Nvidia’s 337.5 ‘Wonder Driver’ Raises Eyebrows – Numbers in the Official Slides are only Selectively

[Gregster]

When Nvidia showed us official slides for its upcoming DX 11 ‘Optimization’ Drivers we saw massive performance increase all over the board in nearly all situations. Now of course, Nvidia’s 337.5 Beta Driver was meant to reduce CPU-Overhead so when we saw that it gave a frames per second increase in Single GPU solutions too, we were pretty impressed. However it turns out that the numbers portrayed, although ‘technically’ true, don’t really apply with the same magnitude to your average scenarios.
NVIDIA GeForce 337.50 BETA DriverImage courtesy of VideocardZ.com
Nvidia 337.5 Beta Drivers Give You a Massive 71% increase – Just throw in another GPU
To put it bluntly, Nvidia’s official slides are a hair short of being bad faith. See, you do get a 71% increase with these drivers in Rome Total War, as compared to the 335 Driver set. But thats only because the 335 Drivers did not support SLI, and the 337.5 Beta Drivers do. Don’t get me wrong the new drivers do give you a significant increase, but they do so very selectively and Nvidia conveniently failed to mention that. If you play Metro 2033, you will see a significant increase in FPS, but if you are for eg, playing Crysis 3, do not expect much of an increase.
Advertisements


We saw the same thing when the official numbers from Mantle were released, although to be fair to AMD, the exaggeration was no where near this extreme. The thing is Driver overhead reducing drivers help the most where the situation is CPU-Bound enough. So with these Nvidia drivers, if you have a middle – high order single GPU don’t expect much unless you are playing very select games. If you however have a multi-GPU configuration then you should get these drivers asap and expect a 10 – 15% improvement across the board. Which is a somewhat decent increase. Long story short it is a very good update, but just like the Initial Mantle Hype, is sadly over rated. I have an intense distaste of posts like this, but news like these are too important to not be covered.


Read more: http://wccftech.com/nvidias-3375-wo...bers-official-selectively-true/#ixzz2ycGNa2BI

As an SLI user, I have proper seen the gains but looking around at single card users, a little disappointing for them. Still, it is an early beta of the new branch and it could well be the starting point of things to come

 

[setter]

I noticed good gains in bf4, crysis 3 though was pretty much the same. But as it's a beta hopefully we can see bigger improvements as this new driver revision matures.

 

[Kaapstad]

As an SLI user, I have proper seen the gains but looking around at single card users, a little disappointing for them. Still, it is an early beta of the new branch and it could well be the starting point of things to come

I never, never, never take anything seriously that the above website writes and even now they still don't know what they are talking about.

As a driver for game optimisation 337.5 is pretty average, TR in quad sli actually runs worse than it did on older drivers as there is no CPU bottleneck for 337.5 to help with. Give it a game like Sleeping Dogs and 4 cards and you get gains over 40%.

As for RTW2, NVidia could have quoted the gains you get on a quad sli setup (about 300%) but they did not.

Edit

I am just off to check out MLL and Batman on 4 GPUs.

 

[Gregster]

Very true Kaap but thought it worth mentioning, as I seem to post a fair chunk of wccftechs stuff

 

[Kaapstad]

Very true Kaap but thought it worth mentioning, as I seem to post a fair chunk of wccftechs stuff

Metro seems to run nicer on these new drivers, before it was just total rubbish.

 

[Gregster]

Metro seems to run nicer on these new drivers, before it was just total rubbish.

I need to fire it up. I have only played it for about an hour but it does look decent.

 

[triss]

Free gains are nice even if there small , I dont think anyone truly believes either sides marketing dpet when they put out we got 60% plus stuff.
Its always under exactly the best bit of the game they found with the ideal setup to get a gain

 

[drunkenmaster]

Nvidia claim wonder drivers that make Mantle pointless and coincidentally therefore DX12... while also talking up DX12. Numbers come out and Nvidia's claims are instantly rubbished completely. Noteworthy gain from enabling SLI in one game where SLi failed to work before, outside of that effectively none of their claimed numbers are even slightly true.

Round up from various places hints at 10% maybe in a few games, absolutely not general DX11 but 3 very specific games, two that have Mantle in them and they've clearly just put lots of extra work into optimising because supposed DX11 optimisation doesn't translate to other games.

Sound like BS when they claimed it and sounds even worse now. It was a "Mantle isn't that good" publicity drive, then not delivering anything remotely close to what they promised.

It's funny that Nvidia feel the need to talk up DX12 as game changing and brilliant, but a couple weeks later try to tell everyone they can do most of it in DX11 with magical drivers in a way they simply never bothered to do before. Turns out to be complete and utter BS.

It was nothing more than a standard driver update, thing is that is fine, but why the insane dishonest PR campaign around them. The only people to look stupid after this driver update is Nvidia. Why not say "we've got a good driver coming, and when DX12 is around we'll be awesome at it.

This comes across as nothing but lying and shines a light on Mantle. Nvidia try in media to say Mantle is a bit crap and they can do the same thing with a new driver.... turns out they can't do remotely close to the same thing in driver, so Mantle is better than a driver update they can do to DX11, which really just highlights that they won't match Mantle till DX12 is available(maybe not even then). It's shining a light on how DX11 will simply not match Mantle and purely because Nvidia had to open their dishonest trap for no reason at all.

 

[Gregster]

I take it you didn't see the gains that Me, PGI and Whyscotty posted then DM?

[convenient]image[/convenient]

http://forums.overclockers.co.uk/showthread.php?t=18591985

If you missed them, there they are

 

[Silent_Scone]

I've not used them so ignore the below, i'm just expressing thoughts from within DM land

Ok noted. There are currently two games using Mantle Drunkenmaster. One of which has a bench thread. I suggest you venture into it and look at the table as it currently stands.


Single card gains are very small agreed, but the gains were going to be from CPU limited scenarios which you see a lot more predominantly with SLi setups. I've seen some great gains spanning over three cards.

 

[Kaapstad]

Nvidia claim wonder drivers that make Mantle pointless and coincidentally therefore DX12... while also talking up DX12. Numbers come out and Nvidia's claims are instantly rubbished completely. Noteworthy gain from enabling SLI in one game where SLi failed to work before, outside of that effectively none of their claimed numbers are even slightly true.

Round up from various places hints at 10% maybe in a few games, absolutely not general DX11 but 3 very specific games, two that have Mantle in them and they've clearly just put lots of extra work into optimising because supposed DX11 optimisation doesn't translate to other games.

Sound like BS when they claimed it and sounds even worse now. It was a "Mantle isn't that good" publicity drive, then not delivering anything remotely close to what they promised.

It's funny that Nvidia feel the need to talk up DX12 as game changing and brilliant, but a couple weeks later try to tell everyone they can do most of it in DX11 with magical drivers in a way they simply never bothered to do before. Turns out to be complete and utter BS.

It was nothing more than a standard driver update, thing is that is fine, but why the insane dishonest PR campaign around them. The only people to look stupid after this driver update is Nvidia. Why not say "we've got a good driver coming, and when DX12 is around we'll be awesome at it.

This comes across as nothing but lying and shines a light on Mantle. Nvidia try in media to say Mantle is a bit crap and they can do the same thing with a new driver.... turns out they can't do remotely close to the same thing in driver, so Mantle is better than a driver update they can do to DX11, which really just highlights that they won't match Mantle till DX12 is available(maybe not even then). It's shining a light on how DX11 will simply not match Mantle and purely because Nvidia had to open their dishonest trap for no reason at all.

I don't think your sources have tested quad sli

Here is the link to the Sleeping Dogs Bench, before 337.5 my 4 Titans were struggling to beat the 4 290Xs, after 337.5 it is so one sided it is embarrassing. The new drivers don't give big gains on everything but where there was a bottleneck the difference can be quite large.

http://forums.overclockers.co.uk/showthread.php?t=18536130

Sleeping Dogs is a gaming evolved game too.

 

[Rroff]

Nvidia claim wonder drivers that make Mantle pointless and coincidentally therefore DX12... while also talking up DX12. Numbers come out and Nvidia's claims are instantly rubbished completely. Noteworthy gain from enabling SLI in one game where SLi failed to work before, outside of that effectively none of their claimed numbers are even slightly true.

Round up from various places hints at 10% maybe in a few games, absolutely not general DX11 but 3 very specific games, two that have Mantle in them and they've clearly just put lots of extra work into optimising because supposed DX11 optimisation doesn't translate to other games.

Sound like BS when they claimed it and sounds even worse now. It was a "Mantle isn't that good" publicity drive, then not delivering anything remotely close to what they promised.

It's funny that Nvidia feel the need to talk up DX12 as game changing and brilliant, but a couple weeks later try to tell everyone they can do most of it in DX11 with magical drivers in a way they simply never bothered to do before. Turns out to be complete and utter BS.

It was nothing more than a standard driver update, thing is that is fine, but why the insane dishonest PR campaign around them. The only people to look stupid after this driver update is Nvidia. Why not say "we've got a good driver coming, and when DX12 is around we'll be awesome at it.

This comes across as nothing but lying and shines a light on Mantle. Nvidia try in media to say Mantle is a bit crap and they can do the same thing with a new driver.... turns out they can't do remotely close to the same thing in driver, so Mantle is better than a driver update they can do to DX11, which really just highlights that they won't match Mantle till DX12 is available(maybe not even then). It's shining a light on how DX11 will simply not match Mantle and purely because Nvidia had to open their dishonest trap for no reason at all.

These drivers are certainly not just another driver update - while they are never going to be comparable to mantle and aren't intended to be they do stabalise performance by a marked degree in many games and increase performance by a moderate amount in many CPU or API limited circumstances.

 

[PhysicsMan94]

Nvidia claim wonder drivers that make Mantle pointless and coincidentally therefore DX12... while also talking up DX12. Numbers come out and Nvidia's claims are instantly rubbished completely. Noteworthy gain from enabling SLI in one game where SLi failed to work before, outside of that effectively none of their claimed numbers are even slightly true.

Round up from various places hints at 10% maybe in a few games, absolutely not general DX11 but 3 very specific games, two that have Mantle in them and they've clearly just put lots of extra work into optimising because supposed DX11 optimisation doesn't translate to other games.

Sound like BS when they claimed it and sounds even worse now. It was a "Mantle isn't that good" publicity drive, then not delivering anything remotely close to what they promised.

It's funny that Nvidia feel the need to talk up DX12 as game changing and brilliant, but a couple weeks later try to tell everyone they can do most of it in DX11 with magical drivers in a way they simply never bothered to do before. Turns out to be complete and utter BS.

It was nothing more than a standard driver update, thing is that is fine, but why the insane dishonest PR campaign around them. The only people to look stupid after this driver update is Nvidia. Why not say "we've got a good driver coming, and when DX12 is around we'll be awesome at it.

This comes across as nothing but lying and shines a light on Mantle. Nvidia try in media to say Mantle is a bit crap and they can do the same thing with a new driver.... turns out they can't do remotely close to the same thing in driver, so Mantle is better than a driver update they can do to DX11, which really just highlights that they won't match Mantle till DX12 is available(maybe not even then). It's shining a light on how DX11 will simply not match Mantle and purely because Nvidia had to open their dishonest trap for no reason at all.

Of course these latest drivers were not supposed to be a competitor to Mantle and Nvidia never implied as such. Mantle is a whole new graphics API built from a very low-level basis, if a new set of DX11 based Nvidia drivers had superseded the performance gains given by Mantle then something would be awry and that would be incredibly embarrassing for AMD.

Indeed, that didn't happen and I don't think anyone was realistically expecting that to happen. You seem to be viewing this from a flawed standpoint, it's not surprising that one would find the gains from this latest batch of drivers lacklustre if they expected Mantle level gains (which as I emphasised earlier, nobody projected nor implied) - it's all relative to what you are/were expecting. And with that said, from the gains I have seen people post (haven't installed them yet myself) I am impressed - particularly with the SLI results, not so much with single card systems but there's still some nice gains anyway.

Thus, given these fairly impressive gains relative to previous driver performance improvements it will be interesting to see the kind of gains we shall see with DX12 as it is developed and enhanced further within the next few years, as remember that these gains were achieved just using DX11 and, I suspect, by reducing function calls to particularly slow structured within DX by replacing the functionality of these methods with inline functions adapted from the DX11 code by Nvidia.

 

[drunkenmaster]

Ok noted. There are currently two games using Mantle Drunkenmaster. One of which has a bench thread. I suggest you venture into it and look at the table as it currently stands.


Single card gains are very small agreed, but the gains were going to be from CPU limited scenarios which you see a lot more predominantly with SLi setups. I've seen some great gains spanning over three cards.

Really because Nvidia kept saying massive gains across the board for single CPU.

I know where the gains come from and why and have no problem with them, it's Nvidia who were banging on about massive gains in all scenarios.

These drivers are certainly not just another driver update - while they are never going to be comparable to mantle and aren't intended to be they do stabalise performance by a marked degree in many games and increase performance by a moderate amount in many CPU or API limited circumstances.

Of course these latest drivers were not supposed to be a competitor to Mantle and Nvidia never implied as such. Mantle is a whole new graphics API built from a very low-level basis, if a new set of DX11 based Nvidia drivers had superseded the performance gains given by Mantle then something would be awry and that would be incredibly embarrassing for AMD.

Greg posts the news of new awesome drivers with the title of his post being

DX11 could rival Mantle

http://forums.overclockers.co.uk/showpost.php?p=26048430&postcount=1123

During a handful of conversations with NVIDIA on DirectX and Mantle, there was a tone from some that leaned towards anger, but hints at annoyance. It’s possible, according to NVIDIA’s performance improvements in DX11 efficiency shown here, that AMD could have accomplished the same thing without pushing a new API ahead of DirectX 12.

It's possible to get the same effects of Mantle in DX11 without doing a new API according to Nvidia.

So are they saying that some low level features already existed in DirectX 11 and that NVidia will release a driver exposing that performance?

In a nutshell, yes. They used the Star Swarm demo on the new drivers and it showed the same shift in performance that Mantle did. Thief is another one.

I don't see how Mantle will be able to compete if NVidia and Microsoft have a magic switch to enable Mantle-like performance improvements on a firmly entrenched DirectX 11, it sounds too good to be true tbh!

Hmm, nope Nvidia weren't talking BS, comparing Mantle saying it can be done in DX11 and multiple Nvidia guys on the forum weren't banging on and on about how Nvidia will now match Mantle in DX11 because they'll just use low level stuff already in DX11. Nvidia users on this forum were all over talking up low level api like features, Mantle comparisons, Nvidia were talking about Mantle comparisons so yes, the Mantle comparison DID happen was implied by everyone who posted similar such nonsense.

AS for results I missed, Nvidia fixed some SLI areas, wow, well done. But Nvidia is claiming 64% in single gpu and 71% in sli. Almost exclusively the gains are in SLI, only one that I've seen is remotely close to what is gained.

The issue isn't if there are or aren't any performance increases. It was the die hards banging on about Mantle being beaten, Nvidia themselves consistently talking about Mantle like improvements in DX11 and Nvidia's completely bogus PR campaign.

They improved quad sli scaling, woo for supposedly worse drivers that scale worse with CPU overhead, why were sli titans ever behind 290's? So these certainly aren't "just another driver update" except an average driver update will improve a few games by 5-10%, check, screw up one or two games, check, and fix SLI profiles for a bunch of games, check....... but they aren't just another driver update because the 4 people on the forums who benchmark 4 cards and don't game saw bigger general increases. No offence to Kaap but me and 95% of people don't care about performance of 4 cards in benchmarking. Nvidia marketed these as Mantle equalling, most of the Nvidia guys here said the same, and 98%+ of AMD and Nvidia users run single gpu setups, Nvidia didn't come close to Mantle like improvements, they were across the board like claimed and in single GPU basically nothing changed.

Single gpu, Mantle is game changing, this set of drivers for single gpu's was absolutely nothing different to any other set. Whack in more than average SLI fixes and you have what the driver consists of.


http://forums.overclockers.co.uk/showpost.php?p=26051415&postcount=1260

another gregster post quoting an article which says

While showing the driver to driver improvement, these benchmarks also show something very very interesting. They show AMD’s Mantle API faring worse than NVIDIA counterparts, on DX11. That’s right, not DX12 but DX11 through driver updates.
10151614_10152287088300049_1135428562_n

Upcoming NVIDIA Driver to Deliver Faster Performance Than Mantle API in Thief and More

They also showed this, Nvidia that is, by showing bogus AMD numbers in both DX11 and Mantle in Thief in order to "beat" them. Don't forget that almost all the slides shown with the initial "epic dx11 drivers beat Mantle) information were SINGLE cards. Nvidia absolutely 100% implied on multiple occasions that Mantle like performance was doable through DX11 optimisations, they think they can beat Mantle, they can do it single card. Multiple forum users said the same, posted this information and talked about it like the performance difference would be epic.

I think it's incredibly disingenuous to pretend Nvidia didn't imply it was bringing Mantle like performance to DX11 because they specifically compared them, they talked about it repeatedly and multiple of the loudest Nvidia users said they expected it. TO turn around a month later when the drivers turned out to be completely bogus, and pretend this was never about competing with Mantle is laughable.

 

[PhysicsMan94]

Greg may have implied that DM, but as far as I'm aware Greg does not represent Nvidia nor does he work for them. Until you can find me suitable statements from Nvidia themselves saying that these drivers would/will supersede Mantle performance gains, then I will just assume that you are making stuff up to suit your own agenda.

Quite frankly I am not concerned by what other members of the forum might or might not have said, I am more interested in what Nvidia, Microsoft or any other involved party said about these drivers - and I care not for quote mining or taking statements out of context, I have enough debates with creationist crackpots to see through that nonsense.

Essentially, I stand by what I said - from my observations, this set of drivers (beta mind you) seem to be very successful, especially for SLI users, and that's what really matters when it comes down to it. Hopefully such optimisations are indicative of the future for Nvidia drivers.

 

[Kaapstad]

No offence to Kaap but me and 95% of people don't care about performance of 4 cards in benchmarking.

That is the comical part, for benching these drivers are not that good for Heaven 4 and Firestrike if anything they are noticeably worse than 320.18 with 4 cards. All the gains I have seen (apart from a nice boost in 3dmark11) are in games.

 

[Calin Banc]

I take it you didn't see the gains that Me, PGI and Whyscotty posted then DM?

[convenient]image[/convenient]

http://forums.overclockers.co.uk/showthread.php?t=18591985

If you missed them, there they are

You and PGI need to test that BF 4 în the same spot on SoS, where the bottlenecks are.

 

[Martini1991]

OP's just flame bait. Typical AMD fanboy.

No wait..... It's from an Nvidia owner

 

[drunkenmaster]

Greg may have implied that DM, but as far as I'm aware Greg does not represent Nvidia nor does he work for them. Until you can find me suitable statements from Nvidia themselves saying that these drivers would/will supersede Mantle performance gains, then I will just assume that you are making stuff up to suit your own agenda.

Again, most of what I linked were lines and information Greg posted from articles/previews of the driver. The bits where Nvidia said Mantle like optimisations can be done on DX11, the bits where they said cpu overhead can be drastically reduced, the bits where Nvidia claimed 64% increase on the CPU. Here's a hint, that was all NVIDIA saying and absolutely and completely obviously implying this driver set can come along and do at least some of what Mantle did.

I'm not blaming Greg, greg and multiple other Nvidia users got this impression directly from what Nvidia said on the subject. They absolutely 100% implied and kept mentioning Mantle when talking about their upcoming uber drivers.

AS for the idea that if you can raise SLI performance once, you can bring the same performance increase in the future, that isn't how life works. If you have an engine misfiring and take it to the mechanic and get it working in the scenario it wasn't working, you can't then just make it 10% faster every time you go back. Nvidia FIXED a problem with SLI, and packaged it up as an all encompassing DX11 updated driver that improves performance everywhere.

There WILL be future games that SLI is borked, and same for xfire, and there will be future games where a new game has dodgy single card performance and future drivers will fix those issues. You won't get a continued 10% faster across the board SLI driver time and time again.

This is a standard driver, with a few games getting 5-10% boost in single card, and a seemingly fairly major fix to excessive SLI overhead in Nvidia drivers. Once you fix a problem, it's not there to be fixed again.

But again if Nvidia came out and said "opps, we noticed an excessive CPU overhead bug in SLI, next drivers will be killer for SLI" I'd say good for Nvidia users.

But instead of that Nvidia firstly had a couple of their pay to give good PR sites post up multiple things about Mantle, DX11, AMD being able to do the same stuff under DX11, their epic all over performance improving driver. They made utterly unsubstantiated claims that are well beyond borderline ridiculous and just flat out lying.

 

[drunkenmaster]

That is the comical part, for benching these drivers are not that good for Heaven 4 and Firestrike if anything they are noticeably worse than 320.18 with 4 cards. All the gains I have seen (apart from a nice boost in 3dmark11) are in games.

That's just me with my gamers head on then. I have absolutely no interest at all in how fast Nvidia/AMD can get Heaven/3dmark to run. They've been screwing with drivers and IQ to improve performance for years. I haven't run any non game benchmarks in years, I should say, I can't remember the last time I ran a game benchmark either. I should say that I only check real game benchmark performance when I read gpu reviews and the like. 3dmark/heaven performance doesn't interest me in the slightest.

For me Nvidia is to be commended for working on real game improvements in SLI. But this also points to the obvious. Excessive amount of optimising for benchmarks means they have got the benchmark profiles right and nothing to gain there. While they spend time doing that they ignore the lack of sli optimisation in real games hence there was more performance to be gained.

If AMD/Nvidia spent less time optimising for 3dmark/heaven their driver teams would have more time to fix things like sli/xfire performance for games like Sleeping Dogs and giving their users real performance benefits.

I'm extremely anti non game benchmarks, because it takes focus away from driver teams optimising for games. Considering 99% of gamers who buy a gpu game rather than bench, IMHO their time should be focused on the majority usage of their gpus.