Rent an RTX 3080 in the Cloud With New GeForce Now Plan

[Master Disaster]

7 minutes ago, Blademaster91 said:

$99 for 6 months, and you'll need a decent internet connection, this seems like a waste of money IMO.

And I think it seems kinda suspicious that graphics cards have been unavailable for most people, yet Nvidia has enough of them to start a cloud rental service.

This seems like a slippery slope for Nvidia to keep raising prices of GPU's,  then get consumers that can't afford to purchase a dedicated GPU to pay for cloud services.

Ding, ding, ding. Congratulations, have 10 internet points and a cookie

[porina]

10 hours ago, Brooksie359 said:

Wait you bought a 2080ti then a 3070? Why? I guess if you could sell the 2080ti for a good price and then buy the 3070 it might be worth but the 2080ti and the 3070 have very similar performance so it's not like the 3070 is an upgrade over the 2080ti. 

Short version: 30 series supports beyond 60 Hz over HDMI with my TV, the 2080 Ti does not. I did re-sell the 2080Ti for close to what I bought it for so it was essentially free while I had it.

 

Long version: 

 

 

[Brooksie359]

6 hours ago, porina said:

Short version: 30 series supports beyond 60 Hz over HDMI with my TV, the 2080 Ti does not. I did re-sell the 2080Ti for close to what I bought it for so it was essentially free while I had it.

 

Long version: 

 

 

Oh I didn't know that. Is it the new hdmi standard that supports 4k 120hz? That makes alot more sense. 

[Quackers101]

Nvidia be like, need a GPU?? we have one, in the cloud through our services!

You want a real GPU? just wait for our SUPER cards that will up the price and out of stock even more!

Green has the eye on the money tree.

[MageTank]

21 hours ago, kuddlesworth9419 said:

Make GPU's too expensive for normal people to buy them so their only option is to invest in a subscription plan?

Nvidia's FE cards are the only cards that remain at "MSRP" through reputable retailers. This is because Nvidia has clauses in their contracts that prohibit retailers from selling these cards at higher prices if they are sourced directly from them. One of the reasons I've been looking to integrate with their cards (though Best Buy seems to consume a ton of the supply here in the US).

 

9 hours ago, Master Disaster said:

Semantics. Nvidia have ultimate control over who can and can't buy their products and they can somehow find enough of them to start a rental service while to everyone else they are rocking horse shit.

I think you are overthinking this conspiracy. This isn't the first time Nvidia has launched a service like this. Those of us with the original handheld Shield devices remember this offer packaged under the Grid K1/K2 cards which offered "4 Kepler GPU's" and "2 Kepler High-End GPU's" respectively: https://www.nvidia.com/content/cloud-computing/pdf/nvidia-grid-datasheet-k1-k2.pdf

 

We have no reason to doubt that they are not implementing a similar virtual GPU configuration now, using cards not commonly available to consumers but giving them a label for a performance metric they would understand, such as the RTX 3080. Let's use the K1 as an example. These had 4 GK107-450-A2's in them, the same GPU die used in the GTX 650, but used slower DDR3 memory instead of GDDR5, putting them closer to the GT 640 in memory bandwidth performance. The K2 had two GK104-895-A2 dies, each having 1536 shader cores's, 128 TMU's and 32 ROP's (all x2), meaning this K2 is actually a GTX 690 (offering similar performance to dual GTX 770's).

 

It would be more feasible for Nvidia to design another GRID card that specializes in virtualization and run multiple GPU dies on a single card, cutting down what isn't needed to deliver 1440p 120hz (or 4k 60) and save on space, plugging several of these multi-GPU cards into a single system.

 

Their blog almost suggests that this is the case: https://blogs.nvidia.com/blog/2021/10/21/geforce-now-rtx-3080/

Quote

The next generation of cloud gaming is powered by the GeForce NOW SuperPOD, built on the second-gen RTX, NVIDIA Ampere architecture.

What is a "SuperPOD"? https://images.nvidia.com/aem-dam/Solutions/Data-Center/nvidia-dgx-superpod-datasheet.pdf

A massive cluster of A100's: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf. A100's specifically having support for up to 7 Multi-Instance GPU users simultaneously. This, coupled with their relaxed advertised settings of 120hz at 1440p and their lack of mentioning "max settings" leads me to believe they'll go with medium-high and DLSS to use as little of these GPU's resources as they can while delivering on that promise.

 

"RTX 3080-Class Power" is also a very weird thing to say if it's legitimately an RTX 3080. You'd just say "RTX 3080 Power" or "Powered by RTX 3080". At this point, it's almost as if they turned "RTX 3080 " into a marketing term, lol.

9 hours ago, Blademaster91 said:

$99 for 6 months, and you'll need a decent internet connection, this seems like a waste of money IMO.

And I think it seems kinda suspicious that graphics cards have been unavailable for most people, yet Nvidia has enough of them to start a cloud rental service.

This seems like a slippery slope for Nvidia to keep raising prices of GPU's,  then get consumers that can't afford to purchase a dedicated GPU to pay for cloud services.

There is one glaring problem with this logic though. Nvidia doesn't want to lose market share, nor do they want to give AMD and Intel an untapped price segment that they themselves are not dominant in. If they price too high, AMD and Intel will gladly offer lower prices in an attempt to steal back that market from Nvidia.

 

Let's humor the idea that Nvidia themselves is intentionally driving up the prices of GPU's in order to turn a quick profit. Why would it make sense for a company this obsessed with money (per this theory) to alienate their biggest customers (crypto-miners) by releasing LHR cards? Would it not make more sense to just crank out an infinite supply of non-LHR, mining-optimized GPU's and sell them at a premium to customers that won't even complain about the prices? I am not defending Nvidia here, but I just don't see this being the case if Nvidia was doing this simply for the money, especially when far more lucrative offers presented themselves to them and they walked away from it in favor of catering to gamers.

 

Now speaking to what I said above, Nvidia is extremely strict on their GPU pricing. I work in system integration and buy GPU's in bulk for this process, yet they are the only ones offering consistent pricing. AIB pricing has gotten absurd and it's practically the wild west every time you place a PO. This is assuming one of your trucks didn't get robbed for GPU's (yes, this is a thing that happens now).

 

If Nvidia themselves were charging higher prices for their GPU's, I'd have to imagine I would see it on the Founders Edition PO's, but they are always at "MSRP".

 

Now what IS curious about this business model is whether or not Nvidia can feasibly deliver this service without tremendous input lag and compression artifacts. Having owned the original shield (and current generation Shield TV Pro), they've yet to make their gamestreaming service seamless. Add in this lofty goal of offering 1440p120/4k60 to a larger audience and it sounds like a recipe for disaster trying to stream that to tons of people. Given my experience with their previous attempts at gamestreaming, I just can't see this working out well for them.

 

Also begs the question, does Nvidia plan on offering free upgrades to newer hardware when this comes out? Or if this like a cell phone plan that needs upgraded for a later model?

[Mark Kaine]

On 10/21/2021 at 6:59 PM, BobVonBob said:

Mice and any latency have never gotten along for me. Controllers are fine, but mouse control feels like you're controlling the game via rubber band.

Try playing hatsune miku on a TV with 30ms input lag with a controller...

 

 

OT: "an RTX 3080 in the Cloud"

 

"How about no?!"

[Anarchy101]

An RTX 3080ti (used mining card) for only a $100?? Can anyone tell me if this is for real??

 

https://aboutsuperior.com/product/rtx-3080-ti-gigabyte-geforce-rtx-3080-ti-vision-oc-12g-graphics-card-12gb-384-bit-gddr6x-gv-n308tvision-oc-12gd-video-card-used-mining-card/

[Fnige]

On 10/23/2021 at 7:58 PM, Anarchy101 said:

An RTX 3080ti (used mining card) for only a $100?? Can anyone tell me if this is for real??

 

https://aboutsuperior.com/product/rtx-3080-ti-gigabyte-geforce-rtx-3080-ti-vision-oc-12g-graphics-card-12gb-384-bit-gddr6x-gv-n308tvision-oc-12gd-video-card-used-mining-card/

heheheh.

 

No.

[Nizkus]

On 10/21/2021 at 10:52 PM, porina said:

I have to wonder about practicalities too.

 

For example, 8 hour a day limit. Might be ok for casual gamers, but if you want to binge on a weekend for example, that might not cut it.

It's 8 hours per session not day, you just relaunch the stream after that and play for another 8h. Unless they changed from how things currently work and I missed it. 

[porina]

28 minutes ago, Nizkus said:

It's 8 hours per session not day, you just relaunch the stream after that and play for another 8h. Unless they changed from how things currently work and I missed it. 

That would be better. The "8 hours per day" was from the quote in OP, and I didn't check further. Still, it implies they need to periodically kick people off to free resource, and don't just afk 24/7.

[Lightwreather]

2 hours ago, Nizkus said:

It's 8 hours per session not day, you just relaunch the stream after that and play for another 8h. Unless they changed from how things currently work and I missed it. 

Huh, I guess I misread that. Thanks

[igormp]

On 10/22/2021 at 2:56 PM, MageTank said:

A100's specifically having support for up to 7 Multi-Instance GPU users simultaneously. This, coupled with their relaxed advertised settings of 120hz at 1440p and their lack of mentioning "max settings" leads me to believe they'll go with medium-high and DLSS to use as little of these GPU's resources as they can while delivering on that promise.

Just a FYI, the A100 wouldn't be capable to be used in those scenarios since it doesn't support DirectX and also lacks RT and all of those fancy things gamers like. (this applies to any GA100 based chip, such as the A30), so it's likely a A40 or any other GA102-based device.

 

But yeah, it makes no sense to use a "regular" 3080 inside a data center.

[Nizkus]

4 hours ago, porina said:

That would be better. The "8 hours per day" was from the quote in OP, and I didn't check further. Still, it implies they need to periodically kick people off to free resource, and don't just afk 24/7.

They already have time limits for free and paid tiers, 1h and 6h respectively, where you need to reconnect and you are put to the bottom of the queue, so I was basing it on that.

 

Granted I haven't read more than headline as far as this new tier goes so I might as well be making an ass out of myself.