if you happen to get stuck with a laptop with an ATI graphics chip? Well, all I can say is GOOD LUCK. Bad enough that if you don't have .NET Framework installed when you install the driver, you end up with 5 more hours of diagnosing and repairing a broken installation of ATI CCC that it was ignorant enough to install anyway. Add in the fact that ATI doesn't even support anything older than ~1 year old (they don't support my laptop with a Radeon X1200 even though it's still under MFG warranty). Now that's a company I want to buy products from.
I hope ATI goes bankrupt from their ignorance.
Yeah, it's 32 cores of x86 overhead.
Why not use a small RISC core alongside the new 512-bit vector unit? No more x86 decoder overhead (non-trivial on a Pentium-level core replicated 32 times), remove the cruft, tighten up the ISA, etc.
Right now it looks like 2x the die area to achieve the same in 2010 as NVIDIA achieved in 2008, and rumoured power consumption figures that make a GT200 look lean and athletic.
However it is a major improvement for Intel, and Larrabee 2 or Larrabee 3 will get it right. Also there are lots of Intel fans who will buy it regardless. My major worry for Intel would be the drivers, these are already rumoured to be why Larrabee is 2010 instead of late 2009 now.
Own NVIDIA shares do you? Or just a fan boy?
Sorry, but the NVIDIA solder problems have been well documented, so you must have chosen to ignore this.
NVIDIA is going to use GDDR5 on their upcoming 40nm "midrange" parts. ATI actually did a lot of work on GDDR5 with the memory manufacturers, hence they got the technology early.
Quite simply with die sizes, the bigger the die, the more costly it is. AMD/ATI and NVIDIA both use TSMC for their products, so it comes down to die size. The bigger the die, the bigger the chance of a flaw affecting it, so yield drops, never mind getting fewer dies per wafer.
NVIDIA have to use a wider memory bus to compete on bandwidth using DDR3, that costs money, it requires a PCB with more layers, it increases complexity. There's a reason that AMD is selling decent graphics cards for $99 - $199, and NVIDIA have been forced to rebrand their old generation parts TWICE in the past year or two.
And ATI's drivers have been fairly good since they went with the monthly updates (it's not 2001 anymore), and even better since they've been part of AMD. There are issues still - hardware video transcoding is unusable, and thus the $30 NVIDIA hardware transcoder is a better option, for those who use it.
A key part of DirectX 11 is hardware tesselation, and ATI have had that since the 360's R600 chip. DirectX 10.1 includes some non-trivial performance enhancements. NVIDIA has been holding back the market by not being able to support 10.1. ATI have DX11 hardware already, they showed it off at Computex a couple of weeks ago.
NVIDIA aren't the same as the good old days when they had awesome chipsets like NForce2, the best graphics cards, etc. However they do push onwards with GPU computer (CUDA) and PhysX, but OpenCL will overtake the former, and the latter won't catch on unless it is cross-GPU.