RUMOR: NVIDIA Reportedly Cancels Launch of RTX 3080 20 GB, RTX 3070 16 GB

[Master Disaster]

1 minute ago, Bombastinator said:

I was pointing out that the implication of the rumor is even more tenuous that the rumor but it won’t stop it from being built on.  Makes an argument for the taken position of “it hasn’t happened till it happens”.  It’s a reasonable attitude to take.  Not the only one, but not unreasonable.  It’s got it’s own problems.  It can be abused too.  The very existence of the 16gb 3070 could have been BS the whole time.  If this is so, the rumor of it being removed is impacted BS.

This seems unlikely since we have seen official lineups from all the big AIBs listing them and even Silverstone listed them in PSU compatibility charts. The chances 4 or 5 AIBs and a PSU manufacturer all listing cards that never existed is tiny.

[Bombastinator]

1 minute ago, SupaKomputa said:

AMD usually support non proprietary open based software like OpenCL / OpenGL instead of making their own CUDA. Example, Freesync which AMD chose to make it open / non restrictive (even Nvidia can use it).

True.  They often write it though.  Making such stuff open lowers costs.  Other people and companies help. Very occasionally it’s so good for whatever reason it more or less kills the proprietary stuff, the way freesynch is eating gsynch. 

[Master Disaster]

12 minutes ago, SupaKomputa said:

AMD usually support non proprietary open based software like OpenCL / OpenGL instead of making their own CUDA. Example, Freesync which AMD chose to make it open / non restrictive (even Nvidia can use it).

Mantle says hello. Yeah I do realise AMD only created Mantle to push the industry into Vulkan but it was still proprietary.

 

Also AFAIK Freesync is simply AMDs branding of VESAs open variable refresh rate standard. I don't think AMD actually created it.

 

Edit - Nevermind. While the software side of Freesync is based on VESAs standard AMD did created some hardware to control the panels refresh rate.

[Bombastinator]

4 minutes ago, Master Disaster said:

This seems unlikely since we have seen official lineups from all the big AIBs listing them and even Silverstone listed them in PSU compatibility charts. The chances 4 or 5 AIBs and a PSU manufacturer all listing cards that never existed is tiny.

Like I said, it’s got its own problems.

[Moonzy]

either nvidia is sick of the leaks and decided to give consumers a middle finger just to spite them (i can see nvidia doing this, that's how petty i think they are)

or big navi wasnt as big of a deal as they thought, so no point releasing a "better" card

or maybe they'll use different manufacturer for those cards because samsung was disappointing(?)

[Master Disaster]

24 minutes ago, Moonzy said:

either nvidia is sick of the leaks and decided to give consumers a middle finger just to spite them (i can see nvidia doing this, that's how petty i think they are)

or big navi wasnt as big of a deal as they thought, so no point releasing a "better" card

or maybe they'll use different manufacturer for those cards because samsung was disappointing(?)

Or maybe they realise they cannot keep up with demand for current SKUs so launching new ones right now is not the best idea.

[AlTech]

Not An Apple Fan talked about this last week and basically said VideoCardz were making this up.

[Spicedaddy]

3080 should have 16GB and 3070 should have 12GB. These cards don't have enough ram.

 

I'm waiting for next year. nVidia will probably release new cards built on TSMC 7nm with more ram.

 

 

[Random_Person1234]

3 hours ago, Moonzy said:

or maybe they'll use different manufacturer for those cards because samsung was disappointing(?)

Theres rumors that Nvidia is going to use TSMC 7nm for an Ampere refresh next year.

[tikker]

Those alleged cards never made sense to me anyway, nor the hate around the "only" 10 GB.

[Blademaster91]

42 minutes ago, Random_Person1234 said:

Theres rumors that Nvidia is going to use TSMC 7nm for an Ampere refresh next year.

Perhaps a 3080 Super or tI card? Nvidia would have to differentiate a 7nm Ampere somehow or that would be really confusing lol.

14 hours ago, VeganJoy said:

exactly my thought process throughout this whole thing lol, even with all the leaks this would have made absolutely zero sense. 20gb of vram is completely ridiculous for gaming purposes, and that gets close enough to a 3090 that a 20gb 3080 would need to be 1300 bucks or something to not infringe on its sales. and 16gb of vram on an xx60/xx70 sku? those cards probably wont be powerful enough to leverage that kind of capacity anyways

 

For gaming yeah 20GB is overkill, but 10GB doesn't seem like enough because we had 11GB on a consumer GPU  with the 1080Ti, IMO the 3080 should have 12GB vram minimum for the price they're asking.

[pas008]

6 hours ago, Master Disaster said:

This seems unlikely since we have seen official lineups from all the big AIBs listing them and even Silverstone listed them in PSU compatibility charts. The chances 4 or 5 AIBs and a PSU manufacturer all listing cards that never existed is tiny.

Maybe its an AIB thing only for them to decide

Many series have had nvidias reference modelthen aib double vram models

Example 512 to 1gb

896 to 1.7

1gb to 2

2 gb to 4

 

Also maybe aibs copy and paste method for cooling and designs and fact mcm is around the corner might be a waste of resources for aibs to even do this right now

[pas008]

24 minutes ago, Blademaster91 said:

Perhaps a 3080 Super or tI card? Nvidia would have to differentiate a 7nm Ampere somehow or that would be really confusing lol.

For gaming yeah 20GB is overkill, but 10GB doesn't seem like enough because we had 11GB on a consumer GPU  with the 1080Ti, IMO the 3080 should have 12GB vram minimum for the price they're asking.

Isn't nvidia using tensor memory compression now though

[porina]

Been giving this some thought. If we assume the claim of the 3080 20GB (and 3070 16GB) cards being "cancelled" are correct, what does that mean? Firstly let's separate the perspective from the masses like us, and AIBs. To us, the high VRAM cards do not exist. To the AIBs who would be making them, they do exist in some form. nvidia would have given them some design info to prepare for it. I think it is therefore possible for it to be cancelled from that perspective. What was previously known as those cards is no longer in nvidia's future.

 

We have about a week before we know what the real AMD offerings are, but let's assume for now AMD will be releasing big VRAM card options at positions lower than 3090 equivalent. nvidia will still want to have a counter to that in some form. Logically to me, they will still produce a card in that space, just that it wont be called 3080. For example, they may pull forward a 3080 Super, or 3080 Ti, or 3085. Whatever it is called, it will still be the high vram card. By not keeping the same model name, they're less tied to the same core configuration. If nvidia expect the AMD cards to perform higher than they did previously, this allows nvidia an opportunity to increase the core counts a bit to also attack that.

 

Assuming the GPUs are drop in compatible, and I don't see any reason for this to not be the case, there is not really new technical design work to be done by AIBs. They'll just get the updated GPUs and drop them in the existing board design. There will be some extra work in other activities, like marketing content, regulatory filings, and such.

[Briggsy]

57 minutes ago, porina said:

Assuming the GPUs are drop in compatible, and I don't see any reason for this to not be the case, there is not really new technical design work to be done by AIBs. They'll just get the updated GPUs and drop them in the existing board design. There will be some extra work in other activities, like marketing content, regulatory filings, and such.

Quite likely drop in compatible, so long as a new SKU uses the same 102 die as the 3080 and 3090, then there should be a lot of configurations available to board partners with existing designs. As a few youtubers have mentioned, a GDDR6 non-x card with 12GB would have roughly the same bandwidth as a 10GB with GDDR6X. Nvidia have had multiple SKUs using the same card name before, albeit on the low end. They could also have more cores unlocked to edge out over the 3080, but I have doubts.

 

There are 2 problems with a theoretical 3080ti. One is the lack of supply, and two is just how much faster this "3080ti" would actually be over a 3080. Even if it landed between a 3080 and 3090 for performance, that's not a very compelling purchase given that the 3090 performs 0-20% better than the 3080 depending on the scenario, and the yields for the 3090 chip can't be very good to begin with, so having a higher cuda core count 3080ti just doesn't make economical sense right now using Samsung. 

 

There is one caveat though: If a 12GB model using GDDR6 was created, the memory would use a lot less power than GDDR6X, giving more headroom to the GPU to clock just that much higher. This still doesn't seem compelling though. Nvidia really seem to have painted themselves into a corner, and now they're stuck (and so are we) waiting for the proverbial paint to dry.

[porina]

1 hour ago, Briggsy said:

As a few youtubers have mentioned, a GDDR6 non-x card with 12GB would have roughly the same bandwidth as a 10GB with GDDR6X.

A move from 10GB to 12GB isn't really a response to AMD apparently releasing 16GB cards though, even if it does give more flexibility if memory availability is a limiting factor.

 

1 hour ago, Briggsy said:

There are 2 problems with a theoretical 3080ti. One is the lack of supply, and two is just how much faster this "3080ti" would actually be over a 3080.

nvidia are not competing with themselves, they're going to be targeting AMD. Forget the 3090 exists, the battle will be fought around the 3080 area. If nvidia need a little extra power to go against AMD, that seems a likely area to do so.

 

1 hour ago, Briggsy said:

Nvidia really seem to have painted themselves into a corner, and now they're stuck (and so are we) waiting for the proverbial paint to dry.

I wouldn't say having all these options open are a bad thing. They're in a position where they have made the first move of this timeframe. AMD will respond to that. How, we have to wait and see exactly. So nvidia's consideration is if they need to respond to that. 

 

21 minutes ago, Abyssal Radon said:

AMD better have taken notes on nVidia's fuck ups here, and hopefully will be ready for the demand. 

Orders at a fab are placed far in advanced and scheduled accordingly, especially as busy and popular a fab as TSMC are on their more advanced processes. You do not just call them up and order some extras in the short term. Whatever AMD has scheduled, is what they have. From their behaviour, it is not looking likely they'll have enough in the short term either.

[ewitte]

Could be they just see more value in doing a refresh.  AFAIK this is an AIB configuration.  I'm not sure *IF* the contracts with AIBs states they are allowed to that they can backtrack without renegotiating new contracts.

 

There are no DDR6X shortages NVIDIA has huge stockpiles.

 

 

[Helpful Tech Witch]

1 minute ago, ewitte said:

There are no DDR6X shortages NVIDIA has huge stockpiles.

 

but there are hella gou die shortages.

[ewitte]

30 minutes ago, HelpfulTechWizard said:

but there are hella gou die shortages.

All that points to on purpose but things may change soon...

[CTR640]

There are words on the streets nVidia prepared a new GPU with codename Ampérea, it should feature 9216 CUDA cores, 1987MHz and 32GB VRAM. Highly likely to be named 'RTX3080 Ti' or 'RTX3080 Super'.

 

Source:

Spoiler

 

 

[pas008]

2 hours ago, HelpfulTechWizard said:

but there are hella gou die shortages.

Leather jacket said it's a demand issue 

[Helpful Tech Witch]

28 minutes ago, pas008 said:

Leather jacket said it's a demand issue 

Lether jacket says a lot of things

[GoodBytes]

This is my analysis:

 

Samsung is unable to make 3080 with good yields, as expected (their strength is memory chips, not processors). This is clearly the case due to low avail of chips. In addition, leaks suggests that Nvidia has plenty of 3070's... so those are clearly uncertifiable 3080's, cut down for 3070's.

 

Initial dealing with TSMC didn't work out, so they had to go with Samsung. Rumors suggests that Nvidia is talking again with TSMC. So I expect that they cancel their plans of the 3080 20GB, to focus more on a 3080Ti 20GB model which is made with TSMC 7nm process. This will take time to adapt... so probably Q2 of next year you'll see it at the earliest.

 

The faulty chips that are working but don't meet the 3080Ti will be the 3070Ti. Nvidia can only hope that AMD GPUs aren't good and/or low quality drivers problem continues. People aren't going to wait that long for Nvidia. 

[Bombastinator]

9 minutes ago, GoodBytes said:

This is my analysis:

 

Samsung is unable to make 3080 with good yields, as expected (their strength is memory chips, not processors). This is clearly the case due to low avail of chips. In addition, leaks suggests that Nvidia has plenty of 3070's... so those are clearly uncertifiable 3080's, cut down for 3070's.

 

Initial dealing with TSMC didn't work out, so they had to go with Samsung. Rumors suggests that Nvidia is talking again with TSMC. So I expect that they cancel their plans of the 3080 20GB, to focus more on a 3080Ti 20GB model which is made with TSMC 7nm process. This will take time to adapt... so probably Q2 of next year you'll see it at the earliest.

 

The faulty chips that are working but don't meet the 3080Ti will be the 3070Ti. Nvidia can only hope that AMD GPUs aren't good and/or low quality drivers problem continues. People aren't going to wait that long for Nvidia. 

Assumes tu104 is a or can be a further cut tu102.  Not impossible.  Nvidia talking to TSMC could just be about manufacturing capacity opening up.  Nvidia is already making stuff with TSMC, namely their 101 dies which may be in great demand.  I don’t know.

[Serin]

If true, it'll be a real shame for the non-commercial vfx artists and beginners.