Nvidia Claims 3X Improved Performance In Certain Applications With New Driver Update for Titan Xp (Vega Response?)

[sgloux3470]

8 hours ago, RagnarokDel said:

 

If Nvidia actually supported freesync, it would prevent a lot of non-cryptomining market share loss to AMD.

Nvidia wants you to buy an expensive GSYNC monitor and then continue to buy Nvidia GPU's for the lifespan of that monitor instead.

 

Come upgrade time those of us with these monitors get to decide if we want to buy the video card we'd prefer or the one compatible with our $600+ monitors.

 

 

 

 

[Sniperfox47]

23 minutes ago, Morgan MLGman said:

Huh, feature level? Then how can one say that their GPUs "fully" support DX12 if they don't support all of its features?

Again, because DirectX 12 is not in any way related to feature level 12 other than that the specs both released at roughly the same time.

 

Feature level 12 is a DX10/11 thing, as much as a DX 12 thing. Pascal supports all of the mandatory features for feature level 12_0 and 12_1, and all of the optional features other than "Stencil reference value from Pixel Shader" and Tier-2 of the resource heap capabilities. The non-titan cards also don't support reduced precision (16-bit and 10-bit) shaders, which again is an optional feature in 11_0, 11_1, 12_0, and 12_1.

 

There are no cards that support everything from feature level 12_1, and by the time there is, there will very likely be a feature level 12_2 specifically because feature levels are designed to be a moving goal for hardware developers. But more importantly there's no reason for a desktop graphics card to support everything optional from 12_1 or even 12_0 since a number of the features in those levels will likely never see use outside of mobile.

 

 

[zMeul]

15 hours ago, Max_Settings said:

This is only for the actual Titan Xp, not the Titan X (pascal). And it's not in games, only certain professional applications.

it's for both:

• Titan X
• Titan X_p

they basically enabled Quadro functionality on Titans

[Valentyn]

Looks like the Titan X P also got decent increases!

User at HardOCP tested with the latest drivers.

https://hardforum.com/threads/vega-rumors.1933112/page-86#post-1043139625

[The Benjamins]

5 minutes ago, Valentyn said:

Looks like the Titan X P also got decent increases!

User at HardOCP tested with the latest drivers.

https://hardforum.com/threads/vega-rumors.1933112/page-86#post-1043139625

these numbers defiantly looks like a response to Vega FE seeing how they are now relativity even now.

 

lol that 600% gain.

[System Error Message]

12 hours ago, LAwLz said:

Hahah yeah, fuck Nvidia for releasing drivers that increase performance, right?

I hate when my graphics cards gets better with newer drivers. They really should stop doing that.

/sarcasm (you'd think that it wouldn't be needed, but I wouldn't be surprised if some AMD fans genuinely thought this way)

 

 

Maybe I am misinterpreting you, but it sounds like you've already made your mind up on what you're going to write in your review, before you even have the cards so you can test them. Having predefined conclusions is a very bad thing to have. It will inevitably lead to unconscious bias (because you want to verify that you're correct, instead of looking at evidence and then make your mind up).

no, its my initial hypothesis. If nvidia proves better than so be it. Nvidia cuts a lot from their gtx line even their titans but AMD only nerfs the double precision performance of their consumer cards so AMD is looking good at the moment as far as some architectural nerfs go. fp16 for instance is nerfed on the gtx line which some games and graphics can use to reduce memory and processing use.

 

However my test isnt a performance test of games, rather the professional applications that are useful to me, and some architectural benchmarks to show some of the numbers not shown (like int8, fp16), and if i have time write a test code to test other instructions.

 

Im hoping that AMD will do well this time because it would help competition and keep the prices in check. Because AMD didnt have anything decent nvidia were able to sell their current line more with no benefit especially the titans, so this test is to compare the titan to regular gtx, to AMD's equivalent and to an older AMD GPU that doesnt have any architectural nerfs or cuts to it.the only thing the pascal titan has over its geforce are more ROPs, TMUs, L2 cache and 1GB more vram but still retains the various nerfs like lowered fp16 performance, half duplex bus (which can reduce performance for nvidia optimus and eGPU solutions especially at higher resolutions).

 

Right now im just waiting for my AMD GPU to arrive and setting up the test. Its not going to be a great test as im using the GPUs externally but i will try to cover as many common professional applications. Currently 1 major loss for nvidia is not giving the same features for the titan xp to the titan x pascal as its still a workstation card as there are still plenty of workstations with titan x pascals out there. Its only been a year.

[vanished]

Awfully suspicious if you ask me.  I mean, I suppose it's possible that they discovered a flaw and were able to legitimately increase performance that much, but I feel like it's more likely they knew how to do this all along and were just intentionally holding it back.

[System Error Message]

It looks like nvidia was holding back performance for professional applications, they did so in the past so you'd buy a quadro. Some applications in the past would only accelerate on quadro or tesla and not accept geforce.

 

none of the softwares however seem relevant to me. Blender and openCL programming are my main.

[RagnarokDel]

2 hours ago, sgloux3470 said:

Nvidia wants you to buy an expensive GSYNC monitor and then continue to buy Nvidia GPU's for the lifespan of that monitor instead.

 

Come upgrade time those of us with these monitors get to decide if we want to buy the video card we'd prefer or the one compatible with our $600+ monitors.

 

 

 

 

Nvidia is like if phone carriers sold GPUs.

[vanished]

Just now, RagnarokDel said:

Nvidia is like if phone carriers sold GPUs.

Nah, not nearly that bad.  To truly match phone carriers, they'd have to also charge a subscription fee, force you to install various other software on your PC, disable features of it if you don't pay extra, update drivers to be compatible with new games only 8 months after their release, and charge you a fee every time you change cards.

[zMeul]

some nVidia cards got better results with new drivers - fuck that, boo nVidia

some Radeon cards get better results with new drivers - AMD FineWine^TM

 

because logic

[ravenshrike]

9 minutes ago, zMeul said:

some nVidia cards got better results with new drivers - fuck that, boo nVidia

some Radeon cards get better results with new drivers - AMD FineWine^TM

 

because logic

The difference being that Finewine is from driver optimization. This is from nVidia ungimping performance that they purposefully gimped in the first place.

[leadeater]

3 hours ago, zMeul said:

it's for both:

• Titan X
• Titan X_p

they basically enabled Quadro functionality on Titans

More like re-enabled what the Titans had which got removed and is now back . It's good that the card actually is what it's marketed as now though.

[vanished]

2 hours ago, ravenshrike said:

The difference being that Finewine is from driver optimization. This is from nVidia ungimping performance that they purposefully gimped in the first place.

It does sure seem that way 

[3DOSH]

Imagine a world were Nvidia is the only company making GPU's. Do you think they would have enabled fp16 on pro-consumer card ? or they would they have kept shafting people to get the Quadros, you answer that. Btw they can enable it on all high end pascal gaming cards snice its the same die but i doubt they well since Volta is near, so why enable feature that you can charge more for later down the line.

[mr moose]

50 minutes ago, 3DOSH said:

Imagine a world were Nvidia is the only company making GPU's. Do you think they would have enabled fp16 on pro-consumer card ? or they would they have kept shafting people to get the Quadros, you answer that. Btw they can enable it on all high end pascal gaming cards snice its the same die but i doubt they well since Volta is near, so why enable feature that you can charge more for later down the line.

 

I think many people don't understand just how much profit companies the size of nvidia have to make to remain viable.   To the average consumer billions in profit looks ugly and they feel it is only generated through unmitigated greed.  But the reality is that nearly all companies need massive profits in order to assure viability long term.  AMD is very lucky they didn't go bankrupt or get bought out in the last few years.   Disabling or delaying features in a product is not greed, it is essentially product revenue management.   What happens if you give your customers a product they don't need to upgrade in the next 5 years?  A. you don't have any revenue for the next 5 years. 

 

Competition is good for customers, don't get me wrong, but just don't make the mistake of assuming unnecessary greed is the motive behind a companies (any company) decisions.

 

 

 

 

[raphidy]

Wow, that ungimping drivers. Thanks AMD. Hopes RX Vega does some damage in the consumer cards

[ARikozuM]

12 hours ago, TidaLWaveZ said:

I wish AMD would stand a little bit closer so their breath would be more noticeable.

If Nvidia doesn't notice the fan on Vega FE, I'd be very surprised. 

[ravenshrike]

10 minutes ago, rrubberr said:

So to clarify, there is no point updating the driver if you're using older Titans? When I search for Titan Z drivers on Nvidia's website, it shows this one as compatible. Why did they include the whole lineup of cards in the update compativility if it was just meant for XP?

Because there's probably other minor driver updates with it. The performance boost itself however is strictly the newer cards.

[Polakki]

That's real f... Finewine!

[DocSwag]

11 hours ago, Billy_Mays said:

Yeah that's the problem AMD should have made it lots more better than a Titian X so then Nvidia would start shitting bricks

But is that really worth it? As far as I can tell Vega doesn't have fp64 hardware on board which means with Vega as the architecture you would probably be looking at a 650 mm^2+ die in order to beat the 1080 ti, not to mention there's extra costs from interposer and stuff which would probably make Vega expensive af. The top end gpu would probably be in the range of $900 at least. Who is gonna want to pay that?

[DocSwag]

18 hours ago, ravenshrike said:

As soon as they saw that AMD had their head screwed on straight again with Ryzen and how Intel got suckerpunched nVidia started to move. That's why that MCM whitepaper with MIT came out recently. They do not want to play the part of Intel. As it is, assuming Navi is a success in its MCM implementation similar to Ryzen then you should see AMD able to saturate the GPU markets from low to high end with little resistance. IPC is a much lesser concern on GPUs. nVidia has started to move now so they're not left out in the cold for 3+ years. However they have got to be pumping a shitload of money into that particular R&D program if they want a successful implementation taped out within the next 2 years.

Nvidia has been moving for years. Maxwell, Pascal, that stuff has probably been in development since 2011.

 

You also have to remember doing an MCM implementation is not perfect. Latency sees a jump due to having more stuff to jump through. We may see inconsistency, just like how we sometimes do on Ryzen and epyc.

[Jito463]

12 hours ago, Max_Settings said:

almost another doubling in performance like we saw from Maxwell to Pascal.

That seems....high.  Do you have some benchmarks from then to show this?  I don't recall the performance jump being that vast, unless you're referring to professional applications, not games.

[leadeater]

1 hour ago, DocSwag said:

Nvidia has been moving for years. Maxwell, Pascal, that stuff has probably been in development since 2011.

 

You also have to remember doing an MCM implementation is not perfect. Latency sees a jump due to having more stuff to jump through. We may see inconsistency, just like how we sometimes do on Ryzen and epyc.

And a GPU needs a shit load more bandwidth than the 30GB/s in Zen.

[sgloux3470]

8 hours ago, zMeul said:

some nVidia cards got better results with new drivers - fuck that, boo nVidia

some Radeon cards get better results with new drivers - AMD FineWine^TM

 

because logic

There's a strong perception that Nvidia was just sitting on this update until they had a reason to release it.  Most likely it's optimizations that were already on the Quadro line that they are happening to bequeath to the Titan card.  

 

Whether that's true or not is only something Nvidia knows.