Pascal consumer = Maxwell on speed confirmed

[79wjd]

6 minutes ago, Briggsy said:

uses 61.7% less power compared to an R9 390.

 

uses 74.4% less power compared to a GTX 980.

It uses 61.7/74.4% of the power or 38.3/26.6% less power* 

[Prysin]

56 minutes ago, AluminiumTech said:

think of VEGA as big Polaris because it will just be bigger Polaris with GDDR5X or HBM2.

nah, VEGA is "polaris+".... it will have some additional efficiency gains that will make the perf per watt between maxwell and pascal. relative to performance ofcourse

[2FA]

1 hour ago, AluminiumTech said:

think of VEGA as big Polaris because it will just be bigger Polaris with GDDR5X or HBM2.

Why did you reply to me?

[HalGameGuru]

People still ignore one of the elephants in the room. AMD will always have a higher hard floor to their power draw due to the inclusion of a hardware scheduler. Big help for certain workloads, big detriment to power draw crowns, and I wonder if that extra draw and work performed on die means it also has a harder time keeping temps down? @patrickjp93 I was going to ask that of you in another thread but forgot, does the hardware scheduler put other stresses on the GPU besides just greater power draw?

 

Now independent of the scheduler the nVidia uArch is probably still a more efficient design "core" for "core". It clocks higher, and pushes a lot of brute force performance. AMD uArch for a while now has seemed to be more about torque than horsepower. Keeping revs low and moving big loads by focusing on the needs of difficult tasks over revving to 10k RPM and smoking tires.

[patrickjp93]

1 minute ago, HalGameGuru said:

People still ignore one of the elephants in the room. AMD will always have a higher hard floor to their power draw due to the inclusion of a hardware scheduler. Big help for certain workloads, big detriment to power draw crowns, and I wonder if that extra draw and work performed on die means it also has a harder time keeping temps down? @patrickjp93 I was going to ask that of you in another thread but forgot, does the hardware scheduler put other stresses on the GPU besides just greater power draw?

 

Now independent of the scheduler the nVidia uArch is probably still a more efficient design "core" for "core". It clocks higher, and pushes a lot of brute force performance. AMD uArch for a while now has seemed to be more about torque than horsepower. Keeping revs low and moving big loads by focusing on the needs of difficult tasks over revving to 10k RPM and smoking tires.

No, but AMD has other problems too, such as weaker, fewer ROPs it has to address too.

The problem with AMD's strategy is that leaves it weaker in HPC environments where the chief driving purchase metric isn't performance/$ or performance/watt. It's performance/watt/$. Nvidia's platform just provides more for your money as a GPGPU algorithm engineer. Not only do you have all the original 32-bit hardware, but you also get the dedicated 64-bit SPs, and that allows a vastly more efficient design than to have the control logic that snaps 2 32-bit SPs together into a 64-bit-capable SP for AMD. AMD is not famous for properly addressing the problems it faces. Just throw more complex hardware at the problem instead of addressing individual problems...

[AlTech]

1 minute ago, patrickjp93 said:

No, but AMD has other problems too, such as weaker, fewer ROPs it has to address too.

The problem with AMD's strategy is that leaves it weaker in HPC environments where the chief driving purchase metric isn't performance/$ or performance/watt. It's performance/watt/$. Nvidia's platform just provides more for your money as a GPGPU algorithm engineer. Not only do you have all the original 32-bit hardware, but you also get the dedicated 64-bit SPs, and that allows a vastly more efficient design than to have the control logic that snaps 2 32-bit SPs together into a 64-bit-capable SP for AMD. AMD is not famous for properly addressing the problems it faces. Just throw more complex hardware at the problem instead of addressing individual problems...

Weaker in what sense? They are better fed with memory bandwidth.

[patrickjp93]

Just now, AluminiumTech said:

Weaker in what sense? They are better fed with memory bandwidth.

Actually they've been perfectly well fed. They're actually overfed/under pressure. Fury/X was actually a classic case of a design bottleneck of GCN. The ROP:SP ratio is not optimal for the strength of ROPs AMD provides. Nvidia provides roughly double the number of ROPs for each SP, allowing it to get away with fewer, faster SPs while Delta Color Compression and other tech makes up for the lesser bandwidth relative to AMD's offerings.

[Dresta]

3 hours ago, laminutederire said:

Dude, the rx 480 already owns the 970, do your research before trying to start a flame war.

Theres a difference between owns and crushes

[laminutederire]

46 minutes ago, Dresta said:

Theres a difference between owns and crushes

It's halfway between a 970 and 980, so that counts as crushing as well in my dictionary 

[dtaflorida]

Regardless that the architecture didn't really change and Pascal is just a node shrink, I'm still impressed with Nvidia. They could have just followed the tradition of adding 10-15% more cores and end up with 10-15% more performance. But they gave us more than 10-15%.

 

Compare that to AMD, shrinking the process and adding exactly the expected 10-15% performance boost.

 

I can only imagine AMD got a real surprise when the 1080 was revealed. Even with their own Vega on the horizon.

[Autoimmunity]

10 hours ago, don_svetlio said:

There are more than a dozen games out and dozens more on the way using DX12. Vulkan is yet to arrive in full force but DX12 is here. Sorry but Nvidia's Pascal is not ready for it. Volta likely will be.

By the time DX12 and Vulkan are required, Pascal will be ancient anyway.  It's not like devs are going to force people to use DX12 immediately, do you remember how long it took for DX11 to be required?

 

Even if Pascal doesn't last once DX12 and Vulkan are mainstream it won't matter, because they won't be fast enough to run the games anyway.  By 2019 we're going to have new consoles and a 1080 will be the equivalent of a 680 today.

[3DOSH]

 

This Roadmap clearly shows Vega isn't gonna be just bigger Polaris chip. As you can see the Perf/Watt is significantly higher than Polaris, HBM is gonna account for some of it but not all of it I'm sure as its close to 1.5x-2x increase.
Polaris 11 was the main focus with Polaris "AMD stated that in interview with PCPer" whats why its scales poorly with clocks from power consumption point of view.
With Vega this shouldn't be a problem, well at least I hope.

[don_svetlio]

14 minutes ago, Autoimmunity said:

By the time DX12 and Vulkan are required, Pascal will be ancient anyway.  It's not like devs are going to force people to use DX12 immediately, do you remember how long it took for DX11 to be required?

 

Even if Pascal doesn't last once DX12 and Vulkan are mainstream it won't matter, because they won't be fast enough to run the games anyway.  By 2019 we're going to have new consoles and a 1080 will be the equivalent of a 680 today.

It took exactly 2 years for DX11 to be a big factor. How long have we had DX12? Oh look, a year. Next year it will be a big thing and people buying 480/1060s are coming from 760s/280s - those people want LONGEVITY - something Pascal does not have.

[Coaxialgamer]

11 hours ago, TrigrH said:

if I had to guess, I think it will win (since its a big 14nm die) but not on power efficiency which isn't a major issue given the target market for vega. 

We really don't know yet about efficiency.  From what I've seen,  it looks like amd may finally be ditching ancient GCN  in favor of a new architecture.  Because Polaris may be quite difficult than tahiti,  it still is GCN at it's core for better or for worse.  The weaknesses of the architecture can be reduced somewhat,  but you can't go on forever with adding features to 2011 tech.  

[Curufinwe_wins]

1 hour ago, don_svetlio said:

It took exactly 2 years for DX11 to be a big factor. How long have we had DX12? Oh look, a year. Next year it will be a big thing and people buying 480/1060s are coming from 760s/280s - those people want LONGEVITY - something Pascal does not have.

You know I'd accept that statement if Pascal didn't have longevity. But it clearly does.

[Nineshadow]

1 hour ago, don_svetlio said:

It took exactly 2 years for DX11 to be a big factor. How long have we had DX12? Oh look, a year. Next year it will be a big thing and people buying 480/1060s are coming from 760s/280s - those people want LONGEVITY - something Pascal does not have.

DX11 wasn't that huge of a change though. Especially for game devs (all hardware and api features from dx10.1 were kept and on top of those were added new features for new functionality). The main things it brought were hardware tessellation, GPGPU support and better multi threading.

 

I think it's going to take a little longer for game devs to switch to DX12. It should also be reminded that, for or a lot of (smaller) games out there, the new features it brings simply aren't necessary and using DX11 would do just the same thing quite a lot easier, with little to none difference in performance.

[Tech Inquisition]

TL:DR...

 

38 minutes ago, Curufinwe_wins said:

You know I'd accept that statement if Pascal didn't have longevity. But it clearly does.

Agreed...

[don_svetlio]

1 hour ago, Curufinwe_wins said:

You know I'd accept that statement if Pascal didn't have longevity. But it clearly does.

We shall speak again next year. Feel free to bookmark this thread

[DocSwag]

10 hours ago, Curufinwe_wins said:

 

Yea you are just blatantly wrong here. There were 0 hardware path changes to the die or anything else (other than the use of newer higher quality components).

 

There were software optimizations (quite similar to the 480x patch actually) that along with the extremely mature process meant card by card performance was MUCH more uniform but the best 390(x) cards paled in comparison to the best 290x 8GB models (consistently reaching 1.15 was fairly easy, but only MSI and XFX models could even more than occasionally go above that.)

 

The biggest single improvement was the tessellation and color compression formulas that were brought over from Tonga. These improvements did make their way to the 290(x) variants a few weeks later (mid august 2015).

 

If you read any of the anandtech article, you would know otherwise. There are quite a few changes to the front end of pascal vs Maxwell.

[KingKeith55]

1 hour ago, Tech Inquisition said:

TL:DR...

 

Agreed...

Even maxwell to a degree has some longevity (Partially because full DX12.1 support, and the fact that it seems very similar to Pascal). Something kepler did not have. Older AMD cards will if AMD decides to work on their software and not drop new drivers for 200 and 300 series cards (which thankfully they haven't). Aside from that DX12 and Vulkan has proven that a lot more performance can be squeezed out of AMD cards. Everyone wins, AMD users get equal or better performance than Nvidia cards in the future, and Nvidia owners have nothing to worry about.

 

Keep in mind, and this is for all graphics cards, Nvidia or AMD. The truth is games aren't going to get any worse performing. Only better. So as we see games get more efficient, these older cards will last longer no matter what. And lets be honest, games don't seem to get any better looking by a huge amount. Crysis still looks better than most games released recently for reference. So please, stop trying to cause a fucking panic and have people going "OH NO MY GRAPHICS CARD WON'T RUN GAMES IN 3 YEARS".

 

[fortsatt regn]

18 hours ago, PerfectTemplar said:

Evidently, it wasn't meant to compete with the 1060 either.

 

 

970* performance.

It's not 970 performance. Closer to 980 than it is 970.

[Prysin]

11 hours ago, Autoimmunity said:

By the time DX12 and Vulkan are required, Pascal will be ancient anyway.  It's not like devs are going to force people to use DX12 immediately, do you remember how long it took for DX11 to be required?

 

Even if Pascal doesn't last once DX12 and Vulkan are mainstream it won't matter, because they won't be fast enough to run the games anyway.  By 2019 we're going to have new consoles and a 1080 will be the equivalent of a 680 today.

actually, the reason DX11 took so long was because the adoption rate between XP/Vista and Win7 took forever.

 

With Win10 however the adoption rate has been unprecedented, and as such, the speed at which developers can start to target DX12 is higher. They do not have to wait 2-3 years this time for Win10 to reach enough market share for them to risk "locking" their game to DX12. I recon already in a year we will see a vast majority of games starting to be DX12 only, as the adoption rate of Win10, thus the requirement for the OS to be able to run DX12, is so high.

 

http://store.steampowered.com/hwsurvey/directx/

 

look at this. Win10 ALREADY has majority share. among the operating systems. I recon A LOT of people jumped onto the free upgrade this week as it was their last chance. So the July HwSurvey should see this market share increase even further.

 

A game developer is only concerned about "target market %". meaning how many % of the market their game can adress with the requirements outlined for the games.

 

Due to a rapid increase in adoption of more powerful hardware lately, we have seen games jump from using GT8800's in their minimum requirements, to HD 7770s and GTX 660s.. A MAJOR improvement from how it used to be. Especially the CPU requirements has improved, as more people now then ever is using a quad core CPU at minimum. Thus the adressable core count, and as such the adressable performance tier, is actually there.

 

 

[PerfectTemplar]

21 hours ago, Dresta said:

Intel started going Nvidia's way,i mean releasing a $1700 CPU with just 2 cores more and some features which many people are not even gonna use is just stupid

 

Yea after about 1 year of drivers,custom coolers it will be the 980 level or even better maybe even a 980 Ti if gaben helps us

After a YEAR of drivers and custom coolers (citation needed desperately)...

 

The GTX 1060 TODAY is 980 level of performance and is the exact same price as even the 4GB RX480 (at least in Australia.) 

[don_svetlio]

5 minutes ago, PerfectTemplar said:

After a YEAR of drivers and custom coolers (citation needed desperately)...

 

The GTX 1060 TODAY is 980 level of performance and is the exact same price as even the 4GB RX480 (at least in Australia.) 

That's weird - the people I know from Aus told me that the 1060 is ludicrously expensive compared to the 480. Can I have a link to the 1060 for 300$ AUS or less?

[Dabombinable]

43 minutes ago, Prysin said:

actually, the reason DX11 took so long was because the adoption rate between XP/Vista and Win7 took forever.

 

With Win10 however the adoption rate has been unprecedented, and as such, the speed at which developers can start to target DX12 is higher. They do not have to wait 2-3 years this time for Win10 to reach enough market share for them to risk "locking" their game to DX12. I recon already in a year we will see a vast majority of games starting to be DX12 only, as the adoption rate of Win10, thus the requirement for the OS to be able to run DX12, is so high.

 

http://store.steampowered.com/hwsurvey/directx/

 

look at this. Win10 ALREADY has majority share. among the operating systems. I recon A LOT of people jumped onto the free upgrade this week as it was their last chance. So the July HwSurvey should see this market share increase even further.

 

A game developer is only concerned about "target market %". meaning how many % of the market their game can adress with the requirements outlined for the games.

 

Due to a rapid increase in adoption of more powerful hardware lately, we have seen games jump from using GT8800's in their minimum requirements, to HD 7770s and GTX 660s.. A MAJOR improvement from how it used to be. Especially the CPU requirements has improved, as more people now then ever is using a quad core CPU at minimum. Thus the adressable core count, and as such the adressable performance tier, is actually there.

 

 

There is kind of a problem with DX12 at the moment however-its apparently a buggy mess.