Nvidia Ampere (2080 Ti, etc. replacements) and other rumors...
tj_1ca9500b
Posts: 2,047
So this was posted up:
| Board ID | Replacement Class | VRAM | Bus Width | Launch Schedule |
| PG132-10 | 2080 Ti | 24 GB | 384 bit | 2H September |
| PG132-20 | 2080 Super | 20 GB | 320 bit | Mid. Sept |
| PG132-30 | 2080 Super | 10 GB | 320 bit | Mid. Sept |
| PG142-0 | 2070 Super | 16 GB | TBD | TBD |
| PG142-10 | 2070 Super | 8 GB | 256 bit | 2H Sept |
| PG190-10 | 2060 Super | 8 GB | 256 bit | TBD |
My first reaaction r.e. the 20 and 24 GB cards is 'We GPU rendering types could only be so lucky', so looking around for other sources:
https://www.tomshardware.com/news/nvidia-rtx-3080-ampere-all-we-know
The Toms Hardware article has slightly different numbers on the VRAM, and mentions the 3090? with 24GB of VRAM and a launch price of $1500??
On the higher end, here's an Octane Bench result of the A100 (6912 Cuda cores, 40 GB HBM2): It looks like it'll set you back over 10 Grand though...
https://videocardz.com/newz/nvidia-a100-ampere-benchmarked-the-fastest-gpu-ever-recorded
Back to the 2080 Ti replacement though. It'd be nice if that 24 GB rumor holds true, I guess we'll see. 11 GB is nice, but I"m gettng tired of hitting the 'CPU only wall' but am too cheap to shell out over $2000 for a grpahics card ATM. There's older cards of course on the Quadro end with more VRAM maybe, if you don't mind Ebay...
The next big question is how long will it take the Daz software folks to enable the new cards in Daz Studio, once they launch, but one thing at a time.
Anyways, I'll post up more stuff as I find it. Feel free to share your own links. It's 'Rumor/Huge Grain of Salt' until the launch actually happens of course. I've heard September 9th as a possible launch announcement date.
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
I hope this is true, because I've been saving up for a Titan RTX, but just moments ago ordered a 2060 Super to tide me over. A $1500 24GB card would be great news.
That scaling is ridiculous and is exactly what I expect from wccftech. I suggest you look at their history of accuracy with previous "leaks."
Legit sources say Nvidia is likely to announce the initial wave of releases on Sept. 9 (that's coming from GN and other sources that do not ever post click bait nonsense) based on that you'd expect pre orders to open in a week or so after and retail to be likely either the next week or the very end of Sept. or Oct. 1ish. So just wait.
My job requires me to monitor RAM prices and the odds still are that 24 Gb of GDDR6 is not going on a $1200 to $1500 card (that would be something like $300 to $350 raw material cost to the board partners).
Yeh that's what I'm worrying about too. Not so much the DAZ people than the Iray people. One would think there would be conversations going on prior to Ampere's release so that Iray compatability would be ready for the release of Ampere.
So we don't have to wait months to get these new cards to work with Iray in DS. I don't know how these things work but that would be nice.
Hence my skepticism, although remember that AMD was selling a card with 16 GB (which had the more expensive HBM2 no less) for around $700-$800 at one point...
Actually, WCCFTech has a pretty good track record with their rumors these days. You may be thinking of the older WCCFTech. Even their 'Lisa Su may be grooming a successor' rumor is several years old at this point, and they do mark rumors as such. Dr. Su did end up getting a retention bonus package from the board of directors around that time worth about 25x her salary at that point, so the timing was interesting...
I've seen that 24GB number for the high end card from a few sources now (note that I cited Tom's as well, which predates the WCCFTech article), and we have about a month to go until the supposed launch date, which of course can always be changed.
Big Navi (which isn't much help to Daz users, unless you use say Blender for your rendering), is purported to have 16GB of HBM2, and is aimed squarely at the 'Ti' lineup, which would give Nvidia stronger reason to increase the VRAM. That and hardware based raytracing...
No question that Nvidia will hike the prices of the cards, mainly because they can.
The WCCFTech article is marked as an exclusive, though, not a rumor, which indicates that they have strong faith in their sources, so how those board IDs translate over to actual product names (if accurate) could be interesting. That high end card could be a 'Titan' class card of course, not a 2080 Ti class card, so I guess we'll see soon enough.
The last big exclusive I remember from WCCFTech last year involved Renoir, which as stated in that exclusive (which was published 3 months before the official announcemnt) turned out to have 6-8 cores in the $700 range (entry level models) with around 12 hours of battery life during lighter use. At the time that exclusive was posted, people were skeptical that Renoir would have more than 4 cores, and based on Picasso, yeah 12 hour battery life didn't sound right. 6-8 hours maybe... And $700? Maybe if it was more like Picasso, only faster... And of course no one predicted that Renoir would be 'on par' with the latest desktop chips that have similar core counts (Both on the AMD and Intel side). It's a pretty incredible little chip which has certainly changed AMD's fortunes in the laptop and mini-PC space...
But this thread is about the latest Ampere scuttlebutt, hence the 'Rumors' tag. Us Dazzians want more VRAM, and better, more affordable options for our graphics cards!
I'm trying to remember how long it took to get the RTX cards working smoothly in the Daz betas. Official 'non-beta' support didn't come until the public release of 4.12.something as I remember, but others can confirm how long it actually took. I remember passing on an RTX card in favor of the 1080 Ti 18 months ago because RTX support in Daz was still a bit shaky/buggy in the DS beta version, and that was four months after the official launch.
AMD is done with HBM2. That stuff was a major part of the high cost and low availability of their last gen of cards, literally everyone in teh industry knows that.
wccftech can label their rumor anything they want. It's a rumor. And based on that chart it isn't even a slide from Nvidia just some stuff they made up or copied down from a phone call. When someone has a slide from a powerpoint then the data makes sense.
That's not what Tom's HW said about Big Navi & HBM2 for some GPUs. They also said the Ray Tracing competive with nVidia was likely in these Big Navi GPUs and they'd also due to release in Sep 2020.
The ProRenderer and the GPU Compute of Cycles in Blender 2.8.x isn't working with my MSI Radeon RX 570 8GB so that's quite lackadaisical regression and test case planning on AMD's part. I was really disappointed.
It's rumours, speculation and guess work.
You'll know what they are once released and results and specs are posted.
Some of what you read will be close to the actual specs, and even very close - but that doesn't mean they had reliable information, only that when you get half the planet guessing - some are going to be right.
"Likely", is an incredibly important word in this statement; they are 'speculating' - speculating means basing guesswork on some data; this is different than pure guesswork.
Yeah and that's why no one pays any attention to Tom's anymore. Middle performance HBM2 costs what top performance GDDR6 costs. Plus you have to add an interposer, which is way more expensive than just laying down circuit paths on a PCB. And I have never seen anything saying that Hynix or Samsung got their yields up. So HBM will stay on phones, and Macbook Pro's.
Well, I guess we'l have to wait & see. I know in a consumer card like I'd buy it'd have GDDR6 RAM. I'm actually more interesting in the ray tracing though if I don't get a response from AMD regarding a question I asked them regarding the Pro Renderer locking up Blender for some Blender Foundation example scenes I will probably buy an nVidia 30XX GPU instead and periodically check Pro Renderer progress with my old AMD card.
I'm surprised this hasn't been posted. Nvidia is announcing *something* on August 31.
https://www.engadget.com/nvidia-geforce-graphics-cards-tease-153636073.html
As far as what they will do, I frankly wouldn't put anything past them. All of those VRAM numbers listed in the first post? I believe they do have designs for ALL of them, so that they can get them out at a moment's notice if need be. The real question is which ones will they actually bring to the market. The answer to that depends on AMD, not you or even Nvidia. If AMD releases the big GPU that people have been hyped forever for and it has 16GB of VRAM, then I do believe Nvidia will move to top that, especially if performance is close.
Nvidia is extremely competitive when they want to be. The more competitive AMD is, the more Nvidia will push in turn. Simple as that. I think Nvidia is planning on launching the smaller VRAM SKUs, and they will hold on the others until they think they know what AMD is doing. These guys are watching leaks just like we are, and if they feel strongly enough, they will react. So if they hear from their sources that AMD is truly doing 16GB, they will do it before AMD even has a chance.
I think they also have the rumored 3090 on the table as well. Probably not for launch, though. I think the 3090 is a card designed purely as a counter punch to AMD if they somehow top the 3080ti. Again, if they feel that AMD can do this, they may just go ahead and announce it, too, even if it is not ready yet. The point being they do not want AMD to have the performance crown for even a short time period, it is that important to them, at least to their CEO.
That is why they are doing this so soon. They feel that AMD will beat the 2080ti, so because they do not want AMD to hold the performance crown they are pushing to launch first. I told you guys they would push as fast as they could, COVID and stock be damned. I even said as early as August, probably September...and here we are looking at an announcement on the last day of August, with the launch to come soon after.
Good times are coming! It is about to hit over drive in the hardware market.
Oh, and on the subject of Iray, I believe Ampere will work on day 1 with Daz and Iray. I mentioned this before, but this is thanks to the new Iray RTX, which is based on OptiX 6.0. Unlike the old Iray that was based on OptiX Prime, OptiX 6 does not need to be recompiled for every new hardware launch. Meaning it should work with just a simple GPU driver update. Down the road Iray may get updated to take better advantage of Ampere. There will be no more months long waiting periods to get Iray working on new GPUs.
No, it does not depend on AMD.
They don't just decide on the spur of the moment what to manufacture. They do all the cards, except the Titans, through partners who have to receive the references cards and then design their own cards. So they cannot change specs at the last minute. The cards are already being manufactured by the AIB's. They have been hammered by the AIB's over Jensen Huang apparently changing MSRP's of the base models back stage before the announcement.
If a card is to launch at the end of September then it is being made right now. They have to get to docks in China and get onto container ships by early September to reach NA and Europe on time. And they always have a limited roll out of just the top 2 or 3 SKU's.
Also there is as of yet nothing on the Nvidia website nor on their press channel. It may be that is just them hyping the anniversary of there first GeForce card coming up. The reliable sources still say the launch announcement of Ampere is Sep. 9 which will be during their online SIGGRAPH 2020 stuff and they have done GPU announcements at SIGGRAPH in the past.
This is a well thought out and presented opinion; it makes sense to me what you've written.
I do feel though that they won't really go with 'spur-of-the-moment'; they will certainly react to what AMD do, as I seriously expect they took notice of what complacency did to Intel.
It looks like Nvidia is doing an "Ultimate Countdown" tweet each day on their Twitter page at: https://twitter.com/NVIDIAGeForce
The new Nvidia header banner posted yesterday at their twitter link, is the 21-days till their new announcement on August 31st, 2020. The tweet earlier today had an old-school modem tone. IMO, that may represent the slow speed of their current GPUs, compared to what's about to be announced.
As @outrider42 said, "Good times are coming!"
I just got an email from nVidia and the tone of the eMail sounds like what they are announcing is much bigger than what we've been led to believe will be the new 30XX Ampere Architecture GPUs performance capabilities.
You Choose:
a) Has the capabilities of the 30XX line been severely underestimated?
b) Is nVidia got something up their sleeve unannounced that's even better?
c) Huge price cuts to the prices on their GPU lines, old & new?
d) Or has nVidia marketing gone completely overboard?
e) Fill in something different you think it might be.
f) Nvidia always sends out OTT emails for everything.
...I'll believe it when I actually see it on the Nvidia site. I still remember all the hype years ago about the 980 Ti getting 8 GB.
Also one of the advantages of the Titan, like the Quadro series, is if it is dedicated to rendering, you can switch to TCC (Tesla Compute Cluster) mode which bypasses Windows WDDM.
Looking at the table in the OP, I'm encouraged to see a 2070 with 16GB VRAM. I really hope that's true.
Do you really believe that Nvidia designed everything in a vacuum, with no concern at all for competition? There are people who work for these companies who will outright shoot that idea down. They spy on each other all the time, and you will find plenty of people in that industry who will affirm that. Jenson in particular is known to be extremely competitive. He is largely the reason behind that drive. And if you will notice something, both AMD and Nvidia have not provided their board partners a ton of information this time around, that's why big leaks have been so few. They are playing it close to the vest. AMD is playing it so close that some are saying they might keep AIBs out of loop at launch entirely, and that is because of their tendency to leak everything.
You also grossly underestimate how this works. There are people who will tell you that they have literally changed things in the presentations the actual day of the presentation, and not just pricing. Dude, they could have GPUs with 12 GB of VRAM in production, but software lock them to 10 for reviewers. If AMD strikes with something, they can then unlock those cards to 12. They wont manufacture them all that way, just the first batch. Maybe you think this is absurd, but do remember that AMD actually shipped GPUs with 8GB of VRAM that were software locked to 4. That sounds crazy, but it is true, their reason were different, but it has been done before.
They could have units of all of the cards all set to go, with enough to cover reviewers. They still have time to go into mass manufacturing for what they prefer.
The same holds true here. Nvidia didn't release the 980ti with 8GB because they didn't need to. AMD did not offer 8, nor did AMD even have a GPU capable of matching the 980ti, so why would Nvidia bother? That would have cost them more, when they already had the performance crown in the bag. I assure you, they absolutely had a 8GB version ready if they wanted it. But like I said...they didn't need it.
So just as back then as it right now, it all hinges on AMD. There is a lot of talk about AMD using 16GB in their top cards, so Nvidia is extremely wary of that. But if AMD's cards are still much slower, what's the point? Nvidia is trying their very best to get a needle on just what AMD's performance will be, and then they can decide on the exact configurations they will go with. If they feel threatened, then yes, they will move forward and release the larger variants...at a premium price, of course.
There is also that one last factor, the new gaming consoles. Both PS5 and Series are getting 16GB of VRAM. Now while 2.5 is taken by the OS, that still leaves a hefty 13.5 GB of VRAM for games. Also, the PS5 in particular is actually going to use its SSD in place of DDR system RAM. That is not speculation, Mark Cerny straight up described it that way during their 'deep dive' a while back. The PS5 SSD is faster than DDR2 RAM, and it is able to access this data very quickly. So when I say the PS5 will have 13.5 GB of VRAM, it is true VRAM, not a combination of VRAM and RAM. The Series X SSD is not quite as fast, but still faster than nearly any out today, and it will also be able to access the SSD much faster than PCs do. So they will also be using their SSD as virtual memory, according to their head Phil Spencer. So the Series X will also be doing this to a degree. Basically both consoles will be able to access data better than any modern PC, and they will have larger pools of VRAM than almost any consumer GPU. This has to aggravate Jenson at least a bit, especially when you consider than Series X and PS5 are AMD based products. Any customer who decides to buy a console instead of a PC is a lost customer for Nvidia. So Nvidia is very much in competition with consoles as well.
Nvidia not only has to beat AMD, they have to convince gamers to buy their GPUs instead of a console. The Series X is touted as a 2080 Super for crying out loud. A 2080 Super with 13.5 GB of VRAM, and it might be $500! Nvidia has to full on ROFL stomp the consoles in performance to have a viable argument. That is yet another reason why Ampere needs to be powerful and probably at least more aggressive on price than Turing. Because if it is not, well, people will just buy consoles instead.
So yeah...competition is coming from all sides for Nvidia. They have plenty of reason to go more all out than before.
....I remember when AMD introduced a version of their Vega GPU with 16 GB of HBM memory a couple years ago. Looked to be a game breaker but instead ran extremely hot at peak output and no water cooling solution was available.
I can see 16 GB being more the top end for a gaming card, 24 GB possibly in the 2090 if that is released, but then you are heading into "Prosumer" territory (possibly replacing the Titan series[?]) as the only users who will benefit from such a leap in VRAM are basically us, and those dabbling in deep learning.
AMD is also not anywhere near release. They are the ones who will respond to Ampere. Nvidia may then respond, as they did with the Super line. But these initial cards will not be a response to some mythical big Navi which if it even happens is still at least a couple of months away.
What we know, and we know it for a fact, is Nvidia wants a flagship release on shelves by Black Friday. They like that buzz in the press and they like that bump in their sales volume. They certainly are not going to hold up a release waiting for AMD. The question that remains unanswered is will AMD push back Navi 2, which is what they say is coming this year not "big Navi" which they have never even said is a thing, to get Ryzen 4000 on shelves by Black Friday as well as this all goes through TSMC's 7nm process and there are only so many wafers. Personally I'm betting we see Ryzen 4000 before any new Radeon cards because that is where AMD is making money right now.
And no they could not release cards with 12Gb of VRAM and software lock it to 10. People would lose their shit. Those GDDR chips can be counted, each one has the capacity right on it (1,2 or 4Gb). The guys who buy the first cards on the market are the ones most likely to WC them. They'd get the coolers off and see 12Gb instead of 10 and it would a huge scandal. Further those things are expensive, a 1Gb GDDR6 chips is right around $12US cost to the manufacturer so putting two of those on that you then turn off is adding a lot ofcost that they can't pass on to consumers unless they want to get sued when the truth comes out.
No, the PS5 is not going to use an SSD in place of RAM. That marketing speak is really really bad. They're claiming the SSD has a very specific custom firmware meant to integrate with the APU such that it is very efficient at fecthing assets. The APU will still have regular RAM and will still use it use as regular RAM and as VRAM. Of course the SSD will be used as virtual RAM. That's standard on every OS for the last 20+ years. That's not some new thing. Your Pc does it and so has every one since Win95 (or maybe Win 3.1 I forget when it was added).
No, Nvidia is not worried at all about the new consoles. The announced graphics specs are beat by Turing cards right now. I think people are getting caught up by some outlets saying the PS5 will support 4k 120fps. that's not what the games will run at. That's just the media describing HDMI 2.1. The actual GPU is described in terms of GPGPU performance and it is nothing all that special (10.28 teraflops) which is roughly a 2080 (10.1). So assuming Nvidia does their usual generational step the 3070 will beat the PS5.
Actually, if you backed off on the voltages on this card a bit, the heat and associated fan noise was quite manageable, and you didn't lose much performance. AMD has been pushing their high end retail cards to their limits lately to eek out a tiny bit extra FPS performance vs. Nvidia in benchmarks, hence why backing off on the voltages a bit can actually give you a very pleasant experience.
And water cooling solutions for this card did show up eventually. Of course, good luck finding a Radeon VII these days. As a comparison, the current Radeon Pro 16 GB card is around $1500 these days, and the Instinct card is around $1800.
I did find a Vega Frontier card just now, which has a radiator and 16GB of VRAM for around $900:
https://www.amazon.com/AMD-Radeon-Frontier-Liquid-Retail/dp/B072XLR2K7
Not sure how that card stacks up these days...
...well considering the major GPU based render engines on the market are pretty much all CUDA based. that puts AMD at a disadvantage for rendering. I recall that Otoy was considering supporting AMD GPUs for the 2020 release of Octane4 but they abandoned the idea.
As I am not into gaming, AMD GPU development is more of a pasing interest for me.
@kyoto kid
Same here. Blender does support AMD GPUs at least, but since I use Daz Studio Iray for my rendering needs...
AMD pushing the value proposition does help us a tiny bit with the Nividia situation though. Since AMD and Nvidia are going toe to toe in the mid and low range segments, at least those prices have stayed somewhat reasonable. Just waiting for AMD to bring more competition to the high end. Right now it's anyone's guess how Big Navi will stack up against the Ampere 3000 series cards...
AMD can't change their design now, but the same is true for NVidia. The type of strategic interaction @outrider42 mentioned surely happened, and worse, just a few months ago. We are lay people and we have no idea what either is doing, but rest assured that both companies or their bankers have entire competetive intelligence divisions full of ex intelligence community officers, and know pretty well what each other are doing.
If you think that the two largest players in a billion dollar industry and their bankers do not treat it like warfare, some good books to read are Confidential by John Nolan and CIA, Inc by F. W. Rustmann and the Society of Competetive Intelligence Professionals at scip.org which is full, at least when I was a member, of ex Defense Intelligence Agency personnel who were looking for areas to apply their unique skills after the defense downsizing that happened after Bush Sr's "peace dividend".
Even the idea of "leaks" that weren't created by the companies themselves is a bit hard to believe. On the other hand, they would be explained perfectly by the Joint Publication on Military Deception, JP 3-13.4, which says that if the enemy doesn't have a "deception channel" through which you can provide him disinformation, you are obligted to provide him one. And in business, unlike warfare, a company listed on the stock exchange lying about upcoming products will get you into trouble with the SEC, but not if it's a "leak" that the company is ostensibly not responsible for. I would not be surprised if the leakers themselves were not competetive intellgence personnel.
It's marketing.
It's purpose is to sell a product; facts are irrelavent
Nothing really new here, just the latest leak essentially confirming the leaked info we've heard so far:
https://videocardz.com/newz/nvidia-geforce-rtx-30-ampere-ga102-series-rumored-to-launch-in-24gb-20gb-and-10gb-variants
I know its all speculation, but I'd be surprised if they make it so there is so little between the 3080Ti and the Titan.. it will have to be enough to justify the extra $1000 or whatever it will be... so I think there will be a bigger gulf in the RAM.. I can't even see 16gb on a 3080Ti..