Micron confirms NVIDIA GeForce RTX 3090 gets 21Gbps GDDR6X memory

[illegalwater]

Both the press release and technical briefing documents on Micron's website have been taken down. Whoever accidentally released them early is likely getting their ass grilled right now as this was a colossal fuck up. Also gives more credence to the 3090 name being legit IMO as these were probably meant to be released along side the 30 Series announcement not weeks before. 

[Bombastinator]

9 minutes ago, illegalwater said:

Both the press release and technical briefing documents on Micron's website have been taken down. Whoever accidentally released them early is likely getting their ass grilled right now as this was a colossal fuck up. Also gives more credence to the 3090 name being legit IMO as these were probably meant to be released along side the 30 Series announcement not weeks before. 

Meh.  They were blurred to hell and gone.  Next to zero information released.   If someone screws up they didn’t screw up much. Here was massive evidence of image massaging.  There wouldn’t be any point to taking them down now.  Drop it, wait for comment which means the data has been scraped, then remove it becaus SHTHCHHH.! It a SEQUETT!  except there’s noting trustable there. 

[illegalwater]

11 minutes ago, Bombastinator said:

Meh.  They were blurred to hell and gone.  Next to zero information released.   If someone screws up they didn’t screw up much. Here was massive evidence of image massaging.  There wouldn’t be any point to taking them down now.  Drop it, wait for comment which means the data has been scraped, then remove it becaus SHTHCHHH.! It a SEQUETT!  except there’s noting trustable there. 

Blurred to hell? Next to zero information released? I’m sorry but did you actually read either document? 
 

Also both documents were hosted on Micron’s official website it doesn’t get any more trustworthy than that. 

[HenrySalayne]

So basically the 3090 is effectively replacing the xx80 Ti and it gets a pathetic 1 GB VRAM upgrade compared to a 3.5 year old 1080 Ti.
It's not blurry, just a little bit disappointing and some may hope it's fake news.

[Bombastinator]

46 minutes ago, illegalwater said:

Blurred to hell? Next to zero information released? I’m sorry but did you actually read either document? 
 

Also both documents were hosted on Micron’s official website it doesn’t get any more trustworthy than that. 

I was looking at the images from which the documents were apparently derived.  People can say all kinds of stuff.

 

re micron site

 This is exactly my point.  
Micron posts some pics on their site with very little data in them.  Very heavily censored.  Not zero but very vey little.  The something is taken down after the information has been part of the general information data base.  So the taking down is an action without use.  Closing the gate after the horse has escaped. 

what do we know? Someone censored information from some pictures, including possibly apparently (if you look at the photos very closely) the actual memory dies.  The only part micron would be involved in. The only new information is the number but not type of chips on the board and some clue in what way NVlink is going to change.  That NVlink would absolutely have to change has been known for months.  It’s a factor of hardware scheduling.  It would HAVE TO change.  Nvidia even officially announce they were changing it.  A while ago.  No one cares particularly.  Here is some evidence that they ARE  changing it but not specifics how. So no really useful info.  Now look at the other part of the “revelation”

 

the memory chips.  They appear to have had their names and codes inadequately removed.  Look closer though and there is a rectangle around each and every visible number set.  Could be a compression artifact.  Could be bad photoshop. There’s so little data there it’s hard to work with.  So the number of chips on the board but not what kind they are. Or rather a what kind they are that might or might not be true.  The interesting bit is the rest of the photo is very very heavily edited.  Might this be an “oops I slipped”? It’s not impossible but it’s not trustable either.  Micron given a photo censored originally by someone else of their own chips and then did a poor job of more censoring?  Doubtful.  For it to have any value it would have to be an accidental release.  Behavior is consistent with an accidental release, but there is no confirming evidence it actually WAS.  The number has been surmised for months, and the number of chips does fit with that count.  Assuming the chips are of the type they say they are.  But there’s no way to actually know that. 

It supports previous assumptions in a way that is to the advantage of Nvidia. It says “SEE!!  these particular assumptions that are in our favor are supported!” 

 

Think of it this way; say there is a person who is a known liar.  They lie a lot.  However, in a way you can translate what they say in a way where it is always the truth.  One can do so by adding a preamble: “it is to my advantage that you believe...” this is almost always true even if the statement is a lie.  It is also generally rue for people who don’t lie. The difference is the person not who doesn’t like will keep silent more often as he has nothing true to say that would gain him anything.  
 

So is a release that says nothing about the truth or falsehood of anything except a set of data that supports a different number of chips.  The NVlink stuff confirms an already public statement by the manufacturer?! How is that useful? 
 

The original statement I replied to was that clearly the person who posted the stuff was in serious trouble.  I am not convinced of that.  Very very little real information was released.  So little that I question both how much trouble such a person would actually be in if it was an accidental release and point out that it might not even be one. 

[porina]

8 hours ago, MageTank said:

At this point, I've given up on caring about the performance numbers, I just want a card as fast as my current RTX 2080 Ti with HDMI 2.1 support. Oddly enough, that is the one specification that has yet to be confirmed in any of these leaks, and it's the only one I care about.

Similar boat, although I'm more open on performance levels since I don't have a 2080Ti level GPU at the moment. Guess we both want that 4k60+? I never followed up with you on the manual settings.

 

39 minutes ago, Bombastinator said:

So the taking down is an action without use.  Closing the gate after the horse has escaped. 

They can't take back the info that is out there, but they may still have a moral and/or legal obligation to remove that information. Now if the document isn't there, anyone trying to verify the source of that document will fail to do so, and it increases the uncertainty for them. I happened to click the link when it was posted earlier, and still have a copy in my download folder. 

[Bombastinator]

6 minutes ago, porina said:

Similar boat, although I'm more open on performance levels since I don't have a 2080Ti level GPU at the moment. Guess we both want that 4k60+? I never followed up with you on the manual settings.

 

They can't take back the info that is out there, but they may still have a moral and/or legal obligation to remove that information. Now if the document isn't there, anyone trying to verify the source of that document will fail to do so, and it increases the uncertainty for them. I happened to click the link when it was posted earlier, and still have a copy in my download folder. 

So did techpwerup.com apparently. And who knows how many others completely ignoring the wayback  machine.  Point stands.

[illegalwater]

47 minutes ago, Bombastinator said:

I was looking at the images from which the documents were apparently derived.  People can say all kinds of stuff.

 

re micron site

 This is exactly my point.  
Micron posts some pics on their site with very little data in them.  Very heavily censored.  Not zero but very vey little.  The something is taken down after the information has been part of the general information data base.  So the taking down is an action without use.  Closing the gate after the horse has escaped. 

what do we know? Someone censored information from some pictures, including possibly apparently (if you look at the photos very closely) the actual memory dies.  The only part micron would be involved in. The only new information is the number but not type of chips on the board and some clue in what way NVlink is going to change.  That NVlink would absolutely have to change has been known for months.  It’s a factor of hardware scheduling.  It would HAVE TO change.  Nvidia even officially announce they were changing it.  A while ago.  No one cares particularly.  Here is some evidence that they ARE  changing it but not specifics how. So no really useful info.  Now look at the other part of the “revelation”

 

the memory chips.  They appear to have had their names and codes inadequately removed.  Look closer though and there is a rectangle around each and every visible number set.  Could be a compression artifact.  Could be bad photoshop. There’s so little data there it’s hard to work with.  So the number of chips on the board but not what kind they are. Or rather a what kind they are that might or might not be true.  The interesting bit is the rest of the photo is very very heavily edited.  Might this be an “oops I slipped”? It’s not impossible but it’s not trustable either.  Micron given a photo censored originally by someone else of their own chips and then did a poor job of more censoring?  Doubtful.  For it to have any value it would have to be an accidental release.  Behavior is consistent with an accidental release, but there is no confirming evidence it actually WAS.  The number has been surmised for months, and the number of chips does fit with that count.  Assuming the chips are of the type they say they are.  But there’s no way to actually know that. 

It supports previous assumptions in a way that is to the advantage of Nvidia. It says “SEE!!  these particular assumptions that are in our favor are supported!” 

 

Think of it this way; say there is a person who is a known liar.  They lie a lot.  However, in a way you can translate what they say in a way where it is always the truth.  One can do so by adding a preamble: “it is to my advantage that you believe...” this is almost always true even if the statement is a lie.  It is also generally rue for people who don’t lie. The difference is the person not who doesn’t like will keep silent more often as he has nothing true to say that would gain him anything.  
 

So is a release that says nothing about the truth or falsehood of anything except a set of data that supports a different number of chips.  The NVlink stuff confirms an already public statement by the manufacturer?! How is that useful? 
 

The original statement I replied to was that clearly the person who posted the stuff was in serious trouble.  I am not convinced of that.  Very very little real information was released.  So little that I question both how much trouble such a person would actually be in if it was an accidental release and point out that it might not even be one. 

What photos are you talking about? I'm talking about the PDFs that were hosted on Micron's official website, including the technical brief which by no means contained only "zero" or "very little" information, they described in a decent amount of detail how they achieved the increased speeds of GDDR6X over regular GDDR6.

 

Did they contain little information about the 3090 specifically? Yes. But what info was there (the name and VRAM specs) was something I seriously doubt Nvidia wanted leaked three weeks before the unveil, especially by one of their closest business partners.

[Bombastinator]

14 minutes ago, illegalwater said:

What photos are you talking about? I'm talking about the PDFs that were hosted on Micron's official website, including the technical brief which by no means contained only "zero" or "very little" information, they described in a decent amount of detail how they achieved the increased speeds of GDDR6X over regular GDDR6.

 

Did they contain little information about the 3090 specifically? Yes. But what info was there (the name and VRAM specs) was something I seriously doubt Nvidia wanted leaked three weeks before the unveil, especially by one of their closest business partners.

I’m talking about the 3 photos and the tech power up article about them the thread references.

[soldier_ph]

So is the 3090 supposed to be the replacement of the Titan Card or is it the new Highest End "Consumer" GPU ?

[naturallyaspirated6]

11 hours ago, HenrySalayne said:

So basically the 3090 is effectively replacing the xx80 Ti and it gets a pathetic 1 GB VRAM upgrade compared to a 3.5 year old 1080 Ti.
It's not blurry, just a little bit disappointing and some may hope it's fake news.

 

I recently watched a YouTube video on this very leak (might be GamerMeld, not sure) - he showed a screenshot of the Micron leak before it was removed from the website and both the 3090 and Titan RTX were listed with 12 GB VRAM - and we know the Titan has 24.

[porina]

50 minutes ago, naturallyaspirated6 said:

I recently watched a YouTube video on this very leak (might be GamerMeld, not sure) - he showed a screenshot of the Micron leak before it was removed from the website and both the 3090 and Titan RTX were listed with 12 GB VRAM - and we know the Titan has 24.

I still have a copy of the document locally and had another look. Actually in that column both it lists both Titan RTX and RX5700 XT as 12GB. How can this be? 

 

Thinking more, where it names the product, it says "Application Type (Example)". And for the VRAM it says "Frame Buffer of Typical System". On further thought, maybe they are using the product names as an example of products using that technology, and a typical configuration, but not necessarily the actual configuration. If so, this would explain the apparent error that would otherwise be present, and also brings some life back into the higher VRAM rumours still floating around.

[KaitouX]

9 minutes ago, porina said:

Thinking more, where it names the product, it says "Application Type (Example)". And for the VRAM it says "Frame Buffer of Typical System". On further thought, maybe they are using the product names as an example of products using that technology, and a typical configuration, but not necessarily the actual configuration. If so, this would explain the apparent error that would otherwise be present, and also brings some life back into the higher VRAM rumours still floating around.

This is probably correct, it wouldn't make much sense for Micron to try to find GPUs with the exact memory configuration when the objective is to give a example on where that memory was/can be used, also because probably there isn't always a GPU that maxes out the possible memory configuration in every way.

[naturallyaspirated6]

1 hour ago, porina said:

I still have a copy of the document locally and had another look. Actually in that column both it lists both Titan RTX and RX5700 XT as 12GB. How can this be? 

 

Thinking more, where it names the product, it says "Application Type (Example)". And for the VRAM it says "Frame Buffer of Typical System". On further thought, maybe they are using the product names as an example of products using that technology, and a typical configuration, but not necessarily the actual configuration. If so, this would explain the apparent error that would otherwise be present, and also brings some life back into the higher VRAM rumours still floating around.

Interesting thought. I think this does actually make a lot more sense, thanks for typing it out this way. Unfortunately I wasn‘t fast enough for a screengrab of said table. 

 

Either way, I haven‘t given up on the dream of a high VrAM GPU Generation. I‘m still wondering if the 3090 is a Titan replacement and comes with the according price tag or not, though. 

 

Hopefully me we can a lot more leaks in the coming weeks and even if not: the official reveal isn‘t that far out. 

[HenrySalayne]

3 hours ago, naturallyaspirated6 said:

I recently watched a YouTube video on this very leak (might be GamerMeld, not sure) - he showed a screenshot of the Micron leak before it was removed from the website and both the 3090 and Titan RTX were listed with 12 GB VRAM - and we know the Titan has 24.

You are right.

2 hours ago, porina said:

Thinking more, where it names the product, it says "Application Type (Example)". And for the VRAM it says "Frame Buffer of Typical System". On further thought, maybe they are using the product names as an example of products using that technology, and a typical configuration, but not necessarily the actual configuration. If so, this would explain the apparent error that would otherwise be present, and also brings some life back into the higher VRAM rumours still floating around.

Good point. But I assume we'll be seeing VRAM configurations somewhat similar to the previous generation. They likely keep the higher VRAM configurations for their "overpriced" top-of-the-line series (two times the VRAM = two times the price).
We've seen in the past some low tier graphics cards beeing shipped with either x memory or 2x memory. I've heard some rumors saying the MSRP of the 30 series may be a bit lower compared to the 20 series. So maybe, just maybe, we will see a return of different VRAM configurations for the same chip as previous leaks have mentioned.

[porina]

1 hour ago, naturallyaspirated6 said:

Unfortunately I wasn‘t fast enough for a screengrab of said table. 

Part of it is included in 1st post of this thread.

[Results45]

Found some interesting insights into what companies consider when picking between GDDR6 and HBM2: https://semiengineering.com/hbm2-vs-gddr6-tradeoffs-in-dram/

[Results45]

3080 just surfaced at 2.1Ghz: https://videocardz.com/newz/nvidia-geforce-rtx-3080-spotted-with-2-1-ghz-gpu-clock