Does Nvidia really gimp their cards?

[dfg666]

380X > 780 Ti

229 USD > 699 USD

Will this be just a case of Kepler being the worst architecture or does Nvidia actually gimp their cards? The 780 Ti competed with the 290X and it outperformed the 290X on launch.

 

[Zeeee]

Holy!!! thats insane and hella interesting. WOW looks like Nvidia forgot about kepler.... thats sad, really sad. What the heck. Im scared for my 1070

[manikyath]

no they dont.

 

at least not on purpose, with that much silicon it happens that issues sneak into a driver, but if they actually did i wouldnt have a gaming capable GT640 now. and in fact my GTX970 has gotten a considerable amount faster since i got it.

 

and i dont need to explain to you that a reviewer has no idea what they're doing when they list SLI configurations for a game that quite clearly does not support SLI.  (beyond that, the CF numbers arent very impressive either)

 

i'd rather blame the game or the person that did the benchmark than blame nvidia.

 

EDIT: should add, this is one benchmark in one game, with you complaining about one card's result.

EDIT EDIT: http://gpu.userbenchmark.com/Compare/Nvidia-GTX-780-Ti-vs-AMD-R9-290X/2165vs2166

[Enderman]

The "nvidia gimps cards" is just BS made up by some AMD fanboys, don't fall for that crap.

 

The reason it seems like nvidia cards "decrease" performance is because nvidia cards get driver updates right after launch, and multiple times, improving performance by a bit.

After the initial bunch of driver optimizations there isn't any huge leaps in performance from drivers.

 

When AMD releases a card, it's pretty bad for a while until they start fixing their drivers and optimizing things better which can take anywhere from a few weeks to a few months.

Since AMD only releases new drivers a few times a year, or sometimes only once a year, they end up performing much better in the future than they did around launch.

 

If you look at benchmarks you will see the GTX____ performs the same or better than it did when it launched years ago, and if you look at an AMD card you will see it have a big improvement between release and where it is now.

 

 

About that benchmark you posted, it seems like this is an isolated case, if you look at other benchmarks (aka more than one) you will see that a 960 does not perform better than a 780ti. It looks like this game just uses certain features which don't work very well on older GPUs. You can blame game developers for that.

 

[Cheddle]

apparently its because Forza 3 has HUGE levels of geometry to process. the older architectures just arn't very good at pushing so many polys.

[Mira Yurizaki]

Well here's another thing. This is just one game on one website. This is not representative of a whole. If you really wanted to make a case, go find benchmarks for several games from several websites.

[Dabombinable]

16 minutes ago, M.Yurizaki said:

Well here's another thing. This is just one game on one website. This is not representative of a whole. If you really wanted to make a case, go find benchmarks for several games from several websites.

Or get a Kepler based card, and run it on the first drivers, then the current ones (spoiler: you see a large and noticeable improvement).

[Mira Yurizaki]

Just now, Dabombinable said:

Or get a Kepler based card, and run it on the first drivers, then the current ones (spoiler: you see a large and noticeable improvement).

You still need to test it on multiple games to make sure it's the hardware, not the software.

[Dabombinable]

Just now, M.Yurizaki said:

You still need to test it on multiple games to make sure it's the hardware, not the software.

I did that with my GTX 650ti OC 2GB. It ended up being able to hit 60fps more frequently, and gameplay was smoother overall.

[Mira Yurizaki]

Just now, Dabombinable said:

I did that with my GTX 650ti OC 2GB. It ended up being able to hit 60fps more frequently, and gameplay was smoother overall.

And with what games? What settings? You have to have all of this splayed out so others can repeat your tests to confirm your findings. Just because you get these results doesn't mean someone else will.

 

[Dabombinable]

1 minute ago, M.Yurizaki said:

And with what games? What settings? You have to have all of this splayed out so others can repeat your tests to confirm your findings. Just because you get these results doesn't mean someone else will.

 

That's my old graphics card-its with the i5 4440 that I donated/loaned, since I managed to replace my previous at the end of last year with this one. Also-ever tried recording with quicksync?

[Deli]

a 960 beats a 780Ti.

 

I feel sorry to own an even slower 780.

[Nena Trinity]

Older Nvidia cards does not handle tessellation as good as Maxwell & Pascal same with the AMD side where Polaris is better than all earlier GCN cards at it...

 

Edit: 285, 380, 380X, Fury & Fury X are also improved when it comes to tessellation but not as much as Polaris...

[Dabombinable]

20 minutes ago, Nena360 said:

Older Nvidia cards does not handle tessellation as good as Maxwell & Pascal same with the AMD side where Polaris is better than all earlier GCN cards at it...

Though if you run Unigine Heaven, and toggle tessellation on and off, you do see that despite the improvements over old architectures, the impact of tessellation is still uite extreme (that and you actually get to see the difference it makes to the graphics, AKA a lot-excluding hair)

[xAcid9]

Rather than gimping it's more like they stop optimizing the architecture because changes from Kepler to Maxwell is pretty significant.

[D2ultima]

I think there's something off...

 

The 380X (Tonga) is just under the 290 (hawaii) despite clearly being significantly weaker card, and the RX 470 (Polaris) is outperforming the 290X and 1060 and almost matching a 980.

 

Since this is Micro$haft UWP exclusive and in DX12 (removes driver overhead from AMD's drivers and generally gives their cards a performance boost for that reason only) it seems like the game is simply optimized more for those two architectures.

 

We saw similar things in the past with UWP-only titles, where say... R9 390 was doing worse than R9 380X and having huge framedrops for random reasons. Besides, the 770 and 780 are so close together it's ridiculous... 2fps difference? The only thing I could consider is that the game uses some form of async compute and Kepler, which cannot handle it, is hugely destroyed in performance... but then Hawaii can do 64 queues so that makes no sense either.

 

Let's just chalk it up to Micro$haft being a terrible company and making games be awful on PC.

 

 

[Morgan MLGman]

5 hours ago, Enderman said:

Since AMD only releases new drivers a few times a year, or sometimes only once a year, they end up performing much better in the future than they did around launch.

Sorry but gotta call BS on that, I'm getting a driver update more than once a month for quite some time now. On a 2013 290X.

[i_build_nanosuits]

i'm sorry but these numbers in the graph don't make any sense what so ever...58/69 FPS on a 290X...BULLSHIT...that thing can barely reach 40FPS...listen to this video you can start watching at 2:40 or just watch the whole thing:

 

290X faster than 970...yeah i call BS on that...also 970 faster than 780ti by 20FPS across the board...also BULLSHIT.

[xAcid9]

50 minutes ago, D2ultima said:

I think there's something off...

 

The 380X (Tonga) is just under the 290 (hawaii) despite clearly being significantly weaker card, and the RX 470 (Polaris) is outperforming the 290X and 1060 and almost matching a 980.

 

Since this is Micro$haft UWP exclusive and in DX12 (removes driver overhead from AMD's drivers and generally gives their cards a performance boost for that reason only) it seems like the game is simply optimized more for those two architectures.

 

Let's just chalk it up to Micro$haft being a terrible company and making games be awful on PC.

 

 

They use AES 256 encryption for the entire Forza Horizon 3 files. CPU usage is pretty high on 1 core which will bottleneck all AMD card especially pre-Polaris since Polaris is more efficient with it's new geometry engine, primitive discard accelerator, improved shader engine. That's why sometime it could out perform the slightly stronger card like 290x/390x or even Fury/Nano in CPU intensive games. Also no ASYNC support.

 

 

[D2ultima]

2 minutes ago, xAcid9 said:

CPU usage is pretty high on 1 core which will bottleneck all AMD card especially pre-Polaris since Polaris is more efficient with it's new geometry engine, primitive discard accelerator, improved shader engine.

The card shouldn't matter. It's the drivers that are the problem, in consideration to AMD's huge driver overhead problem for DX11 titles. However, since the game is in DX12, there is no driver overhead whatsoever. I think the game is simply coded to be more friendly toward Tonga and Polaris.

 

As for the decryption, I heard about that. God, they're so stupid. I don't know why people pay for the stuff they put on UWP. NONE OF THEM are designed to benefit the consumer in any way.

[xAcid9]

3 minutes ago, D2ultima said:

The card shouldn't matter. It's the drivers that are the problem, in consideration to AMD's huge driver overhead problem for DX11 titles. However, since the game is in DX12, there is no driver overhead whatsoever. I think the game is simply coded to be more friendly toward Tonga and Polaris.

 

As for the decryption, I heard about that. God, they're so stupid. I don't know why people pay for the stuff they put on UWP. NONE OF THEM are designed to benefit the consumer in any way.

Going to DX12 doesn't magically eliminate AMD driver overhead imo, but yeah the game/driver probably need some patches for the older cards to work efficiently similar to GoW:UE.

 

I don't have the game to test which setting cause these huge frame drop for AMD cards, my guess probably Geometry/Shadow related. 

Refuse to pay $60 for it... maybe $20. 

[Enderman]

4 hours ago, Morgan MLGman said:

Sorry but gotta call BS on that, I'm getting a driver update more than once a month for quite some time now. On a 2013 290X.

Sorry but I remember when they didn't release a driver update for almost a year and people were really pissed off so they hired more people for their software division and created Crimson. Maybe you don't remember that?

[D2ultima]

4 hours ago, xAcid9 said:

Going to DX12 doesn't magically eliminate AMD driver overhead imo

 

I don't have the game to test which setting cause these huge frame drop for AMD cards, my guess probably Geometry/Shadow related. 

Refuse to pay $60 for it... maybe $20. 

Yeah, it does; it's the only thing that's giving them huge performance boosts from DX12/Vulkan. nVidia killed as much driver overhead in DX11 as possible a couple years ago; around the time Thief was out and using Mantle. It's why their cards worked better with DX11 than AMD cards did with Mantle, as Mantle's low-level access benefit bypassing the CPU was negated. AMD's primary issue is that for their drivers (not saying that optimization doesn't play a part; it does... but the primary issue fixed by the low-level APIs DX12, Vulkan and Mantle, is removal of CPU driver overhead). It's also why nVidia cards perform the same or worse in DX12/Vulkan too. The games aren't having optimizations baked in using the low level API; they're just using the API to kill CPU overhead and get better CPU load balance and reduced usage. In this respect, the games performing worse is more because DX12/Vulkan drivers are not nearly as mature as DX11/10/9/OGL drivers, so the direct performance is usually slightly worse.

 

As for how much the games are worth, honestly I feel that with the level of optimization/work they put in, the fact that it's UWP only, the fact that I need to be on shitty Windows 10, and the fact that I can't do anything with the game files or run custom profiles for SLI etc? They might as well be paying me to play it.

 

38 minutes ago, Enderman said:

Sorry but I remember when they didn't release a driver update for almost a year and people were really pissed off so they hired more people for their software division and created Crimson. Maybe you don't remember that?

They always release a few beta drivers a year. Usually at least 4 or 5. But if you're talking about WHQL drivers, yes. They rarely released those in the past. However, WHQL means absolutely nothing anymore. nVidia has put out "WHQL" drivers that cause BSOD boot loops and break windows installs and all sorts of crap this last year and a half. Constant TDRs in normal PC usage were common as hell around 353.06 all the way until r361 branch drivers. All sorts of garbage. If you're specifically waiting for AMD to release WHQL drivers, you're more or less just shooting yourself in the foot. Might as well see if a beta driver is out, wait a day or two, see if any rampant problems show up, and if not, download it and use it.

[Enderman]

1 minute ago, D2ultima said:

They always release a few beta drivers a year. Usually at least 4 or 5. But if you're talking about WHQL drivers, yes. They rarely released those in the past. However, WHQL means absolutely nothing anymore. nVidia has put out "WHQL" drivers that cause BSOD boot loops and break windows installs and all sorts of crap this last year and a half. Constant TDRs in normal PC usage were common as hell around 353.06 all the way until r361 branch drivers. All sorts of garbage. If you're specifically waiting for AMD to release WHQL drivers, you're more or less just shooting yourself in the foot. Might as well see if a beta driver is out, wait a day or two, see if any rampant problems show up, and if not, download it and use it.

Yes I'm talking about full release drivers.

 

And nvidia made one driver with a bug that would make your PC unable to boot when you had a second monitor plugged in.

You could unplug that monitor to boot properly.

The bug was fixed in a release a week later.

[D2ultima]

Just now, Enderman said:

Yes I'm talking about full release drivers.

 

And nvidia made one driver with a bug that would make your PC unable to boot when you had a second monitor plugged in.

You could unplug that monitor to boot properly.

The bug was fixed in a release a week later.

I've kept up with nVidia drivers pretty well until 365.19 came out. Every single driver release has been plagued with tons of issues which you could see in their release threads. The driver you're thinking of is probably the first r364 branch driver that got pulled, and I can guarantee you, it caused a whole host of other issues. And BSOD boot loops were one of them. The number of my acquaintances who started complaining after installing it of their "PC being broken" and stuff to which I told them run DDU and use 362.00 is not small. Not everybody had issues, certainly, but their drivers from 350.12 all the way through to I'd say 365.10 have for the most part (not completely) been garbage. And every. last. one. has been WHQL. They even asked people to send them their PCs who were getting constant TDRs around 353.06 just to determine what the problems were, because apparently they couldn't figure it out.

 

Don't think WHQL means anything anymore. If your windows install breaks, or you get hardware issues/damage (the first windows 10 driver on all dGPU laptops caused LCD EDIDs to be written to garbage, bricking every dGPU-only laptop's screen if on Windows 10; requiring a new panel or a flashing of a properly working EDID using special hardware to fix... and then it could be re-broken right after), Microsoft isn't doing ANYTHING to help you, and neither is nVidia. Despite WHQL being something WE pay for in Windows where it's not supposed to screw with a system by itself if it's certified, and if it does, somebody has to pay damages. Why do you think Windows is so much cheaper now than it was with W7? Their support and guarantees have taken a huge drop off.