[VideoCardz] NVIDIA to preview 20nm high-end Maxwell tomorrow

[Zanovar]

At the time of release, Intel will also field Skylake with PCIe 4.0.

Erm,cheers for that i guess.

 

[dangerman1337]

NVLink is PCIe based. But its only used for GPU to GPU.

*snip*

It helps to do the homework before complaining.

Does this mean that if you have a PCI-E express 3.0 slots in a motherboard NVLink will be already to work on it?

 

[R0H1T]

At the time of release, Intel will also field Skylake with PCIe 4.0.

You mean Intel will release skylake in 2016 ?

 

[ShintaiDK]

Does this mean that if you have a PCI-E express 3.0 slots in a motherboard NVLink will be already to work on it?

NVLink is not used to link your CPU to your gaming card.

 

[chimaxi83]

With the way they're talking about NVLInk working, and with what the connection method will be like, this seems to be a first step towards not needing a full PCI style card anymore. Almost like a socketed GPU if I'm understanding it correctly, correct me if I'm wrong.

 

[NTMBK]

NVLink is not used to link your CPU to your gaming card.

It can be used to connect a CPU to a GPU, but no, it isn't for gaming rigs.

From http://devblogs.nvidia.com/parallelforall/nvlink-pascal-stacked-memory-feeding-appetite-big-data/

It's meant for supercomputers, hence the POWER link. And cache coherence is coming in the second generation:

EDIT: And just in case you doubt the POWER tie in, it is made explicit in this blog post http://nvidianews.nvidia.com/Releas...ng-Pave-the-Way-to-Exascale-Computin-ad6.aspx

With NVLink technology tightly coupling IBM POWER CPUs with NVIDIA Tesla® GPUs, the POWER data center ecosystem will be able to fully leverage GPU acceleration for a diverse set of applications, such as high performance computing, data analytics and machine learning.

 

[NTMBK]

That's fine, but what about the desktop X86 market. I can't imagine that they don't need to sell to that market as well. AMD will want to have an industry standard (as they always do) for unified memory and will want it to include dGPU. nVidia is not too proud to use standards like Display Port with GSync.

Just a possibility. Curious to see what happens.

Hmm, potentially, but I'm not sure whether AMD would really go for that; they would be much better off making higher-powered APUs with a single shared memory pool. And even if AMD created the standard, I doubt that Intel would support it- Intel want to push their Xeon Phi in the HPC market, not help out Tesla.

 

[ViRGE]

Hmm, the mezzanine connector shown in AT's link doesn't look too practical to me. Looks very.. very expensive.

It's just an example of what a mezzanine connector looks like, according to AT. It's not the NVLink connector.

 

[DeathReborn]

With the way they're talking about NVLInk working, and with what the connection method will be like, this seems to be a first step towards not needing a full PCI style card anymore. Almost like a socketed GPU if I'm understanding it correctly, correct me if I'm wrong.

It is indeed more socket like. With Intel/AMD moving the Southbridge & Northbridge onto the CPU it leaves some room on ATX & E-ATX boards for one of these to be installed. Being point to point & power provided through the board makes this ideal for not just Tesla & Quadro but any HPC/GPGPU based chip would benefit greatly.

 

[Subyman]

You guys ready to upgrade your Audi with new computer parts? lol, and he actually expected an applause from that.

Only a GK110 made with laffy taffy could be stretched further than what Nvidia's done over the last year and a half. I cringe to think what they will do if Maxwell 20nm comes out in 2015. They've already iterated the memory to 6GB, maybe we will get a 780 TI Black NOS version?

 

[tential]

You guys ready to upgrade your Audi with new computer parts? lol, and he actually expected an applause from that.

Only a GK110 made with laffy taffy could be stretched further than what Nvidia's done over the last year and a half. I cringe to think what they will do if Maxwell 20nm comes out in 2015. They've already iterated the memory to 6GB, maybe we will get a 780 TI Black NOS version?

You laugh now but wait until my new Audi has dual Titan Black-Z's in it. It'll be the fastest car on the planet!!!!!

 

[thilanliyan]

You laugh now but wait until my new Audi has dual Titan Black-Z's in it. It'll be the fastest car on the planet!!!!!

Mine has stickers on it, which doubled my horsepower. I'm pretty sure my one is the fastest on the planet.

 

[moonbogg]

I expect mid range Maxwell this year. I also expect people to say that they are the true gaming cards and that the big die isn't meant for gaming, just like people lied to themselves about GK110 not being a gaming chip. That made it easier to shell out $500.00 for a GTX 680. Same thing this time.

 

[Artorias]

Not to bring this completly off topic, but does anyone know the roadmap for AMD 20nm? Its been quite silent at AMD.

Start a separate discussion thread for this. It is not on topic for this discussion.
-- stahlhart

 

[coolpurplefan]

Not to bring this completly off topic, but does anyone know the roadmap for AMD 20nm? Its been quite silent at AMD.

Start a separate discussion thread for this. It is not on topic for this discussion.
-- stahlhart

20nm is the topic. Have you googled "cognitive distortions"?

Warning issued for moderator callout.
-- stahlhart

 

[moonbogg]

20nm is the topic. Have you googled "cognitive distortions"?

Seriously dude?

 

[stahlhart]

Anyone who is interesting in doing so is more than welcome to discuss a 20nm roadmap for AMD -- in a separate thread. Not here. This thread is for discussing Maxwell.
-- stahlhart

 

[xpea]

I have no idea why people believe that it will be high performance 20nm GPUs. And so called "journalists" responsible for such claim should be banned from the web. Really.
They are only one 20nm process at TSMC and it's called 20SoC. This process is not adapted for power hungry GPUs. It's for... SoCs. Period.
We may see a mid range Maxwell this year but it will still be 28HP. For gen2 top range Maxwell, we have to wait for 16FinFet. sadly...

 

[sontin]

Yes, it should be obvious that Maxwell will be still on 28nm. GM107 delivers nearly 2x the performance in the same power range while only 15% bigger than GK107.

 

[ShintaiDK]

I have no idea why people believe that it will be high performance 20nm GPUs. And so called "journalists" responsible for such claim should be banned from the web. Really.
They are only one 20nm process at TSMC and it's called 20SoC. This process is not adapted for power hungry GPUs. It's for... SoCs. Period.
We may see a mid range Maxwell this year but it will still be 28HP. For gen2 top range Maxwell, we have to wait for 16FinFet. sadly...

While the processnode is optimized for low power. It doesnt exclude high performance GPUs as you claim. Just look on Ivy and Haswell for example.

20nm+Maxwell should give amazing performance/watt tho.

 

[sontin]

Huang said yesterday at the investor conference that they will not jump to a new node without other customers like they did it with 28nm.

Considering this and that you you can archive the same perf/watt jump with an architecture 28nm makes more sense than 20nm. At least from a $/perf standpoint.

 

[moonbogg]

I miss the good old days. Best of a generation comes first. GTX 280, GTX 480, GTX 580, GTX 680 (wait WTH?) GTX 690 (Dual mid range wth?) Titan (WTH is this crap?) GTX 780 (Oh good lord, finally I think?) GTX 780 Ti (The real deal, again, huh?!) GTX 780 Ti 6gb (Real deal AGAIN! OMG!)

 

[sontin]

Na, GTX280 came 8 months after G92. GTX480 was released in April - nearly 10months after nVidia's first 40nm products.

 

[Kronvict]

You guys ready to upgrade your Audi with new computer parts? lol, and he actually expected an applause from that.

Only a GK110 made with laffy taffy could be stretched further than what Nvidia's done over the last year and a half. I cringe to think what they will do if Maxwell 20nm comes out in 2015. They've already iterated the memory to 6GB, maybe we will get a 780 TI Black NOS version?

I miss the good old days. Best of a generation comes first. GTX 280, GTX 480, GTX 580, GTX 680 (wait WTH?) GTX 690 (Dual mid range wth?) Titan (WTH is this crap?) GTX 780 (Oh good lord, finally I think?) GTX 780 Ti (The real deal, again, huh?!) GTX 780 Ti 6gb (Real deal AGAIN! OMG!)

Amen to that. nVidia needs to let keplar rest and bring on high end maxwell. All these continued keplar iterations are making my head spin. I see no reason to upgrade my 680s to another keplar chip at this point in the cycle.

 

[spat55]

I just got a Boner!!! (Joking) But this is awesome just aswell I didn't go for a R9 290