Nvidia Discloses Details On Next-Gen Fermi GPU 175
EconolineCrush writes "The Tech Report has published the first details describing the architecture behind Nvidia's upcoming Fermi GPU. More than just a graphics processor, Fermi incorporates many enhancements targeted specifically at general-purpose computing, such as better support for double-precision math, improved internal scheduling and switching, and more robust tools for developers. Plus, you know, more cores. Some questions about the chip remain unanswered, but it's not expected to arrive until later this year or early next."
But does it... (Score:5, Interesting)
... run Linux?
Re: (Score:2)
Yes, but only if your main display is connected to a genuine Nvidia graphics card [slashdot.org].
Re: (Score:2, Insightful)
You understand that these gpu's are made by nvidia, right? So how could they run something on a machine with a non-nvidia gpu if the gpu's the article refers to are made by nvidia/I.?
What exactly were you trying to say? I'm not quite sure.
Re:But does it... (Score:4, Insightful)
Re: (Score:3, Insightful)
Actually, I can think of at least one other major computer manufacturer who makes products that nerf other manufacturers' products. I think they're located in Cupertino.
Re:But does it... (Score:5, Funny)
Re: (Score:2)
IBM isn't located in Cupertino, right? Neither is Sun. Which are you talking about?
(before some pedantic know-it-all kid starts acting up; yes, I do know which company he was talking about, as do all of us).
Re: (Score:3, Informative)
Re: (Score:2)
Thanks for that. I didn't know they were based in Cupertino.
Re:But does it... (Score:5, Interesting)
Some motherboards have more than one PCI Express slot. Some even come with GPUs built onto the motherboard. In either case, it is entirely conceivable that there may be a GPU present other than the one attached to the display. Then there's the Hydra 200 (look on Anandtech, I'm too lazy to find the link) - a chipset which evenly distributes processing power among multiple GPUs from any vendor to render a scene or do GPGPU computing.
Nvidia just released new drivers which explicitly disable PhysX acceleration in the presence of a GPU from another manufacturer. For the above stated reasons, this is evil.
Re: (Score:2)
Really? That's cool.
That sucks. Do you know if there are other hardware physics solutions on the market besides PhysX? I know of a couple of software solutions, but don't know of any other hardwar
Re: (Score:2)
PhysX is the only one that had a true hardware solution (the P1 PhysX cards, previous to nV purchase). That said, AMD has been porting Havok iirc to OpenCL for use on their hardware for fancy things like cloth mesh.
Re: (Score:3, Informative)
Re: (Score:2)
Yeah it's too bad for the two people who have both an AMD and nVidia card on the same machine AND care about PhysX...
Re: (Score:2)
But it won't have an RPU... (Score:2)
They could fit one of Philipp Slusallek's ray-trace processing units in the corner of the chip and never notice the cost in silicon.
Re: (Score:3, Interesting)
FLOPS aren't directly comparable, as the ATI chips are arranged more like the Itanium, while nVidia looks more like a Core Duo. ATI has more raw power, but uses a smaller percentage of it.
This word "detail"... (Score:3, Funny)
...I'm not sure it means what you think it means.
So... (Score:2, Interesting)
Re: (Score:3, Funny)
Nvidia just makes the cards. It isn't their fault if they're not installed, cooled or properly read bedtime stories after use.
Comment removed (Score:5, Insightful)
Re: (Score:2)
Who really wants/need bleeding edge technology anymore?
Numbers...must...go...higher... [berkeley.edu]
Re:Games before hardware (Score:5, Interesting)
They won't want to sell you this chip for a hundred bucks, they want to sell it to the HPC world for a couple thousand bucks (or more... some of NVIDIA's current Tesla products are 5 figures). The only gamers they're really interested in these days are on mobile platforms, using Tegra.
Re: (Score:2)
NVIDIA doesn't care about you anymore.
They never did.
Nor does any other company.
Companies care about your money.
Name me one company that actually cares more about you than your money.
Re: (Score:2)
Re: (Score:2)
That's right. After conquering the competitive but profitable mass-market for their products, where they can make a killing in low-margin high-volume sales, they are going after a tiny niche....
It would certainly have nothing to do with competition from Larabee and the general realisation that as the GPU becomes more general purpose Games will seek to offload more calculations onto them. For graphics it's rare to hit a case that needs double-precision (it happens in HDR), but when you move your physics code
Re: (Score:2)
I mean, look at Agilent. Agilent has twice NVIDIA's revenue and they make testing and lab equipment. As for Roadrunner itself, IBM makes a lot more money on the day-to-day HPC business than they do on the occasional hypercomputer for the US government.
Re: (Score:2)
You do realise your card is less than two years old? In the year 2000 they were releasing the Geforce 2 Ultra, I bet you can't play Doom3 in it.
Graphics are still very far from "realistic", and till then graphic cards will continue to evolve; the 8800GT may seem more than enough for now, but it won't seem that way some years from now, Unless, of course, OnLive takes over the PC gaming market.
Re:Games before hardware (Score:5, Interesting)
People have been saying that forever now. I think only the first 2 generations of 3D cards were greeted by universal enthusiasm, while everything else since had a number of "who needs that power to run game X" crowd. The truth is, yes, you can run a lot of games with old cards, but you can run them better with newer cards. So, it's just a matter of preference when it comes to the usual gaming.
AMD/ATI is at least doing something fun with all this new power. Since you can run the latest games in 5400x2000 resolutions with high frame rate, why not hook up three monitors to one Radeon 58xx card and play it like this [amd.com]? That wasn't something you could do with an older card.
Similarly, using some of the new video converter apps that make use of a GPU can cut down transcoding from many hours to one hour or less... you can convert your blu-ray movie to a portable video format much easier and quicker. Again, something you couldn't do with an old card, and something that was only somewhat useful in previous generation.
In summary, I think the *need* for more power is less pressing than it used to, but there's still more and more you can do with new cards.
Re: (Score:2)
why not hook up three monitors to one Radeon 58xx card and play it like this [amd.com]?
Because not even in the publicity shot could they get that dirty great inch gap from between the top and bottom tiers of screens. The horizontal looks ok (but you lose definition in the resin overlay between the horizontal monitors), but that joint right in the middle of where you're looking would be very similar to constantly having a piece of masking tape over the middle of your current monitor.
Why not just output to a high-def TV?
N.B. Strategy games do not scale well to high-def TVs. The resolution is
Re: (Score:2)
Plus, it brings down the cost on the low end. I love getting massive power for $100.
More enthusiast cards, please!
Re: (Score:2)
They've been saying it forever, but it's starting to come true. Let me quote you from Anandtech's HD 5850 review, benchmarks starting at page 3:
"Crysis Warhead: Warhead is still the single most demanding game in our arsenal, with cards continuing to struggle to put out a playable frame rate with everything turned up."
"Far Cry 2: Thankfully it's not nearly as punishing as Crysis, and it's possible to achieve a good framerate even with all the settings at their highest."
"Battleforge: [Has no real one-sentence
Re: (Score:2)
This generation is noticeably different in that consoles now have
Re:Games before hardware (Score:4, Funny)
Young coders are too lazy and brainwashed by MS and Sony to think anymore. You finally have the bandwidth and cpu and gpu to do something and your stuck dreaming at 640p. Hack the cards with Linux and dream big. Take computing back from the DRM, locked down junk MS and Sony code down to. You have the OS, now get some graphics freedom too.
I hope you don't contribute to the trunk, your code is 2x as long as it should be, and only half as effective!
Re:Games before hardware (Score:4, Insightful)
That's because most games are now being written for consoles and then being ported to PC, so the graphics requirements are based on what's in an X-Box 360. Unfortunately consoles are on something like a 5 year cycle. People are now buying a game console + a cheap PC for their other stuff for cheaper than the ol gaming rig. Makes sense in a way.
Re: (Score:2)
>so the graphics requirements are based on what's in an X-Box 360.
I dont think thats such a limiting factor. Lets say they develop the xbox game first, instead of the PC version. They settle on 1080i for resolution and only a certain level of quality for textures. They also tone down the physics and AI to a level it doesnt slow down the xbox cpus.
Okay, now when you port the PC version, you let the user select the resolution he likes and you up the textures to max and ungimp the physics and AI. Its not
Re: (Score:2)
Re: (Score:2)
There is a danger coming to both Intel nvidea and AMD/ATI within the next 20 years.
NEWS FLASH... When you're 2x as old as you are right now, the world will have changed.
Get off my lawn.
Re: (Score:2)
Re: (Score:2)
even going so far as to eat my own dogfood
You do realize this implies you work for ATI/AMD? If not, perhaps you should do a quick check on what dogfood means.
According to your article, you're obviously price sensitive, so why would you *ever* pay full price for a product (from your competitor)?
Clearly you need to inform yourself before attempting to inform others.
Re: (Score:2)
even going so far as to eat my own dogfood
You do realize this implies you work for ATI/AMD? If not, perhaps you should do a quick check on what dogfood means.
Umm... no, it means to use whatever you're selling and recommending. To take a car analogy (surprise) if you sell Fords for a living and drive a Toyota you're not eating your own dogfood. The implied message is that you'd go to a different store and buy a Toyota, even though you don't work for Ford. Same if you're selling clothes you'd never wear yourself, even though you're a retailer and not working for the brands you carry. So if he was recommending AMD systems to everyone but using Intel systems himself
Re: (Score:2)
Re: (Score:2)
Answers.com has this:
An expression describing the act of a company using its own products for day-to-day operations. (...) This slang was popularized during the dotcom craze when some companies did not use their own products and thus could "not even eat their own dog food".
In other words, to NOT eat your own dog food means you SELL it but you're not using it yourself. If you are using it too, you are eating your own dog food. I'm not sure when that expression changed, if ever, to mean that you're not selling it yet. I guess your use is more a result of the other, because there's so much focus on "eating your own dogfood" you end up eating crap for marketing purposes.
Re: (Score:2)
Most of my customers just rave about how "blazing fast" the new dual and quad AMDs I've been building them are, [...]
That surely puts a whole new perspective on the next generation of GPUs.
Re: (Score:2)
Re: (Score:2)
Or maybe you are just getting old/mature, and are no longer drooling over toys for toys' sake? Just saying...
Re: (Score:2)
The desire for better CPUs is somewhat plateauing, too. I can't recall seeing a game with min reqs higher than a 2ghz Athlon X2 or 3ghz Pentium D. That's despite much faster dual and quad cores becoming the norm.
For a while, I actually played Left4Dead on my old Athlon XP 2400+ and 7800GS. Got about 30fps with VSYNC on, and settings set somewhat low. But now I've got an Athlon II X2, and 8800GS (which cost $30 brand new, one year ago!)
I don't have any upgrade plans at the moment. If a game runs too slowly,
Re: (Score:2)
Real gamers still dream of more than Sony or MS allows you to have
Re: (Score:2)
GPU to network (Score:2)
I wonder when a GPU will be able to directly access a network of some sort. Right now, you would need glue code on the CPU to link multiple GPUs in different systems together. I imagine that some HPC applications would run quite well with 100 GPUs spread over 25-50 machines with a QDR InfiniBand link between them.
Re: (Score:2)
Probably not, without major architectural changes. Currently when you want something to run fast on a GPU you really want to have an algorithm where you load everything up and then just let it run.
You could potentially improve that by cutting out the CPU, PCI, etc., but then you're not really talking about a graphics card anymore and you might as well just market it as a stream processor or a competitor to Cell blades.
Another article here (Score:4, Informative)
http://www.pcper.com/article.php?aid=789 [pcper.com]
Just for a second glance.
More than just graphics (Score:5, Informative)
But this technology isn't straightforward. Someone asked why not replace your CPU with it? Well for one, GPUs didn't use to be able to do ANY floating or double-precision calculations. You couldn't even program calculations directly -- you had to figure out how to represent your problem as texel- and polygon-operations so that you could trick your GPU into doing non-GPU calculations for you. With each new card released, NVIDIA is making strides to accommodate those who want GPGPU, and for everyone I know those advances couldn't come fast enough.
Re: (Score:2)
I'm curious to know how you go about writing code for GPUs. I've been thrown into a project recently that involves programming multicore architectures, so I've been reading about StreamIt (from MIT). It looks really interesting. But they don't mention GPUs in particular (just multicores), probably because the current batch of GPUs don't have a lot of candy that CPUs have (like floats).
Re: (Score:2, Informative)
Start looking at OpenCL as soon as possible if you want to learn gpgpu, cuda is nice but opencl is portable between vendors and hardware types :)
Re: (Score:2)
GPUs (of the past) are basically just massively parallel floating point units. I think the OP meant to say that they lacked integer operations and double precision floats in the beginning.
Re: (Score:2)
Hmmm? Pretty sure that the last/current-gen ATI cards can do double precision floating point at reasonable speed, though it's technically not 100% IEEE-compliant.
Re: (Score:2)
Go and have a look at GPGPU [gpgpu.org]. There's tons of material on there about techniques, some tutorials and a busy forum.
This is the easy way to go (Score:2)
Re: (Score:2)
I work at a physics lab, and demand for these newer NVIDIA cards are exploding due to general-purpose GPU programming. With a little bit of creativity and experience, many computational problems can be parallelized, and then run on the multiple GPU cores with fantastic speedup. In our case, we got a simulation from 2s/frame to 12ms/frame. It's not trivial though, and the guy in our group who got good at it... he found himself on 7 different projects simultaneously as everyone was craving this technology. He eventually left b/c of the stress.
That sounds like a physics lab alright!
Re: (Score:2)
Embedded x86? (Score:2, Interesting)
What I'd like to see is nVidia embed a decent x86 CPU, (maybe like a P4/2.4GHz) right on the chip with their superfast graphics chips. I'd like a media PC which isn't processing apps so much as it's processing media streams, pic-in-pic, DVR, audio. Flip the script of the fat Intel CPUs with "integrated" graphics, for the media apps that really need the DSP more than the ALU/CLU.
Gimme a $200 PC that can do 1080p HD while DVR another channel/download, and Intel and AMD will get a real shakeup.
Re: (Score:2)
Re: (Score:2)
Your proposal sounds similar to the IBM/Sony cell architecture: one general purpose processor core with a collection of math crunching cores. The enhanced double precision FP in this latest Nvidia chip also maps the progression of cell with the PowerXCell 8i over the original cell processor.
Re: (Score:2)
Indeed, I have two (original/80GB) PS3s exclusively for running Linux on their Cells. The development of that niche has been very disappointing, especially SPU apps. And now Sony has discontinued "OtherOS" on new PS3s (they claim they won't close it on legacy models with new firmware, but who knows).
I'd love to see a $200 Linux PC with a Cell, even if it had "only" 2-6 (or 7, like the PS3) SPUs, and no hypervisor, but maybe a separate GPU (or at least a framebuffer/RAMDAC, but why not a $40 nVidia GPU?). Th
Re: (Score:2, Informative)
What I'd like to see is nVidia embed a decent x86 CPU,
They did, its called Tegra. Except its not using the x86 hog, but way more efficent ARM architecture
Re: (Score:3, Interesting)
That's better than nothing. But I want all the x86 packages, especially the Windows AV codecs. That requires an x86.
Though that requirement suggests an architecture of ARM CPU for OS/apps, little x86 coprocessor for codecs, and MPP GPU cores doing the DSP/rendering. If Linux could handle that kind of "heterogenous multicore" chip, it would really kill Windows 7. Especially on "embedded" media appliances.
Not out until Q1 2010 (Score:2)
Emacs of the graphics cards (Score:2, Funny)
Hi thar. We gave you some useful hardware to support general purpose calculations in your graphics accelerator so you can compute while you compute.
The only thing we can't support is decent graphics in games without resorting to special, NVIDIA-specific patches.
Re: (Score:2)
Re: (Score:2)
Completely OT, but your sig just made my day.
How did this get modded "insightful"? (Score:4, Interesting)
"Ignorant" would be a better rating - there's a lot of compute power but it's in the middle of a very different architecture to an x86 CPU. Not usable for running an OS.
Re: (Score:2)
Re: (Score:2)
but it's in the middle of a very different architecture to an x86 CPU
OMG! It's not x86? It's USELESS!!!!!....
Re: (Score:2, Insightful)
Re: (Score:2)
Why bother buying a computer motherboard, cpu and case? Maybe the case to store it in, but you could make a full fledged computer with just a graphics card they are so powerful.
GPU's vs CPU's is a bit like having 5000 highly trained monkeys vs 5 highly trained people. If your task is easy enough for the GPU, it'll do it blazingly fast. On the other hand, for some tasks the CPU is still the better option.
Re: (Score:2)
It's more like comparing a single William Shakespeare with 5000 monkeys that each memorized a small part of Hamlet; The monkeys will be able to write Hamlet in a few seconds, but only Shakespeare is able to write anything other than Hamlet.
Anybody else got any contrived analogies? Something with cars perhaps?
Re: (Score:2)
Bus vs racecar? You'll transport 50 people faster in a bus than the racecar, but the bus isn't winning any speed races and doesn't corner as quickly.
Re: (Score:3, Funny)
That is an astoundingly bad analogy.
What about it's like having a regiment of 5000 soldiers vs 5 ninjas. If the task can be accomplished by rote then the regiment will win on sheer manpower, but it requires adaptability then the ninjas will triumph.
Substitute pirates for ninjas for an instant paradox.
Re: (Score:2)
Why bother buying a computer motherboard, cpu and case? Maybe the case to store it in, but you could make a full fledged computer with just a graphics card they are so powerful.
Why bother with GPU and just go with a CPU they're so powerful?
Re:Honestly, at this point... (Score:4, Funny)
Yes.
Re: (Score:2)
with a UID > 1.6m you clearly know how the mods function.
Yeah, but he's been lurking since day 1. or was it day 2. It's so hard to remember when ... HEY YOU, get of my damn lawn!
Re: (Score:2)
Accept certain inalienable truths, redundancy will rise, mods will
philander, you too will get old, and when you do you'll fantasize
that when you were young redundancy was reasonable, mods were
noble and children respected their elders.
And trust me on the sunscreen.
Re: (Score:2)
Re:AWESOME (Score:5, Insightful)
Re: (Score:3, Insightful)
Re: (Score:2)
Re: (Score:2)
Re:AWESOME (Score:5, Informative)
So no, I do mean the GT200. The GT200 processor supports double-precision, the G8x and G9x processors do not.
Re: (Score:2)
better support
Not only is it "better support", but it now has "Awesome Street Cred" and "Totally amazing user experience", thanks to a magical technology licensed from ________ (Pick your favorite scape-goat) famous for lock-in and proprietary design!
Can also be useful in graphics (Score:3, Insightful)
It depends on what you are doing, but when you get something that involves a lot of successive operations, even 32-bit FP can end up not being enough precision. You get truncation errors and those add up to visible artifacts. This could also become more true as displays start to take higher precision input and even more true if we start getting high dynamic range displays (like something that can do ultra-bright when asked) that themselves take floating point data.
Re: (Score:2)
Re: (Score:2)
http://helmer2.sfe.se/ [helmer2.sfe.se]
Re: (Score:2)
Re: (Score:3, Insightful)
Re: (Score:2)
What kind of performance were you expecting from multi-cores? I'm not too familiar with your application so I was wondering if the code could hit 1 FLOP/cycle for 3-4 GFLOP/core or even 2 FLOP/cycle for 6-8 GFLOP/core? If the programming model for DP on these boards supports what you need then you might be looking at 100x the performance on GPGPU. But then again, if lack of support for saturating error conditions was holding you back before when there was a potential 10x increase performance, is 100x enough
Re: (Score:2)
Re: (Score:2)
I mixed up per-core and per-chip performance in original post, but assuming that the code can be scheduled to despatch 2 DP ops per cycle per core (reasonable if the code has a simple enough structure to run on a GPU), then you hit about 8 GFLOP/core. So yes, the new six-core Nehalems can hit about 50 GLOP/s. Some benchmarks put it slightly higher at about 55. Nvidia are claiming an 8x increase in DP throughput which puts them at about 616 GLOP/s for DP. It's still a ten-fold increase in performance.
When yo
Re: (Score:2)
When you say that Intel can double cores and vector word length in their next family are you confusing Westmere (which is a tick generation) with it's successor (which will be a tock)?
I am talking about Sandy Bridge [wikipedia.org] since that is the one with AVX. They have demoed systems with it already at IDF. It is supposed to do 256 DP GFLOPS. Intel is also supposedly going to implement Fused Multiply Add sometime after that which will double performance again.
Re: (Score:2)
AC posters suck
Fixed that for ya.
I'd have posted AC, but the refresh time is too long.