AMD, Intel and Nvidia driver issues and last recommended version

[NikosD]

nVidia is going to acquire Arm, if some specific companies and countries allow it.

It's not finished yet.
It seems that nVidia wants to become "Apple of silicon" but I don't know what real Apple and China say about this.

Apple is the worst company in the world, but this is a different discussion.

[VBB]

Hello!

GPU NVIDIA Geforce GTX 960, drivers 398.11 win7 x64.
Would this old GPU benefit with latest drivers 452.06?

It's safe to use up to 446.14. Anything newer breaks HDR with madVR.

[SamuriHL]

^^^^

Yea! That's why I want the new driver so badly. LOL

[chros]

The other thing that there are really confusing "leaks" about RDNA2, some state it's crap, some state it's an Ampere killer :D
How on earth can this happen with AMD at this stage that I don't understand ... :)
This is what Coreteks thinks (https://www.youtube.com/watch?v=BR70xbcwB6U) about RDNA2 GPUs. In short: big Navi will be only 3070 + 15% (with AMD preferred titles!) and won't be much more power efficient than Ampere.
(He also tries to answer my question above.)

Well, if that will be the case then only the pricing can save AMD from another disaster :)

[SamuriHL]

That's pretty much what I've been speculating this entire time. I think they beat the 3070 but fail to match the 3080. And they will be forced to lower prices before launch to compensate.

Sent from my SM-G975U using Tapatalk

[el Filou]

Oh look, AMD has teased a 3D render of their cooler design and confirmed 2x 8-pin power.
It would be pretty funny if NVIDIA announced an RTX 3060 on the day of AMD's Navi2 presentation (or just after).

[chros]

It would be pretty funny if NVIDIA announced an RTX 3060 on the day of AMD's Navi2 presentation (or just after).
They will do it before it (AMD is on okt 28).

And here's Jim's ranting about Ampere (https://www.youtube.com/watch?v=tXb-8feWoOE&t=269s): "Ampere is no Pascal", "Ampere is bad choice for everyone" :)
He talks about power consumption that is mainly caused by Samsung 8nm.
He doesn't mention 1 thing though that Coreteks talks about in his RDNA2 video: the power hungry TSMC 7nm+ :)

I still think that lower end Ampere cards (3060/3050/etc) will be the go to for madvr if madvr receives any update at all in the future and those cards will get all the goodies of the higher models (HDMI 2.1, AV1 decoder, etc).

[huhn]

there is no point in saying ampere is power inefficient when it is currently the most power efficient card "on the" market. (we see tomorrow if that's true...)

but what do you expect from someone saying "these clocks weren't fast enough to feed these numbers of shader" ok that's how a shader work sit's the feeding speed thank you...

what we currently know about 7nm is that it performance terrible in navi and can't even beat 16 nm TSMC at the same clock speeds.

and even on zen 2 the clock speeds are not that impressive it's competing with a process that's been refined for like ever but still not that impressive if you consider that 7nm is a much better process but power efficiency is quite impressive.

until AMD releases a GPU which is more power efficient using 7nm his "argument" is absolutely worthless. and even if it is more power efficient which it easily could be that again is not that great if the heavily needed process is to expensive.

[varekai]

It's safe to use up to 446.14. Anything newer breaks HDR with madVR.
OK, thanks for the info.
Do you think there will be any performance improvements with that driver?

[el Filou]

Not with madVR and on Win7 and an older GPU like 960, I don't think so.

Apart from obviously supporting newer GPUs, newer drivers are only good for:
- better supporting newer games
- supporting newer APIs, or WDDM versions in Win10
- sometimes providing better default video timings or functionalities (like some driver a few years ago added some LPCM sound formats for HDMI)
- security issues if you worry about that

But they also sometimes break things in point 3.
If you don't have any bugs bothering you and don't care about the above, I don't see a need.

[NikosD]

Interesting guy that one from Coreteks, although a little nVidia biased.

So, I agree with him like MLID guy that due to internal reconstruction of caches, the Big Navi21 can be faster (a lot ?) than 3070 with the same 256bit memory GDDR6 and with 16GB VRAM
No 320bit and 20G VRAM as I thought before in order to compete with 3080.
I lost this bet against the world, because probably it doesn't need 320bit to come close to 3080.
But he says only +15% over 2080 Ti.
That could be the case before Ampere's announcement with Big Navi21 initially at 275W TDP.
The last conversations at AMD, as MLID and Igor said also, have put Big Navi21 to 300W just a little less than 320W of 3080.
The higher clocks of Big Navi21 at 300W will bring performance a lot closer to 3080 just as power consumption.
I also believe like him that AMD will bring HBM2e Big Navi on Radeon gaming cards to beat 3090 but nVidia could bring an Ampere Titan card to respond.

Two bad things that I kept from Coreteks' analysis that really annoyed me and possibly could be true but I hope not:

1) No smaller/ cheaper cards than Big Navi21 for 2020!
Even the Big Navi22 will be released on late Q1 2021 or early Q2 2021 according to him.
That's a disaster (if true) for smaller wallets.

2) He said that AMD will repeat itself like Vega64 LC (Liquid Cooling) and 5700 XT Anniversary card for the highest Big Navi21 300W card that competes with 3080.
And the "normal" Big Navi21 at probably 275W will be like 3070 (+15% ?)
That's not good but it's not bad.
AMD still (before RDNA2 architecture) hasn't managed to beat 1080 Ti Pascal card.
So, let's wait for tomorrow's real-world reviews on nVidia's cards in order to see the perfect samples of Ampere 3080 GPUs that leather-jacket-man has sent to reviewers in order to be highly overclockable and with the lowest power consumption possible, how will stand against 2080/2080 Ti.

Is it possible for 3080 to be twice as faster in rasterized games than 2080 on average for the ten- fifteen most popular games ?
No.

But measuring the real-world performance of 3080 will probably unlock Big Navi21 benchmarks a lot sooner than 28th of October...

[varekai]

Not with madVR and on Win7 and an older GPU like 960, I don't think so.
But they also sometimes break things in point 3.
If you don't have any bugs bothering you and don't care about the above, I don't see a need.
Thanks for the info.
"if it ain't broke, don't fix it"

Best regards,
varekai

[Pastol]

Hi.

What is the alternative to madvr on the latest nvidia drivers for hdr material

[SamuriHL]

Hi.

What is the alternative to madvr on the latest nvidia drivers for hdr materialIMO the isn't one. The problem exists in other players too.

Sent from my SM-G975U using Tapatalk

[NikosD]

Reviews are out for 3080.

14 Game Average of Hardware Unboxed:

1440p
3080 is 20% faster than 2080 Ti and 50% faster than 2080

4K
3080 is 30% faster than 2080 Ti and 70% faster than 2080

More or less, the expected results.

[nevcairiel]

More interesting, in heavy OpenCL/CUDA workloads (eg. Blender tests), there is a strong >100% uplift over 2080. This might benefit compute-heavy scenarios like running very heavy madVR settings quite a bit, I could imagine, since any other factors that could limit performance are pretty much non-existant and the huge number of FP32 cores can really do their work.

[huhn]

there are benchmarks where the 5700 XT wins against a 3080 so production benchmarks are sadly hard to judge. as far as we know madVR is shaders "only" so yes could be great but well polaris anyone?

and yes it's the most efficient card on the market.

[NikosD]

After three reviews I have seen so far, it's pretty obvious that nVidia failed this time badly to deliver a gaming (Geforce) card.
Ampere architecture is basically a computing architecture for data center and this is obvious in gaming as well.
The performance of 1440p - a really high resolution for most gamers as still the majority uses 1080p monitors - is not so good compared to 2080 Ti and even 2080.
The power consumption is huge, analogous to the Data Center destination of Ampere architecture.
Real-world power measurements of 3080 give numbers between 330W to 350W as peak power consumption vs 250Watts of 2080 Ti.
The result is that 3080 is only 8% better in performance per watt compared to 2080 Ti at 4K resolution.
The 3080 is strictly a 4K card not a reason to buy it if you have a 1440p monitor or below.
But at the same time 4K is going to need more than 10GB of VRAM in the near future, so a disastrous combination.
AMD has a real opportunity here to attack viciously this non gaming friendly card with Big Navi.
And with AV1 HW acceleration added to AMD Radeon RX 6000 cards, I have really all the good reasons to wait for a good price to performance VFM Big Navi card than to buy an Ampere card to run Blender!
For God's sake!

[SamuriHL]

Yes....you ALL want to wait for AMD. /waves hand like Jedi

Then I can get my 3080 tomorrow. :)

[huhn]

well i wonder how high the number of user is that buy a 3080 for 700 and have a 1080 screen?

[nevcairiel]

If I actually planned on gaming at 4K, I might wait for a 20GB variant, or a Ti, or whatever they are going to have, but right now I'm on 1440p, and likely to stay there for the time being. But I also have other reasons to get a 3080, because I do use productivity tools, and of course I need something to work on AV1 decoding, so that I can finally make a worthwhile LAV release again. :)

I agree that on 1080p its mostly nonsensical, you can get a 3070 there for high-end performance.

[NikosD]

Interesting that Digital Foundry accepted their mistake or "mistake" to benchmark Doom Eternal to highest Texture Settings at 4K that were simply above the 8GB of 2080 and the performance difference from 3080 was about 100% or even more due to limited memory of course.
Using a slightly lower setting "ultra" in order for textures to fit in the 8GB VRAM the performance difference is around 70% like the average of 4K gaming.
If 2080 was a PCIe v4.0 card tested on Ryzen platform in order to support PCIe v4.0, the performance difference could be less.
But I also see some reviewers like Digital Foundry that are trying a little desperately to increase the hype, comparing a 3080 vs a 1080 card of 2016!
Oh, come on guys...You can do better than this!

[chros]

For 1080p probably a 3060 will be enough :)
The rest is really good news, this card is a beast in every way.

[SamuriHL]

Yea for 1080p 144hz gaming I have my monster Alienware Area 51m with a 2080 so I'm covered there. I agree about the memory, Nev, but, I'm not waiting. I think that 10gb of GDDR6x will be fine for the majority of 4k gaming. And if after a year or 2 it isn't, ok, well I can upgrade again to something better at that point. I don't necessarily buy the October surprise rumors about a 20gb 3080 unless AMD hits a homerun which is looking increasingly unlikely. AV1 support would be nice, though. And if you have a 3080 I can be sure it'll work well for me when you release it. Sweet :D

[SamuriHL]

Are they seriously going to wait til tomorrow to release the damn driver? SIGH

[CZ Eddie]

https://us.download.nvidia.com/Windows/456.38/456.38-desktop-win10-64bit-international-whql.exe

https://uk.download.nvidia.com/Windows/456.38/456.38-desktop-win10-64bit-international-whql.exe

https://us.download.nvidia.com/Windows/456.38/456.38-win10-win8-win7-release-notes.pdf

Nvidia released GeForce 456.38 WHQL driver.

Fixed Issues list includes:
[madVR][MPC-HC]: Various HDR issues occur when using the madVR renderer with MPC-HC. [3038381]

https://videocardz.com/driver/nvidia-geforce-game-ready-456-38-whql

[SamuriHL]

At least the driver is released. But no humans were able to buy any cards. They were sold out in microseconds, all snatched by bots and are now up on eBay. WTF Nice job, nVidia. Way to push people to consoles. At least those pre-orders are easier to get.

[CZ Eddie]

This one shows available at Best Buy
https://www.bestbuy.com/site/gigabyte-geforce-rtx-3080-10g-gddr6x-pci-express-4-0-graphics-card-black/6430621.p?skuId=6430621

[CZ Eddie]

This one shows available at Best Buy
https://www.bestbuy.com/site/gigabyte-geforce-rtx-3080-10g-gddr6x-pci-express-4-0-graphics-card-black/6430621.p?skuId=6430621

Not anymore.
Keep checking other cards on BB because I saw another one come up available that had been out of stock earlier.

I'm waiting for something to show up on Dell so I can use my $400 in coupons & credits I have available.
I have no idea when they will start selling them.

[SamuriHL]

Yea they're releasing them periodically but before I can even get through the cart it's gone. It's insane.

[huhn]

HDR works "fine" on this driver even with overlay rendering.

edit: this is not an indeep check if meta data and such.

[SamuriHL]

I'll check metadata tonight. I have to hook the integral 2 back up to do it, but not a big deal.

[chros]

I'll check metadata tonight. I have to hook the integral 2 back up to do it, but not a big deal.
Thanks, can you also check the rest of the bullet points (https://forum.doom9.org/showthread.php?p=1921810#post1921810) as well?

Jim explains Ampere cuda cores (https://www.youtube.com/watch?v=nJBggUfYozY) one more time. Interesting stuff in it:
- 3080 has ~3x cuda cores of 2080, but only 70% faster
- according to nvidia, ~25% of instructions are int out of all
- since the new path can either run fp or int, and int are 25%, this path can be occupied 50% by int

[nevcairiel]

- 3080 has ~3x cuda cores of 2080, but only 70% faster


Games (and game-like benchmarks like 3DMark) are often being limited by something else rather then pure FP32 performance. More shader-intensive workloads would benefit from the additional cores more. Maybe future more graphically intensive games will benefit more from it.
Some productivity benchmarks have shown the real power of the massive number of FP32 CUDA cores, but of course for that to be meaningful an application has to be able to use it.

Interestingly, AMDs Fiji architecture from years ago had the same issue. They had a lot more potential shader power then they could get games to use. Todays use might be limited, but future use, who knows. Free shaders for RTX IO and stuff. But of course one shouldn't buy hardware on account of future what-ifs.

(Also, I can't watch Jims videos, his accent just drives me crazy)

[SamuriHL]

Thanks, can you also check the rest of the bullet points (https://forum.doom9.org/showthread.php?p=1921810#post1921810) as well?


I can check that metadata is being sent, someone else will have to test the rest of the bullet list.

[huhn]

like rop, TMU, cache and so much more.

memory bandwidth is not comparable too because they changed the lossless compression over the years.

[NikosD]

You don't need to watch Jim.
You only need to read my posts here, I've been telling you the same things from day one BEFORE he releases his videos.
Now, regarding the gaming workload analysis, the 25% of INT commands mentioned above is wrong.
The figure given by nVidia when Turing was released back then at 2018 that was the first architecture to execute INT+FP at the same time was around ~35 % or so to speak 1/3.
That's why the 30TFLOPS become 20TFLOPS and the number of CUDA cores follow the same pattern of 2/3 real-world number compared to Turing.
Now I have some info/news/prediction for Big Navi.
It seems that the model that will be announced on 28th of October will be a 60CU - 64CU die that could possibly reach 2.4GHz for the 3 fans edition of 300W+ TDP.
It will have 256bit bus with 16GB of GDDR6.
I'm under the impression that will beat 3080 at both 1080p and 1440p resolutions but will lose a little at 4K.
AMD will release later the full die of 80CUs using HBM2e memory and will beat even the 3090 card even at 4K.

[SamuriHL]

So, mostly good news with this new driver. More or less. When you play an HDR title now, it'll start out as SDR and then switch to HDR mode and then enable HDR. Those of you who aren't familiar with the work around for the previous few drivers of starting a movie, stopping it, and playing it again are probably wondering wtf I'm talking about, but, they've essentially appeared to "automate" that process. The metadata is being sent correctly now, as well. Hooray for small miracles. Now for the not so great stuff. If you bring up your player's UI, it'll take you out of HDR into SDR. But we've seen that behavior quite a bit before. It's whatever. I can live with this current driver. And hey, if I ever do manage to land a 3080 in the next year, I'll be all set. LOL

[VBB]

Thanks for testing so quickly, bud! So, are we back to using NVidia's API for HDR with this driver, or does it still switch to OS HDR? The player UI bug has been around since early this year, so I guess that hasn't been fixed yet.

[Klaus1189]

Any improvements of the out-of-the-box 23 Hz refresh rate? Dropped/repeated frames...

[huhn]

i have nor rebooted yet but i'm back to 23.980

[Klaus1189]

Please reboot :)

[huhn]

same.

[Klaus1189]

Have they added new options, like 10 bit? Huge new branch so possible in my head and a workaround for madVR ... very nice of Nvidia.

So the new driver behaves like that?
445.75 2020-03-23 This version and all 445.XX and higher: Playback at greater than 8-bit drops back to 8-bit (and out of HDR) in FS windowed mode when player GUI is triggered, regardless of bit-depth of the file. Can be circumvented by using FSE mode or by setting madVR to 8-bit.

[huhn]

i can only tell you that "everything" is fine at my end with overlay rendering if it comes to HDR.

and they did not add 10 bit in this driver and they may never add 10 bit for HDMI 2.0 signals jusdt for HDMI 2.1 (they are packet differently).
edit: please report if 23.976 is 23.976 on your system witht he newest driver so i can report it to madshi. one case is far to little to come to a conclusion

[oldpainlesskodi]

See, I can't help myself. Lord forgive me.....AMD=HTPC.

[Klaus1189]

If someone got a 3080 please let us know then if there is 10 bit available. You will of course need a display with HDMI 2.1 input like newer TVs and please report your refresh rate here, info is needed if 23.976 is 23.976 on your end, see previous post of huhn.

[huhn]

See, I can't help myself. Lord forgive me.....AMD=HTPC.

so you found a way to send BT 2020 to the display at will ?

[DMU]

@huhn
what are you speaking about?

[oldpainlesskodi]

@ Huhn...that's loaded question, so Touché. And, no, I see no reason to send BT2020 at will.