DX 12 performance increase comes at a cost

[That Norwegian Guy]

Because AMD has everything included on-board, while the Nvidia GTX series can't do much async compute without loading the CPU more.

 

As it turns out, this means a GPU power consumption increase on AMD and a CPU power consumption increase on Nvidia. Shocking, I know.

 

So just to address this:

37 minutes ago, kurahk7 said:

I wonder if this is an early driver issue

It's working as intended. Nvidia will do the same once they have the same level of async compute capability AMD has. Which may or may not be Pascal.

[Sharkyx1]

27 minutes ago, don_svetlio said:

A power Virus should force 250W through it. A game using it to the max should be using rated power as minimum and then based on OC anywhere from 200 to 225W

Some games utilize different parts of the card differently

if the card is hitting its max boost bin and only using 150w then its not like it can consume more.

In normal testing TH found that same 390X used 294w and the 980 used 175 in their individual reviews

Spoiler

 

[don_svetlio]

The issue here is that the 390X has a reworked power delivery so it should be lower than a 290X, not higher, as every other reviewer seems to believe.

[Sharkyx1]

10 minutes ago, don_svetlio said:

The issue here is that the 390X has a reworked power delivery so it should be lower than a 290X, not higher, as every other reviewer seems to believe.

Its a custom board, MSI could have made a board that will deliver more power. in toms original review of that MSI model it was about 50w more and techpowerup found an increase of 96watts over a reference 290X in average and 76 in peak readings

 

the problem is that low 222w reading when all other reviews are showing closer to the 330 number

[don_svetlio]

Just now, Sharkyx1 said:

Its a custom board, MSI could have made a board that will deliver more power. in toms original review of that MSI model it was about 50w more and techpowerup found an increase of 96watts over a reference 290X in average and 76 in peak readings

Jayz reviewed the same one but said the opposite.

[Doobeedoo]

It may depends some on drivers though hardware works more fully on new DX so. There will be further optimizations so it will all be closer.

[NateGSR117]

At the very least it nice to know for a fact that it a thing, although it not really a surprise.

[GoodBytes]

You know, some people pay extra to have heated seats, or buy a room heater, and you complain... Shiiish, what world we came into.

[patrickjp93]

2 hours ago, don_svetlio said:

How is the 980 using less power than it's rated for ? I find this awfully conspicuous.

TDP =/= power consumption. For AMD it usually underestimates how much is used. For Nvidia it is usually dead on or slightly higher.

[don_svetlio]

6 minutes ago, patrickjp93 said:

TDP =/= power consumption. For AMD it usually underestimates how much is used. For Nvidia it is usually dead on or slightly higher.

Odd you should say it underestimates considering AMD rate TDP with +20% based on the power target so a stock card uses 20% less than rated TDP more or less.

Source: http://www.guru3d.com/articles-pages/amd-radeon-hd-7850-and-7870-review,6.html

[PlayStation 2]

[Curufinwe_wins]

4 hours ago, don_svetlio said:

Jayz reviewed the same one but said the opposite.

Put it this way...  Given that both the fury x and 980ti use fairly similar power pre-async it stands to reason properly that the card that takes full advantage  of async would use more power under said loading while the other (which notably basically can't use it at all) is unaffected. 

 

All the cards here seem to be less than super peak (or even tdp)  utilization under dx11 which suggests to me that AoS  doesn't hit the gpu nearly as efficiently as it could. 

 

We all already knew Hawaii preformed particularly poor under power viruses (Furmark spikes to 500W), so perhaps the load being attached by a sync similarly stresses (to a lesser extent obviously) the gpu. 

[Trik'Stari]

They forgot the whole "forced onto a specific operating system" drawback.

 

Although I guess that's always applied to Direct X

[Paranoid Kami]

Desktop users don't care about power consumption. This only affects laptop users who don't plug it in a wall when they game.

[Big-G]

We are all being lied to...

 

j/k...

[Kinda Bottlenecked]

6 hours ago, don_svetlio said:

Jayz reviewed the same one but said the opposite.

 

It's clocked at 1100mhz though. If it's higher than stock power consumption, thats quite reasonable 

[Mew]

6 hours ago, That Norwegian Guy said:

Because AMD has everything included on-board, while the Nvidia GTX series can't do much async compute without loading the CPU more.

 

As it turns out, this means a GPU power consumption increase on AMD and a CPU power consumption increase on Nvidia. Shocking, I know.

 

So just to address this:

It's working as intended. Nvidia will do the same once they have the same level of async compute capability AMD has. Which may or may not be Pascal.

But the Fury X's power consumption doesn't shoot up as much as the 390x did. 

[Belgarathian]

I knew I bought my 1200W PSU for a reason. Maybe now my config will actually make it to the sweet spot. 

[BobbyG2]

10 hours ago, Nena360 said:

Not that it is a big deal but was it not meant to make the hardware more efficient use its power?

Supposed to, at the same settings and framerate.

[Humbug]

the advantage of these new gen APIs like Vulkan and DX12 is that (when implemented properly) they are able to do a much better job of keeping the GPU fed with work under adverse conditions. In the case of AOTS benchmark we are looking at an extreme scenario with loads of stuff going on on-screen which under old APIs like DX11 results in a CPU bottleneck hence less load on the GPU. But CPU power consumption would certainly have been higher under DX11.

[Yoinkerman]

Power is cheap bring it on

[zMeul]

14 hours ago, patrickjp93 said:

TDP =/= power consumption. For AMD it usually underestimates how much is used. For Nvidia it is usually dead on or slightly higher.

I don't know how AMD rates their GPUs, but nVidia states the power consumption for the video card: http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-980/specifications

 

---

 

I see that AMD fanboys still can't understand that 390(X) isn't anything more than a rebranded 290(X)

the "power improvements" are not at HW level, but at driver level - and they've done it by downclocking based on workload going trough the GPU - and that caused an effect where it creates uneven performance in games; see:

 

---

 

back on the topic at hand

when nVidia removed the possibility of overclocking notebook GPUs I said that this might be the reason - specifically: power usage increase in DX12 games

[Stefan1024]

That was expected. I saw a presentation of a software developer talking about DX12 and vulcan about 3 month ago. I allways said the goal is to use all resouces and produce as much heat as possible, as the equals to the GPU does as much as it can.
This is the resaon I tested my passive cooling solution with furmark. It's not a power virus, it's how a perfectly optimized game should run.

[DXMember]

daymm... who would of thought that utilizing all that hardware would cause power consumption to go up... we all thought GPUs are made of pixie dust and magnets, right?

 

the article is good and very much needed, but I don't think this has to be so sensational

should put more emphasis on the fact that hardware is being utilized better than the fact that power consumption increased

 

also goes to show the gutted maxwell architecture

[Humbug]

The CPU power consumption for a given workload will be down though.