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]

Originally Posted by varekai

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]

Originally Posted by chros

The other thing that there are really confusing "leaks" about RDNA2, some state it's crap, some state it's an Ampere killer
How on earth can this happen with AMD at this stage that I don't understand ...

This is what Coreteks thinks 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]

Originally Posted by el Filou

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: "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]

Originally Posted by VBB

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]

Originally Posted by el Filou

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]

Originally Posted by Pastol

Hi.

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

IMO 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?