• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

NVIDIA Updates GPU Roadmap; Announces Pascal

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
During a keynote speech at our annual GPU Technology Conference in San Jose, Calif., NVIDIA CEO Jen-Hsun Huang updated our public GPU roadmap with the announcement of Pascal, the GPU family that will follow this year’s Maxwell GPUs. Named for 17th century French mathematician Blaise Pascal, our next-generation family of GPUs will include three key new features: stacked DRAM, unified memory and NVLink.

  • 3D Memory: Stacks DRAM chips into dense modules with wide interfaces, and brings them inside the same package as the GPU. This lets GPUs get data from memory more quickly – boosting throughput and efficiency – allowing us to build more compact GPUs that put more power into smaller devices. The result: several times greater bandwidth, more than twice the memory capacity and quadrupled energy efficiency.
  • Unified Memory: This will make building applications that take advantage of what both GPUs and CPUs can do quicker and easier by allowing the CPU to access the GPU’s memory, and the GPU to access the CPU’s memory, so developers don’t have to allocate resources between the two.
  • NVLink: Today’s computers are constrained by the speed at which data can move between the CPU and GPU. NVLink puts a fatter pipe between the CPU and GPU, allowing data to flow at more than 80GB per second, compared to the 16GB per second available now.
  • Pascal Module: NVIDIA has designed a module to house Pascal GPUs with NVLink. At one-third the size of the standard boards used today, they’ll put the power of GPUs into more compact form factors than ever before.
  • Pascal is due in 2016.
 
Hopefully the upcoming generation will be useless to bitcoin(or whatever replaces it) miners so they can't ruin the party for everyone else live they've done with this generation.
 
The "Pascal Module" looks interesting... Do the board manufacturers implement their own version this or is this the equivalent of the new reference board? Also, the cards will still need large heatsinks/coolers, right?
 
Wow, why did I just buy a socket 1150 motherboard? It looks like the GeForce 800 series and Haswell-E will rule the world.
 
I'd bet money that none of the current hardware will work with pascal boards. That interlink is going to be a physical connection to cpu, and no current hardware has that connection. The server boards have something similar but no where as fast or efficient as nvidia is claiming the link will be.

The interesting twist is that if Nvidia is publicly announcing it, they have an agreement with someone making a CPU that will work with this design. So the question is who? Snapdragons are almost fast enough 800 series and later to be considered, Intel might have done it to get help with their own board cards, and some how I can't see AMD and Nvidia working together at this point, but AMD does have cross IP licenses with Intel so who knows.

Oh and Pascal was the first language I learned in High School.
 
Last edited:
drakken said:
I'd bet money that none of the current hardware will work with pascal boards. That interlink is going to be a physical connection to cpu, and no current hardware has that connection.
Click to expand...
Well, here's the thing, they never clearly defined which CPU they were talking about.

We know they're planning on putting ARM CPU's on these newer graphics cards, they could be talking about THAT rather than the main system CPU. A bus that wide between the on-card CPU and on-card GPU is pretty easy to imagine.
 
Unknown-One said:
Well, here's the thing, they never clearly defined which CPU they were talking about.

We know they're planning on putting ARM CPU's on these newer graphics cards, they could be talking about THAT rather than the main system CPU. A bus that wide between the on-card CPU and on-card GPU is pretty easy to imagine.
Click to expand...


They could indeed, and if that was the case it would probably make more sense. Does the current GPU dies use PCIe as their fabric though? Most people would immediately think the lanes between the GPU > System CPU, just based off how they explained it they are clearly pointing out that being a major bottleneck that developers avoid. Plus there is no limit on PCIe lanes AFAIK for internal fabrics. If they needed more they could just add more lanes.
 
Some of that list isn't really very exclusive. If Nvidia has 3D memory coming, AMD probably does too. TSMC announced that as a feature for its customers last year. :p

The NVLink part stands out only because AMD went back to PCIe inter-GPU communication. NVLink should be a nice feature for certain GPGPU applications, if direct GPU to GPU communication is exposed in the APIs.

These next gen announcements always make me want to finally upgrade at a bang for the buck peak of current generation GPUs.
 
Titan Z, Pascal, etc, really makes me wish I won the lottery so I could go back to Uni and do some research on hpc just to play with all this stuff.

In the mean time, back to minesweeper.
 
So, if Pascal (cough...Volta...cough) is due in 2016, is it safe to assume that 800 series Maxwell parts aren't hitting shelves until very late this year or sometime next year? Sigh...
 
Meanwhile, I have this gorgeous chunk of tropical swampland in Antarctica that will also be available in 2016...yea, buddy...;)
 
drakken said:
I'd bet money that none of the current hardware will work with pascal boards. That interlink is going to be a physical connection to cpu, and no current hardware has that connection. The server boards have something similar but no where as fast or efficient as nvidia is claiming the link will be.

The interesting twist is that if Nvidia is publicly announcing it, they have an agreement with someone making a CPU that will work with this design. So the question is who? Snapdragons are almost fast enough 800 series and later to be considered, Intel might have done it to get help with their own board cards, and some how I can't see AMD and Nvidia working together at this point, but AMD does have cross IP licenses with Intel so who knows.

Oh and Pascal was the first language I learned in High School.
Click to expand...


IBM

Let me elaborate.

They are collaborating with ibm to bring nvlink to IBM Power CPUs on servers.

I don't think we'll see nvlink in consumer MB anytime soon if ever.
 
Last edited:
pxc said:
Some of that list isn't really very exclusive. If Nvidia has 3D memory coming, AMD probably does too. TSMC announced that as a feature for its customers last year. :p

The NVLink part stands out only because AMD went back to PCIe inter-GPU communication. NVLink should be a nice feature for certain GPGPU applications, if direct GPU to GPU communication is exposed in the APIs.

These next gen announcements always make me want to finally upgrade at a bang for the buck peak of current generation GPUs.
Click to expand...
I think NVLink is Nvidia's response to AMD's hUMA and inter-GPU Crossfire via PCI-Express.

AMD already has in place the general idea of unified memory between CPU and GPU thanks to HSA before Nvidia came out with theirs. I can only assume future boards using future AMD APUs will have full HSA and hUMA integrated into them.

So, now there'll be two differing versions and I believe Nvidia is going to be a proprietary link that'll be added to future motherboards connected directly to the CPU or via the chipset and the graphics card that will work only with Nvidia Pascal GPUs and models following after that.
 
striker444 said:
Anyone else think of the programming language when they saw Pascal?
Click to expand...

I did - because I actually used to write programs in it.

While PASCAL (the programming language) predates even DOS by quite a bit, like COBOL, and even FORTRAN, it also transitioned to the world of the GUI, with compilers and IDEs available for Linux distributions, UNIX, and Windows.
 
PGHammer said:
I did - because I actually used to write programs in it.

While PASCAL (the programming language) predates even DOS by quite a bit, like COBOL, and even FORTRAN, it also transitioned to the world of the GUI, with compilers and IDEs available for Linux distributions, UNIX, and Windows.
Click to expand...

I just thought about because I took a class on it in year 11... I was terrible at it (no debug tools worth a damn).
 
I'm more and more impressed with innovation of the GPU nowadays compared to CPU's. GPU's seem much more aggressive at this point. Perhaps I'm not reading enough.
 
coolpurplefan said:
Can anyone pull some statistics out of their butt and guess when we'll have a GTX 850 Ti?
Click to expand...


Well I stand by my assessment of saying the release of the 750/750 Ti could mean one of two things. Either Maxwell is in route, or Maxwell is nowhere close to ready but they wanted to test the architecture with minimal impact by placing it in the 700 Series in a market segment focused on price and power constraints to see how the architecture performs.

Right now all this tells us (lack of Maxwell news) is that it isn't in the horizon. Nvidia has hyped Maxwell for 5 years since Project Denver was announced and probably have been working on it longer then that. It's a very important release unifying programming and memory. They will not just drop it randomly and unless I'm missing some big event Nvidia usually attends between now and the Fall, then I can't see it dropping before or during the summer.

The fact they didn't mention Maxwell ONCE with any solid details is very telling. I bet everyone in that audience was shocked, including myself.
 
GTC hasn't really ever been used for significant consumer gaming card related announcements. If they actually talked about Maxwell and upcoming gaming cards that would've actually been a first and a rather large departure of what is usually presented.

Volta is supposedly still there as a project codename and is for what is coming after Pascal. Really it should be kept in mind these are just codenames for projects, not set in stone products.

Liger88 said:
They will not just drop it randomly and unless I'm missing some big event Nvidia usually attends between now and the Fall, then I can't see it dropping before or during the summer.
Click to expand...

Computex in June and SIGGRAPH in August are two.

But in general both Nvidia and AMD have not been linking consumer desktop graphics card launches to these types of events in recent years. The increase and prevalence of the internet has considerably lowered the impact of launching products at these types of events and still allowing large scale media coverage (at least for the relevant audience).

GTX 7xx was rumored to launch at Computex for example but it ended up launching earlier. There was rumors about Kepler launching at various events such as GDC 2012 but it ended up not being tied into any of them either. I think all of Nvidia's recent announces at trade shows/tech conferences have only been Professional (Tesla/Quadro) level products? (not too sure on this, maybe Titan was which straddles the line a bit). Which makes more sense since you can actually reach out directly to that market at those events. But consumer desktop card buyers are generally not at these events and will get their information online regardless.
 
Pretty sure we will see those card in end of the year but there's no point of talking about them right now. Last time the new card came out in november so
 
Ah, Pascal. It's so fitting that nVidia would chose the name of an obsolete computer language to name its products after:

http://en.wikipedia.org/wiki/Pascal_(programming_language)

OK, I know it wasn't the language nVidia had in mind--but seriously, how do you name something computer-related today "Pascal" and not think about Pascal the programming language?
 
coolpurplefan said:
Can anyone pull some statistics out of their butt and guess when we'll have a GTX 850 Ti?
Click to expand...

Probably not till next year. The 8 series will most likely start with the 880 high end card at the end of summer or early fall. They could probably release it sooner but there is no reason to.

Your best bet would be either an 860 or a price cut 7 series.
 
octoberasian said:
I think NVLink is Nvidia's response to AMD's hUMA and inter-GPU Crossfire via PCI-Express.

AMD already has in place the general idea of unified memory between CPU and GPU thanks to HSA before Nvidia came out with theirs. I can only assume future boards using future AMD APUs will have full HSA and hUMA integrated into them.
Click to expand...
No, it's far better and that's the point. Any GPUs connected through pcie, if capable of mastering transfers, can run through that 16GB/s link without main memory or the CPU. It was just a few months between AMD's HSA and CUDA 6, so they were likely in development at the same time. :p
 
WaltC said:
Ah, Pascal. It's so fitting that nVidia would chose the name of an obsolete computer language to name its products after:

http://en.wikipedia.org/wiki/Pascal_(programming_language)

OK, I know it wasn't the language nVidia had in mind--but seriously, how do you name something computer-related today "Pascal" and not think about Pascal the programming language?
Click to expand...

Because it sounds cooler than BASIC or C++. :p
 
PRIME1 said:
Probably not till next year. The 8 series will most likely start with the 880 high end card at the end of summer or early fall. They could probably release it sooner but there is no reason to.

Your best bet would be either an 860 or a price cut 7 series.
Click to expand...

Agreed. With the 6gb cards now and the titan "z", if they put out a new family of cards they wouldn't make much money on the "new" card releases.
 
WaltC said:
OK, I know it wasn't the language nVidia had in mind--but seriously, how do you name something computer-related today "Pascal" and not think about Pascal the programming language?
Click to expand...
Most people have no idea Pascal was a programming language, even many enthusiast nerds. (Object Pascal, the language used by Delphi, had a much longer life.)

For people who don't have problems with only interpreting things in one way, Pascal (like the language, named after early mechanical calculator inventor Blaise Pascal) fits in with previous Nvidia codenames:
Fahrenheit
Celsius
Kelvin
Rankine
Curie
Tesla
Fermi
Kepler
Maxwell

No offense, but Blaise Pascal >>>>>>> you.
 
Since NVlink will require a new connector from the motherboard it sounds like PCI Express will go the way of the DoDo?

I guess I'm going to wait another year before I upgrade. Pascal with the its stacked memory sounds too good to miss and by then I imagine DDR4 will be more affordable. Upgrading now will mean, I'll need an entirely new motherboard, CPU again in 2016. Wish I was rich but 2016 looks like a good time for an entirely new build financially.

Hoping my CPU can last me till then...hopefully it doesn't bottleneck Maxwell cards...
 
PCIe is not going the way of the DoDo bird, in fact NVLink is DOA on the consumer side. It isn't meant as a replacement for PCIe except in the enterprise and distributed computing market where it makes sense because the hardware is mostly modified anyways for specific needs.

I wouldn't put Pascal on a stick. Nvidia already did that with Maxwell and so far all that hype got pushed back to Pascal. Wait and see. I'd expect DDR4 to come down a little, but it wont come down drastically until Skylake which is around the same time. I'm actually surprised we aren't seeing GDDR6 to hold us over. It appears that's also turning into a lame duck.
 
the unified memory will be a huge step forward.
I buy a high end SLI every 3 or 4 years at maximum, GTX980 SLI is my next target but, should I wait the pascal cards with their unified memory.
who knows... :(
 
You must log in or register to reply here.
Back
Top