AMD, Intel and Nvidia driver issues and last recommended version [Archive] - Page 38

Asmodian

25th September 2020, 15:02

Posts from nVidiatel fanatics.

Why do you assume people have to be fanatics if they disagree with you? (Projection?) I really do not think of any of these companies as entities to be fans of. The idea seems weird to me, billion dollar corporate entities are not something that I can be a fan of.

I am buying a 3090 asap, but for no good reason. A 3080 is a much better buy for my use cases. If big Navi turns out to be better than the 3090 I will buy one.

Anyone without an AMD tattoo seems anti-AMD to you. :rolleyes:

Manni

25th September 2020, 15:24

Where is the like button when you need it :)

Klaus1189

25th September 2020, 15:59

https://www.youtube.com/watch?v=7YQ7rNgoqMA

videoh

25th September 2020, 16:43

If you are going to link to Germish stuff, how about an executive summary in English? The language for this forum is English.

Klaus1189

25th September 2020, 16:53

See in the description of the video:
https://www.igorslab.de/en/what-real-what-can-be-investigative-within-the-crashes-and-instabilities-of-the-force-rtx-3080-andrtx-3090/

videoh

25th September 2020, 17:06

Thank you, Klaus.

SamuriHL

26th September 2020, 03:52

Asmodian

26th September 2020, 05:47

I think companies simply use release as a QA step today. In house QA at that level would be expensive, digital updating is cheap, and everyone who bought one can do it so the PR isn't too bad. Everyone who can does it now. :p

ryrynz

26th September 2020, 08:10

This launch feels rushed even though it shouldn't have been. We know why that is

Yeah Nvidia was scared of big Navi [emoji23]

huhn

26th September 2020, 09:13

huhn you asked why I'd even consider an FE over an AIB. Given the latest reports out there, hopefully it's painfully obvious now. The AIBs, at least some of them, it looks like they cheaped out on capacitors and the crash to desktop problems we're hearing about MAY be caused by voltage problems on those cards when they boost. nVidia used the good capacitors. ASUS did, as well, which is the only other card I was even looking at. And...I think I'll stick to that. LOL Drivers are also not super awesome for these cards combined with the problems you've found about power settings sticking. This launch feels rushed even though it shouldn't have been. We know why that is, but, nVidia really didn't have any room for a screw up here. Hopefully these issues get sorted out.

yes i still do the 3080 FE is part of this problem and the 3080 FE is so bad that it will thermal throttle with let's say furmark that's not the case with a decent AIB.

there where AIB so terrible that they would die if you flash a stock bios on them and run furmark and just to make that absolutely clear i said die not maybe die.

clock rate are currently taking into account why the card is crashing. AIB cards often reach higher boost clock then FE cards just by been massively cooler with the same bios it is as it is.

so just lay back wait for some test it's sadly not unusual for stuff like this to happen. someone wants an exploding EVGA card even through the VRM are totally fine? there where and there will always be terrible AIB cards too.

nevcairiel

26th September 2020, 09:31

NikosD

26th September 2020, 09:31

as for Intel vs AMD, when AMD starts supporting Thunderbolt in more than a couple motherboards, I might consider it.

In the meantime, for video editing, especially with After Effects, a threadripper is a waste of money and Intel provides better performance at the same price point.
I really do not think of any of these companies as entities to be fans of. The idea seems weird to me, billion dollar corporate entities are not something that I can be a fan of.

I am buying a 3090 asap, but for no good reason. Thank you.

Manni

26th September 2020, 11:54

Thank you.

You're welcome :)

Manni

26th September 2020, 12:22

As nobody seems to mind the off-topic posts, an interesting review of the 3090 (no idea if it can be trusted) seems to show that, as expected, it's more efficient compared to a 3080 (2-4%), 2080ti, Titan RTX and RX5700XT (20-33%!), despite its higher TDP: https://www.eurogamer.net/articles/digitalfoundry-2020-nvidia-geforce-rtx-3090-review

****

Watts Per Frame RTX 3090 FE RTX 3080 FE RTX 2080 Ti FE Titan RTX RX 5700XT
Death Stranding 3.209 3.349 3.562 3.670 3.846
Percentage Diff 100% 104.4% 111.0% 114.4% 119.9%
Watts Per Frame RTX 3090 FE RTX 3080 FE RTX 2080 Ti FE Titan RTX RX 5700XT
Gears 5 4.055 4.156 4.266 4.291 5.403
Percentage Diff 100% 102.5% 105.2% 105.8% 133.2%
"The results here are fascinating. Despite sporting a higher TDP, the 3090 actually consumes fewer watts to generate a single frame than the 3080 in the titles we tested: In Death Stranding, the RTX 3080 requires four per cent more power per frame, and this drops to just under three per cent in Gears 5. This may sound implausible bearing in mind the beastly nature of the RTX 3090 - but the new card's total graphics power is 12.5 per cent higher overall and it is capable of a performance bump in excess of that number."

****

If confirmed, this should be the case for madVR too: with the same settings, the 3090 should be more efficient (though a 3080 or lower remains a more sensible choice for most use cases obviously). Now of course if you pump everything up to the max, it won't be :)

I'm interested to see what I can get with 4K60p HDR with the 3090/3080, and even more so when we'll be able to send this in 4:4:4 8 or 10/12bits to an HDMI 2.1 display / projector, as well as what it can do with poor content such as upscaled SD/HD.

As some in this thread seem to have trouble understanding what I write, I'll state again that I don't advise to buy a 3090 for madVR or games. For madVR, a 3080 10Gb is a far better choice at the top end (anything lower according to your budget will do too). For high-end games, the upcoming 3080 20Gb might be a better long term investment than the 10GB version, as quite a few games already make use of more than 10Gb of RAM. If a 3080 20Gb was available to buy today, that's what I would get, not a 3090, as it would give me all I need for around 30% less money (guesstimate). That's what I will be buying in November, when it's most likely released, if I'm not happy with the 3090.

If you prefer AMD for whatever reason, by all means buy an AMD GPU, I couldn't care less, but don't complain if later down the line madVR starts using tensor cores for new algos and you can't make use of that extra power!

NikosD

26th September 2020, 12:36

If you prefer AMD for whatever reason, by all means buy an AMD GPU, I couldn't care less, but don't complain if later down the line madVR starts using tensor cores for new algos and you can't make use of that extra power! It started as a joke but since you mention it for a second time, which is the API inside Geforce drivers that exposes tensor cores to be leveraged by a developer outside of DLSS for games ?

Manni

26th September 2020, 12:52

It started as a joke but since you mention it for a second time, which is the API inside Geforce drivers that exposes tensor cores to be leveraged by a developer outside of DLSS for games ?

I have no idea and I honestly don't care. Ask madshi or nVidia :).

The only thing I can tell you (again!) is that madshi has, over the last couple of years, publicly stated that he planned to make use of tensor cores in the future for madVR/Envy algos. I work as a consultant for madVR Labs, and as far as I know this is still his intention and it has been stated publicly a few times.

For this reason, he has informally advised anyone asking him the question in forums to buy RTX over GTX or AMD if they wanted to future-proof their purchase, though there is still no commitment at this stage that 1) he will make use of tensor cores in madVR/Envy and 2) that even if he does end up using tensor cores in Envy, this will translate to madVR. His focus is currently on finalizing Envy, not on new algos, so we won't know for a while.

However, what we know is that the GPUs in Envy are all RTX and will remain RTX in the foreseeable future. The GPU in the top of the range Envy Extreme will be a 3080 10Gb whenever stocks can be secured. As a result, there will clearly be more testing going on with RTX than with anything else. There are already some features (such as supporting the BT2020 flag to signal SDR BT2020 content for automatic calibration switching or HDR/SDR switching through API) that is only supported with nVidia GPUs.

These are facts. If you don't like them, there is nothing I can do. I am not posting here for you, I'm posting here for everyone else.

If you still want to buy an AMD GPU for madVR use, why should I care? It's your money, your choice, your decision.

I only provide information so that buyers make an informed decision. MadVR use is, for many, only one factor in their decision. They might have very good reasons to buy AMD, or even Intel.

As I said, we are all different, with different preferences, priorities and budget.

This is my last reply to you. I am not pro nVidia, pro Intel or anti-AMD. I don't care about these companies. I don't own shares in any of them. In this thread, I don't represent madVR Labs, only myself. I am not sharing any insider info, only what madshi has publicly stated in many occasions, or what has already been announced in press releases.

As a consumer, I am pro-me and I buy what suits my needs, my priorities, my budget and my personal preferences. I don't expect anyone to have the same as mine.

I only expect everyone with a brain to do the same, and make their decisions based on THEIR needs, priorities, budget and preferences :).

el Filou

26th September 2020, 12:59

It started as a joke but since you mention it for a second time, which is the API inside Geforce drivers that exposes tensor cores to be leveraged by a developer outside of DLSS for games ?Do you by any chance know about something called CUDA? It's an API provided by NVIDIA to develop natively for their hardware. (I'm sure you also know about something called Google that would give you the answer to the question you're desperately trying to troll with)

NikosD

26th September 2020, 13:02

I have no idea and I honestly don't care. Ask madshi or nVidia :).

For this reason, he has informally advised anyone asking him the question in forums to buy RTX over GTX or AMD if they wanted to future-proof their purchase, though there is still no commitment at this stage that 1) he will make use of tensor cores in madVR/Envy and 2) that even if he does end up using tensor cores in Envy, this will translate to madVR.

These are facts. If you don't like them, there is nothing I can do. I am not posting here for you, I'm posting here for everyone else.

I expect everyone with a brain to do the same :). Tensor Cores are out in the wild for three years now.
Do you know any software outside of DLSS and games leveraging Tensor Cores using a Geforce card ?
So, the fact is that madVR labs even in partnership with nVidia hasn't provided yet a single piece of code demonstrating that is feasible to use Tensor Cores for madVR.
Buy you are still expecting people with brain, as you said, after years of promises to buy RTX cards because of some future, nebulous use of Tensor Cores for madVR ?

OK...

Manni

26th September 2020, 13:33

Tensor Cores are out in the wild for three years now.
Do you know any software outside of DLSS and games leveraging Tensor Cores using a Geforce card ?
So, the fact is that madVR labs even in partnership with nVidia hasn't provided yet a single piece of code demonstrating that is feasible to use Tensor Cores for madVR.
Buy you are still expecting people with brain, as you said, after years of promises to buy RTX cards because of some future, nebulous use of Tensor Cores for madVR ?

OK...

Can you please stop putting words in people's mouths?

There is no madVR Labs / nVidia partnership.

madshi has not even STARTED to look into making use of tensor cores.

He has only advised, over the last couple of years, that it was his INTENTION to do so at some point, but never committed to do it.

There has been no work on new algos over the last couple of years, except improvements to HDR to SDR tonemapping. As I said, and as everyone knows, madshi's focus has been on releasing Envy, not on developing new algos with tensor cores...

I'm only making people aware of this INTENTION with no commitment, and you turn it into a FAILURE over THREE YEARS?

Do you realise how twisted your post is? Otherwise, you really need professional help.

Whatever nonsense you decide to post, I will not reply.

NikosD

26th September 2020, 13:57

There is no madVR Labs / nVidia partnership. There is no other way for such a difficult task than get in touch with nVidia.
It's that simple and it's not something to be embarrassed of.
madshi has not even STARTED to look into making use of tensor cores.

I'm only making people aware of this INTENTION with no commitment, and you turn it into a FAILURE over THREE YEARS? You must stop putting words in others people mouth.
I've never used the word FAILURE over THREE years for madVR intention of using tensor cores.
I said that tensor cores are out for three years and there are years of promises/rumors/intensions - call it whatever you want - around it.
Do you realise how twisted your post is? Otherwise, you really need professional help. Judging from your first huge post and how you followed up next, I would advise you the same.
Whatever nonsense you decide to post, I will not reply. You have said that before in this thread. Promises, promises, promises...

huhn

26th September 2020, 14:17

what is this:
https://github.com/tensorflow/tensorflow/tree/v2.3.1

it links to this:
https://docs.nvidia.com/deeplearning/frameworks/tensorflow-user-guide/index.html

The TensorFlow container includes support Tensor Cores starting in Volta’s architecture, available on Tesla V100 GPUs. Tensor Cores deliver up to 12x higher peak TFLOPs for training. The container enables Tensor Core math by default; therefore, any models containing convolutions or matrix multiplies using the tf.float16 data type will automatically take advantage of Tensor Core hardware whenever possible.

and this also exist:
https://github.com/igv/FSRCNN-TensorFlow

does this prove to be using tensor cores no i have to say to be fair.

SamuriHL

26th September 2020, 14:42

There is no real information on the crashing issue at all currently. It might be driver. People might be running their PSUs close to their limit. AIB boards may be OCed a tad too aggressively and a BIOS update will sort that out. The capacitor theory is just that, a theory. I see no conclusive tests in any of that reporting.
Some AIBs are also reporting that they actually changed the capacitor layout in a final board revision to improve the situation, which might not be the one that was sent to reviewers.

The FE on the other hand is reportedly running quite hot and more noisy then good AIB cards. But I suppose if you were happy with previous FEs, this is still an upgrade over that.

And I wouldn't say the release was necessarily rushed by NVIDIA. The problem is the stringent secrecy, which didn't give the AIB much time to do full validation. So AIB cards might be rushed, because they wanted to be part of day 1 sales, but the reason for that being getting drivers, for example, only days before the actual launch.The drivers not getting released until day of launch was weird. It will be interesting to see what the next version fixes / breaks. Lol

As for the capacitor theory, yes it's a theory but it also makes sense given how much power these cards are consuming. You are right that the aib cards are driving higher than the fe. And it's been theorized that a vbios update to drop the max speed even 15 to 20 mhz might do the trick. Yes all still theory. Especially since I can't seem to get my hands on a card yet anyway. Lol I'm sure by the time I can they'll have this figured out.

I would love to get my hands on one of the asus cards and would gladly take one of those over the fe but I've not seen those anywhere yet. I'll keep looking until they become more widely available.

Sent from my SM-G975U using Tapatalk

NikosD

26th September 2020, 19:35

I'm reading a lot of rumors regarding ray-tracing performance of Turing/Ampere vs RDNA2 cards and I have to report a big-big-big disclosure regarding RT cores or ray-tracing cores of RTX cards.

Contrary on many beliefs, they do not exist in the form you think they do, meaning that they are not actually what most of the people believe:
An ASIC, fixed-function HW unit fully independent and dedicated on ray-tracing, like video HW decoders/encoders inside the GPU.

RT cores are actually an optimized, low-latency datapath inside the GPU heavily dependent from the well known CUDA cores, not only for hybrid RTX code like most of the RTX games, but even for fully path-tracing games like Minecraft and Quake II RTX.
The dedicated silicon for ray-tracing only calculations is desperately small.

The RT performance follows the classic CUDA core performance in both Turing and Ampere cards, it's not based on a separate block of fixed-function HW doing fully independently and separately ray-tracing calculations.

It's a little, dirty secret of nVidia that doesn't want the world to know, but I decided to disclose it here in my doom9 forum friends.

Don't tell it to the others!

nevcairiel

26th September 2020, 19:55

Your "secret" information is part of every official NVIDIA block diagram of both Turing and Ampere.
There is one RT core as part of every SM (Streaming Multiprocessor), which is the unit that contains the CUDA cores (128 per SM on Ampere), the Tensor cores (4 per SM), and also the RT core (1 per SM). Their position inside the SM very closely coupled to all other math functions clearly defines their role. Its not a secret. They don't have to be very huge, similar to Tensor cores, if you make something designed specifically for a certain type of math, it can do this much faster then eg. the CUDA cores.

A 3080 has 68 SMs, which makes for 68 x 128 CUDA cores, 68 x 4 Tensor Cores, and 68 RT cores.

NVIDIA even has block diagrams of what the RT core does internally, its not secret magic. It just does BVH traversal (eg. finding out which object is in a certain space), and ray intersection. Relatively simple math, which is just expensive to run on non-specialized hardware.

tyguy

26th September 2020, 20:09

I was told that nvidia changed their hdr implementation and madvr needed to be updated for it to not have a washed out picture. So today I loaded up a movie and forgot to change to os hdr and noticed the picture looks great again. Did nvidia change their hdr implementation once again?

Klaus1189

26th September 2020, 20:11

Look on post 2 ;)

nevcairiel

26th September 2020, 20:19

Manni

26th September 2020, 20:45

As for the capacitor theory, yes it's a theory but it also makes sense given how much power these cards are consuming. You are right that the aib cards are driving higher than the fe. And it's been theorized that a vbios update to drop the max speed even 15 to 20 mhz might do the trick. Yes all still theory. Especially since I can't seem to get my hands on a card yet anyway. Lol I'm sure by the time I can they'll have this figured out.

I would love to get my hands on one of the asus cards and would gladly take one of those over the fe but I've not seen those anywhere yet. I'll keep looking until they become more widely available.

Yes if the theory is true, Asus seems to have built their cards above the specs.

Re EVGA, I'm glad I went for the 3090 (2 MLCCs out of 6 (https://i.imgur.com/20VTCrx.jpg), same as the FE) vs the 3080 (1 MLCC out of 6). See here for a full table per AIB (https://new.reddit.com/r/nvidia/comments/izhpvs/the_possible_reason_for_crashes_and_instabilities/g6jivta/?utm_source=reddit&utm_medium=web2x&context=3).

That's if Igor's theory for the desktop crash is true of course. It has been confirmed by EVGA (https://imgur.com/1l3PTBN), so it's most likely true, though it looks like only reviewers got cards with 6 POSCAPs.

I never overclock, but I'll report how my 3090 XC3 does when tested with stock settings (if it fits in my case!).

Hopefully by the time they release the 3080 20Gb the reason for this will be known and the issue will be addressed.

claw

26th September 2020, 21:53

Yes if the theory is true, Asus seems to have built their cards above the specs.

Re EVGA, I'm glad I went for the 3090 (2 MLCCs out of 6 (https://i.imgur.com/20VTCrx.jpg), same as the FE) vs the 3080 (1 MLCC out of 6). See here for a full table per AIB (https://new.reddit.com/r/nvidia/comments/izhpvs/the_possible_reason_for_crashes_and_instabilities/g6jivta/?utm_source=reddit&utm_medium=web2x&context=3).

That's if Igor's theory for the desktop crash is true of course. It has been confirmed by EVGA (https://imgur.com/1l3PTBN), so it's most likely true, though it looks like only reviewers got cards with 6 POSCAPs.

I never overclock, but I'll report how my 3090 XC3 does when tested with stock settings (if it fits in my case!).

Hopefully by the time they release the 3080 20Gb the reason for this will be known and the issue will be addressed.

You might want to check this thread on the EVGA forum concerning performance in games. Don't know if it would affect MadVR performance as well. Thread title is 3080 but there is one poster who said it affects 3090 as well. Maybe an issue with the new driver and a workaround.

https://forums.evga.com/EVGA-XC3-ULTRA-3080-UNDERPERFORMING-m3096945.aspx

I have an EVGA 2080 Ti so I don't intend to upgrade unless a 3080 20G is helpful with any new algorithms Madshi might introduce.

SamuriHL

26th September 2020, 21:59

Yea, I'm inclined to believe there's at least some truth to the capacitor issue. And yes, Asus tends to engineer very good stuff. I'm leaning more and more towards an Asus. The FE, as Nev pointed out, is not able to boost as much because of thermal issues. Not that I OC, either, but if a card is CAPABLE of doing so then running at normal clock rates means it'll be quite a bit cooler (and maybe quieter?). My 680 was an ASUS and it lasted a *LONG* time until I replaced it with a 1060. The 2070 FE has been a great card for me which is why I was considering the 3080 FE. However, I've been convinced that Asus has their stuff together as usual. And with these cards pushing this much power, engineering matters more than ever.

SamuriHL

26th September 2020, 22:00

madshi has said memory won't be a problem which is why the Envy is going with 3080 10gb. The memory will be an issue for gaming, but, not madvr.

Manni

26th September 2020, 22:15

You might want to check this thread on the EVGA forum concerning performance in games. Don't know if it would affect MadVR performance as well. Thread title is 3080 but there is one poster who said it affects 3090 as well. Maybe an issue with the new driver and a workaround.

https://forums.evga.com/EVGA-XC3-ULTRA-3080-UNDERPERFORMING-m3096945.aspx

I have an EVGA 2080 Ti so I don't intend to upgrade unless a 3080 20G is helpful with any new algorithms Madshi might introduce.

Thanks, I use Studio drivers anyway, so it looks like the Precision X1 settings resolves this when/if it happens.

Note that I didn't go for the ULTRA because I don't want any factory overclock, or the FTW3 because of the form factor. I went for the XC3 gaming (no OC, but better binning than the black).

I'll report back any issue. I upgraded my PSU to 1000W a while back to prepare for my GPU upgrade. I'm glad I decided to overspec it, at least I should be fine on that front.

20GB is a waste of RAM/Money for video playback. There is zero need to go above the 3080 10Gb for madVR use, and even the 3080 power headroom over the 2080 super in Envy won't be used for a while, unless you stack multiple algos at 4K60p. So if you decide to upgrade, save some $ and get a 3080 10Gb.

If you don't need HDMI 2.1, you're good to go with your 2080ti for a good while, at least until prices stabilize and availability becomes a non-isssue. You might even skip a gen entirely. :)

ryrynz

27th September 2020, 06:21

Imagine buying a 30 series for potential tensor core functions with MadVR... Nobody is doing this. It's all about the raw performance and improved thermals/noise levels for the workload they require. I believe everything ever stated as being in MadVR 1.0 is now frozen, in the distant future we may see some more but given what I've see happening I don't expect much. The goals appear quite clear.

NikosD

27th September 2020, 11:06

Your "secret" information is part of every official NVIDIA block diagram of both Turing and Ampere. Can you share an official link showing the block diagram with RT cores which are placed inside SMs ?
I have seen and know very well the BVH diagram.
I want to see the building block diagrams describing RT cores as one per SM inside it.

nevcairiel

27th September 2020, 11:31

Techpowerup, for example, has it in their architecture rundown. The diagrams shown are officially from NVIDIA.
https://www.techpowerup.com/review/nvidia-geforce-ampere-architecture-board-design-gaming-tech-software/3.html

Full block diagram (red parts are disabled in a 3080, in comparison to a full Die):
https://tpucdn.com/review/nvidia-geforce-ampere-architecture-board-design-gaming-tech-software/images/rtx-3080-block-diagram.jpg

Couldn't find a proper high-res version in this second, but the yellow RT cores can be seen in every SM.

Single SM, build up of 4 warps with 32 CUDA Cores and 1 Tensor core each, and the single RT core.
https://tpucdn.com/review/nvidia-geforce-ampere-architecture-board-design-gaming-tech-software/images/ampere-sm.jpg

If you want even more official, this Whitepaper about Turing contains a rather similar block diagram, just for Turing instead of Ampere, but the location of the RT core is the same:
https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf

huhn

27th September 2020, 11:36

is there a diagram showing them outside?

https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf

Ray tracing acceleration is performed by a new RT Core processing engine within each SM (RT
Core and ray tracing features are discussed in more depth in Turing Ray Tracing Technology
starting on page 25).

more details are on 30/31.

NikosD

27th September 2020, 12:09

The message was never communicated like the post above by me.

The message from nVidia and tech journal at Turing release was as they managed to accelerate in HW and in real-time the ray-tracing procedure .

And the message still remains the same, even today.

In terms of video decoding acceleration, the implementation of RTX ray-tracing is what we call a very partial ray-tracing acceleration, a HW assisted implementation.

It implements in HW a tiny, little part of the ray-tracing algo, resulting a performance level, deeply dependant and proportional by the shader performance.

This is not a HW accelerated ray-tracing implementation.

It looks like the good old days of GPU assisted video acceleration, by shaders or small dedicated HW not able to accelerate the whole video decoding pipeline to make an analogy, not the nowadays of fully hardware accelerated video decoding/ encoding.

It was and still is a rushed move just to justify the price of the whole RTX 2000 series and especially asking 1200€ for 2080 Ti with a big performance hit due to the partial acceleration.

Take a look at the size of RT cores in the photos of the block diagram of SMs
It looks huge compared to CUDA cores, while we know it's a tiny little silicon inside.

Ampere improved the ray-tracing acceleration mainly because of faster shader performance (~70% for games at 4K vs 2080) and only around 10% for ray-tracing HW acceleration by itself.

This is the message that I want to make it clear.

butterw2

27th September 2020, 13:27

When any discrete gpu can handle 1080p60 gaming, but you still want to sell expensive gpus, what are you to do ? :)

RT Cores, Nvidia RTX ON, always sounded like marketing BS to me.

I wish they would at least bring some efficient/affordable gpus to the market.

For GPU accelerated machine learning training, however, my understanding is that Tensor Cores and VRAM quantity are actually useful.

Klaus1189

27th September 2020, 15:40

https://www.youtube.com/watch?v=RlDC7u7GpR4
and following link also in the description of the video - the english "review"
https://www.igorslab.de/en/nvidia-geforce-rtx-3080-und-rtx-3090-and-the-crash-why-the-capacitors-are-so-important-and-what-are-the-object-behind/

Klaus1189

27th September 2020, 16:01

https://www.youtube.com/watch?v=THMukcOzB8g

NikosD

27th September 2020, 19:22

It is now reported by many HW review sites, that nVidia decided to rush in an extreme way the release of 3080/3090 cards by pushing the AIBs to their limit.

They got the drivers in the last minute and they had time to test only power on/off and basic stability issues.

The AIB cards used the owners/ buyers of 3080 cards like guinea pigs for real workloads and games, mainly because of nVidia's aggressive behavior against them.

But after the last leaks it's time to forget all these and leave them behind, in the past.

We have to speak about the elephant in the room - Big Navi21

It seems, according to the latest leaks, that AMD is going to release two flavors - 21a and 21b - both of them maxed out at 80CUs.
The first flavor 21a has a huge clock of 2.0GHz at only 200W (!) TDP and 21b has even faster clock at 2.2GHz at only 240W (!!))
According to the latest leaks, the 21b is already faster than 3080 and is closing the small gap of 3090.
This is HUGE news.
A card that is faster than 3080 with a 240W TDP is simply unbelievable, it's like that RDNA2 architecture is the 8th Miracle of the World.
Of course we have to see the real performance, not leaks, but it seems confirmed by many sources.
The 28th of October could be a day that nVidia will remember as a nightmare, if the leaks proved to be valid but for all the others could be a day for celebration.

SamuriHL

27th September 2020, 20:04

OMG your "conspiracies" and "inside information" is beyond laughable. None of what you posted is "new information". We already know they didn't get the drivers until a couple days before release. We already knew they didn't have time to fully test before cards were sent to reviewers. They used REVIEWERS as advanced testers, yes, but they did NOT as you seem to be stating use owners as such. And those reviewers that got pre-release boards? Oh noes! What are they to do? Oh, right, wait until the AIB replaces the board for them with production versions...which they are doing case by case. Your attempts to make it seem SO MUCH WORSE than actual reality is literally hilarious. I'm certainly not apologizing for nVidia here, far from it as I still can't buy a card, but, you go the EXTRA mile to make it seem like the biggest conspiracy in history. LOL

All hail the almighty AMD to rescue us from the TRAVESTY of the evil nVidia. LMAO

oldpainlesskodi

27th September 2020, 20:09

Me, still waiting for 16k....heard it's going to be mind blowing......

NikosD

27th September 2020, 20:28

@SamuriHL Are you OK my friend ?

The cards have problems right here right now, they are in the hands of BUYERS and OWNERS not REVIEWERS.

What conspiracy theories and inside information are you talking about ?

GAMERS not reviewers, mentioned problems with these cards, although if you read the videocardz article you will see that one AIB mentioned some problems before anyone else.

The guinea pigs are the frustrated GAMERS, OWNERS not REVIEWERS.

I made a summary of more than five articles that I read on this subject for users of this forum that didn't have the time to read them.

AIBs released the cards without testing them in games and real-world workloads.

It's just the simple truth and nothing else.

I can't feel pity for your frustration reading my posts, but the truth is that I don't remember such a problematic release of super high-end cards in the last many years.

28th of October is coming....

SamuriHL

27th September 2020, 21:19

Me, still waiting for 16k....heard it's going to be mind blowing......

LMAO :D Imagine 8K in each eye for 3D?! WOA LOL

SamuriHL

27th September 2020, 21:20

@SamuriHL Are you OK my friend ?

Yup. Perfectly fine. And grounded in reality. Thanks for asking.

chros

28th September 2020, 10:47

AMD is going to release two flavors - 21a and 21b - both of them maxed out at 80CUs.
The first flavor 21a has a huge clock of 2.0GHz at only 200W (!) TDP and 21b has even faster clock at 2.2GHz at only 240W (!!))
According to the latest leaks, the 21b is already faster than 3080 and is closing the small gap of 3090.
This is HUGE news.
A card that is faster than 3080 with a 240W TDP is simply unbelievable
Indeed, if it will be true then this will be a huge slap in the face :)

el Filou

28th September 2020, 12:45

When any discrete gpu can handle 1080p60 gaming, but you still want to sell expensive gpus, what are you to do ? :)
RT Cores, Nvidia RTX ON, always sounded like marketing BS to me.Probably like 32-bit colour rendering sounded like marketing BS to some people (mostly fans of 3dfx) in 1998. I mean, you're just describing the normal evolution of tech, the first implementations of a new tech are always lower performance than the older standard but after a few generations nobody would want to go back to the old standard. Would you have have wanted to go back to 16-bit rendering in the mid-2000s?