Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
Just for fun, hot off the rumormill...
Rumor: Project Canis (Joint Intel/AMD GPU project...)
https://wccftech.com/amd-project-canis-flagship-intel-joint-venture/
I'm thinking April Fools. HBM-4 isn't due until after 2020... the specs are pretty though.
Rendering is computing. And science is often running simulations that are just as visual. They also demoed a car simulation on the new box, with real time photoreal graphics.
VRAM is what it is because of its ability to be accessed quickly by the GPU. That is all. This is all about speed. How fast is the NVSwitch? It allows the GPUs to access the collective VRAM across ALL connected GPUs at 300 GB/s. How fast is that??? To put this into perspective, that is as fast the the GTX 1080 can access its own VRAM. NVSwitch can run this fast to every GPU it is connected to! And that is precisely why NVSwitch is able to pool all the GPUs VRAM together. No other multiGPU connection was ever this fast.
And there are practical reasons. VRAM is expensive. Production is tight. NVSwitch would negate the need to pile tons of memory on a single board. For the first time ever, you could actually upgrade your VRAM...buying another GPU, yes, but that would be part of the master plan to sell more GPUs, LOL.
If they are not lying, NVSwitch is one of greatest GPU innovations...ever. This is really exciting stuff.
But the question is if they will bring this to the GTX line. Odds are they probably wont do it. It would really shake the market, and gamers are not really demanding more memory right now. The other issue is will the software support it if they do bring it, and that is a big if.
...as I recall there was some type of limitation involved that prevented stacking of VRAM between cards for render purposes as a copy of the complete scene would be loaded into each card's memory, therefore the single card VRAM ceiling we have been dealing with for years. The one benefit of multiple cards was more cores to hasten the process along or for gaming, swapping between cards to improve frame rate.
When NVLink was introduced with the GP100, on one CG forum I brought up the same question and was told that it for rendering purposes the maximum scene size would still be limited to a single card's memory resources, but would stack for pure computational purposes like running scientific simulations & such.
Be that as it may, this is pretty much moot for us as I don't see NVLink filtering down to consumer GPUs anytime soon. For one it would require HBM2 memory which would boost card prices (like the Titan-V coming in at around 2.4 times the MSRP of the Titan Xp the latter which is already pretty rich for many people's budgets). Second it would mean adding about another 900$ into the system cost just for the two NVLink connectors (which is 200$ more than the MSRP of a 1080 Ti). Third, well you already mentioned it, and it would make sense why Nvidia would consider keeping it exclusive to their top end Quadro and Tesla lines. True also, more VRAM isn't as important to gaming as it is to what we do and we are a relatively small segment of the GPU buying public.
So unless any of us here hit a lotto sometime in the very near future, we will continue to be still stuck with the single card memory limit.
Some more info on the upcoming Nvidia GTX 1180/1170 cards..
...well was right on the GDDR6 and also that the base MSRP will be higher due to development and production costs for the newer generation processor and memory. Now the only matter is how fast will these sell out if the memory speed will be even more attractive to ETH miners?
Oh and Nvidia cards with GDDR5, those probably won't be in production anymore (the Nvidia store has been out of stock for a couple weeks now), so we will be stuck with buying used.
There may be a small window where prices as stable at the outset before things get out of hand again as miners will most likely snap these new cards up by the dozens to upgrade their rigs. So for many of us it will be continue to make do with what you have (which means CPU rendering for myself), or trust that used card you got on ebay hasn't has most of the life sucked out of it. Most definitely don't buy any used 980 Ti's.
I've waited a year; i can continue to wait.
Keeping my cash in my bank account benefits me - I'm fine with that.