NVIDIA Readies RTX 3060 8GB and RTX 3080 20GB Models

[btarunr]

A GIGABYTE webpage meant for redeeming the RTX 30-series Watch Dogs Legion + GeForce NOW bundle, lists out eligible graphics cards for the offer, including a large selection of those based on unannounced RTX 30-series GPUs. Among these are references to a "GeForce RTX 3060" with 8 GB of memory, and more interestingly, a 20 GB variant of the RTX 3080. The list also confirms the RTX 3070S with 16 GB of memory.

The RTX 3080 launched last week comes with 10 GB of memory across a 320-bit memory interface, using 8 Gbit memory chips, while the RTX 3090 achieves its 24 GB memory amount by piggy-backing two of these chips per 32-bit channel (chips on either side of the PCB). It's conceivable that the the RTX 3080 20 GB will adopt the same method. There exists a vast price-gap between the RTX 3080 10 GB and the RTX 3090, which NVIDIA could look to fill with the 20 GB variant of the RTX 3080. The question on whether you should wait for the 20 GB variant of the RTX 3080 or pick up th 10 GB variant right now, will depend on the performance gap between the RTX 3080 and RTX 3090. We'll answer this question next week.



View at TechPowerUp Main Site

 

[Raendor]

There also 3070S with 16 gigs on that same list.

 

[SIGSEGV]

I am gonna patiently wait for the 1Tb version.

 

[my_name_is_earl]

And it's gone in 3.... 2.....1

 

[btarunr]

I am gonna patiently wait for the 1Tb version.

The logical next step to DirectStorage and RTX-IO is graphics cards with resident non-volatile storage (i.e. SSDs on the card). I think there have already been some professional cards with onboard storage (though not leveraging tech like DirectStorage). You install optimized games directly onto this resident storage, and the GPU has its own downstream PCIe root complex that deals with it.

So I expect RTX 40-series "Hopper" to have 2 TB ~ 8 TB variants.

 

[BArms]

Maybe they should launch the original 3080 first? I really don't consider meeting less than 1% of the demand a real launch.

 

[hat]

Am I the only one that thinks it's a little bit of a reach to posit that normal (even NVMe) storage and the CPUs handling it are too slow, and that the idea of GPU-assisted storage is a little bit ludicrous?

 

[BArms]

Am I the only one that thinks it's a little bit of a reach to posit that normal (even NVMe) storage and the CPUs handling it are too slow, and that the idea of GPU-assisted storage is a little bit ludicrous?

Not sure but I thought the idea was to load textures/animations/model vertex data etc into VRAM where it needs to go anyway.

 

[R0H1T]

the idea of GPU-assisted storage is a little bit ludicrous?

Well technically it's the storage assisting GPU (DirectStorage) unless you meant something like this?

AMD announces Radeon Pro SSG: A GPU with 1TB of SSD

While it may not seem much this will reduce a lot of CPU overhead & free it for more FPS.

 

[AusWolf]

If nVidia taught us something with the 20 (Super) series, it's the fact that early buyers get inferior products. That's why I'm going to wait for the 3070 Super/Ti with 16 GB VRAM and a (hopefully) fully unlocked die, unless AMD's RDNA 2 proves to be a huge hit.

 

[nikoya]

they should release the 3080 20gb ASAP
or at least give us a date a price now.

that we can make a decision.

 

[AusWolf]

they should release the 3080 20gb ASAP
or at least give us a date a price now.

that we can make a decision.

I doubt that they will. With all the hype around the 10 GB version, they can easily milk the cow twice by making informed buyers wait a bit longer.

 

[R0H1T]

they should release the 3080 20gb ASAP
or at least give us a date a price now.

that we can make a decision.

NVidia's not in the business of making it easy for you, as a consumer you have to decide what's in your best interests.

I doubt that they will. With all the hype around the 10 GB version, they can easily milk the cow twice by making informed buyers wait a bit longer.

Yup, they're making a pretty penny right now so why would they stop?

 

[InVasMani]

It's not listed, but could see a RTX 3060S 16GB being possible as well eventually. That probably depends a lot on the memory controller bandwidth however, but knowing Nvidia if they can get some better binned GDDR chips down the road they'd slap on the extra ram density and faster chips and start selling them to fill whatever gaps can.

 

[Hossein Almet]

I say the 3080 20G will cost somewhere between $1000 and $1200, so buying the 3080 10G now is not a bad thing to do.

 

[ZoneDymo]

yay more SKU's just what I always wanted, THANKS NVIDIA!

If nVidia taught us something with the 20 (Super) series, it's the fact that early buyers get inferior products. That's why I'm going to wait for the 3070 Super/Ti with 16 GB VRAM and a (hopefully) fully unlocked die, unless AMD's RDNA 2 proves to be a huge hit.

Nooo be like the sheep, BUY INTO THE HYPE NOW, heck just get 1 now and another later, you will save money that way!

 

[Valantar]

It's not listed, but could see a RTX 3060S 16GB being possible as well eventually. That probably depends a lot on the memory controller bandwidth however, but knowing Nvidia if they can get some better binned GDDR chips down the road they'd slap on the extra ram density and faster chips and start selling them to fill whatever gaps can.

Why on earth would a 60-tier GPU in 2020 need 16GB of VRAM? Even 8GB is plenty for the types of games and settings that GPU will be capable of handling for its useful lifetime. Shader and RT performance will become a bottleneck at that tier long before 8GB of VRAM does. This is no 1060 3GB.

I doubt that they will. With all the hype around the 10 GB version, they can easily milk the cow twice by making informed buyers wait a bit longer.

RAM chip availability is likely the most important part here. The FE PCB only has that many pads for VRAM, so they'd need double density chips, which likely aren't available yet (at least at any type of scale). Given that GDDR6X is a proprietary Micron standard, there's only one supplier, and it would be very weird if Nvidia didn't first use 100% of available capacity to produce the chips that will go with the vast majority of SKUs.

The logical next step to DirectStorage and RTX-IO is graphics cards with resident non-volatile storage (i.e. SSDs on the card). I think there have already been some professional cards with onboard storage (though not leveraging tech like DirectStorage). You install optimized games directly onto this resident storage, and the GPU has its own downstream PCIe root complex that deals with it.

So I expect RTX 40-series "Hopper" to have 2 TB ~ 8 TB variants.

Does that actually make sense, though? That would mean the GPU sharing the PCIe bus with storage for all CPU/RAM accesses to said storage (and either adding some sort of switch to the card, or adding switching/passthrough capabilities to the GPU die), rather than it being used for only data relevant to the GPU. Isn't a huge part of the point of DirectStorage the ability to transfer compressed data directly to the GPU, reducing bandwidth requirements while also offloading the CPU and also shortening the data path significantly? The savings from having the storage hooked directly to the GPU rather than the PC's PCIe bus seem minuscule in comparison to this - unless you're also positing that this on-board storage will have a much wider interface than PCIe 4.0 x4, which would be a whole other can of worms. I guess it might happen (again) for HPC and the like, for those crunching multi-TB datasets, but other than that this seems nigh on impossible both in terms of cost and board space, and impractical in terms of providing actual performance gains.

Btw, the image also lists two 3080 Super SKUs that the news post doesn't mention.

 

[AusWolf]

yay more SKU's just what I always wanted, THANKS NVIDIA!



Nooo be like the sheep, BUY INTO THE HYPE NOW, heck just get 1 now and another later, you will save money that way!

To be honest, I might have gone with the non-Super skus if I hadn't found 3080 benchmarks so disappointing, compared to initial promises (it's NOT twice as fast as the 2080 Ti, and performance/power data are barely on par despite the new 8 nm process).

Edit: Speaking of which, am I the only one who finds it funny how many more cuda cores the 3080 has compared to any 20 series chip, and still doesn't offer comparably more performance? I mean, the efficiency of single cuda cores must have dropped considerably with Ampere.

 

[londiste]

Why on earth would a 60-tier GPU in 2020 need 16GB of VRAM? Even 8GB is plenty for the types of games and settings that GPU will be capable of handling for its useful lifetime. Shader and RT performance will become a bottleneck at that tier long before 8GB of VRAM does. This is no 1060 3GB.

Because it is not the classic 60-tier GPU. From leaks and details we have so far, both 3070 and 3060 will be based on GA104. When it comes to performance we will have to wait and see but 3060 should be roughly around PS5/XBSX performance level, so it will be enough for a long while. 16GB is more of a marketing argument (and maybe preemptive strike against possible AMD competition).

Does that actually make sense, though? That would mean the GPU sharing the PCIe bus with storage for all CPU/RAM accesses to said storage (and either adding some sort of switch to the card, or adding switching/passthrough capabilities to the GPU die), rather than it being used for only data relevant to the GPU. Isn't a huge part of the point of DirectStorage the ability to transfer compressed data directly to the GPU, reducing bandwidth requirements while also offloading the CPU and also shortening the data path significantly? The savings from having the storage hooked directly to the GPU rather than the PC's PCIe bus seem minuscule in comparison to this - unless you're also positing that this on-board storage will have a much wider interface than PCIe 4.0 x4, which would be a whole other can of worms. I guess it might happen (again) for HPC and the like, for those crunching multi-TB datasets, but other than that this seems nigh on impossible both in terms of cost and board space, and impractical in terms of providing actual performance gains.

PCI-e 4.0 x16 does not really seem to be a bottleneck yet and probably won't be a big one for a long while when we look at how the scaling testing has gone with 3.0 and 2.0. Fitting 4 lanes worth of data shouldn't matter all that much. On the other hand, I think this shader-augmented compression is short-lived - if it proves very useful, compression will move into hardware as it has already supposedly done in consoles.

Moving the storage to be attached to GPU does not really make sense for desktop/gaming use case. More bandwidth through compression and some type of QoS scheme to prioritize data as needed should be enough and this is where it really seems to be moving towards.

 

[T3RM1N4L D0GM4]

Release date or GTFO!
(a real release date, not a paper launch, please)

 

[BluesFanUK]

Are people that desperate they can't wait another month to see what AMD have to offer? Nvidia have a history of mugging off their customers, day one purchasers are setting themselves up for buyers remorse.

Give it a month or two and who knows, Turing cards may start tumbling or AMD could knock it out the park. Personally i'm just being sensible and buying a PS5 for the same cost as what one of these overpriced GPU's cost.

 

[phill]

Whatever happened to just giving the different levelled tiers of GPUs normal amounts of VRAM... 1, 2, 4, 8, 16, 32GB etc.. I don't get it why we need 10GB or 20GB....

 

[Caring1]

Release date or GTFO!
(a real release date, not a paper launch, please)

So can prep your bots?

 

[8BitZ80]

AMD buyer's cycle: "I'm waiting for the next generation" [New product launches] "Seems nice, but I'll wait for the next one which will be even better"

Nvidia buyer's cycle: "I NEED IT RIGHT NOW, GIMME GIMME" [New product launches] "SHIT, I SHOULD'VE WAITED FOR THE BETTER VERSION"

 

[Bubster]

Glad this news update came out to calm the ampere hysteria down...3080 with 20 gig