• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Some bad news

Status
Not open for further replies.
Pieter3dnow said:
Why don't you take the time and stroll down the AMD youtube channel find any video that has been made about GCN. Then you might be a little wiser..
If GCN was so power hungry then why would the same GCN use less power on other cards that does make sense to you ?
Click to expand...

Really, because clock for clock, RX Vega 64 and Fury X perform similar that suggest there is no architectural change (shown in [H] comparison), and which tells me the power saving is most likely result from going to 28nm to 15nm rather than any architectural changes.

Feel free to post the video from AMD youtube channel, because I am not doing the work for you.

Edit, meant to type 14nm instead of 15nm.
 
Last edited:
Chimpee said:
Really, because clock for clock, RX Vega 64 and Fury X perform similar that suggest there is no architectural change (shown in [H] comparison), and which tells me the power saving is most likely result from going to 28nm to 15nm rather than any architectural changes.

Feel free to post the video from AMD youtube channel, because I am not doing the work for you.
Click to expand...
Except the changes made to allow the chip to clock much higher?
 
Chimpee said:
Wouldn't that result from the extra transistor and pipeline change for a much higher clock?
Click to expand...

You are correct.. Vega has already been proven as a marketing scam even by this very site. =)
 
Digital Viper-X- said:
Except the changes made to allow the chip to clock much higher?
Click to expand...

Yes it can get higher clocks, this was due to adding pipeline stages to make the pipeline longer which uses transistors.

Over all though, a Vega 64 liquid cooled uses 300 watts, 25 watts more than a FuryX and gets 38% more performance (tech power up chart below). So pretty much they weren't able to hit what the node was expecting to hit. 60% decrease in power consumption or 40% increase at the same power usage. This might not be a liquid cooled edition, but at stock the liquid cooled only gets a few % points more, so it didn't make the cut either, it needs to be performing 48% more or so then it will hit the node estimations, its close so we can give it to them, but still its only the node that is showing up in power savings, nothing to do with the architecture.

perfrel_3840_2160.png
 
razor1 said:
Yes it can get higher clocks, this was due to adding pipeline stages to make the pipeline longer which uses transistors.

Over all though, a Vega 64 liquid cooled uses 300 watts, 25 watts more than a FuryX and gets 38% more performance (tech power up chart below). So pretty much they weren't able to hit what the node was expecting to hit. 60% decrease in power consumption or 40% increase at the same power usage. This might not be a liquid cooled edition, but at stock the liquid cooled only gets a few % points more, so it didn't make the cut either, it needs to be performing 48% more or so then it will hit the node estimations, its close so we can give it to them, but still its only the node that is showing up in power savings, nothing to do with the architecture.

perfrel_3840_2160.png
Click to expand...


So then, back to point #1, there are architectural changes :)
 
Digital Viper-X- said:
So then, back to point #1, there are architectural changes :)
Click to expand...

Not exactly, pipeline stages don't affect the architecture what they do is they add more gates switches and at each stage you need less transistors, it decreases transistor density (over all they use more transistors but more sparsely), which help dissipate heat more (simplistic view, but just to get the point across). The architecture itself stays the same. Making pipelines longer didn't save it any power also It made it worse cause of the higher clocks. Not only that, everything AMD has stated for increased performance over Fiji, so far hasn't shown up.

Increasing pipeline lengths to get higher clocks, that is what Intel did with the P4, we saw how far that got them lol.

That is not a smart way to get higher clocks, its an emergency way of getting higher clocks. Vega from an architectural point of view is a failure, its base voltage is too high, this is the type of architectural changes that need to be done. Things to drop core voltages and keep frequency up, that is what nV did with Maxwell and did more with Pascal. To quantify how much they did with Pascal we don't have the tools to do it, for Maxwell we do, they were able to keep the same voltage as Kepler but their clocks went through the roof.

back to Vega only, this goes back to pre Vega launch actually 1 year before launch, what I stated they didn't have time to do any low level transistor layout changes. If they did that the entire architecture will change and we would see that right now.
 
Well, I am pulling 350 Watts from the wall well playing Crysis 3 and that is pretty good. (Ryzen R7 1700X at 1.35v and 3.8 Ghz, Vega 56 at Turbo settings through the the Wattman control panel.) Crysis 3 will typically pull 50 to 100 Watts more power than almost any other game since it makes really good use of all available hardware resources. This is all without me even doing any tweaking or undervolting on the videocard yet. :D :)

That power draw is typical of any high end gaming system, at least as far as Crysis 3 goes. Any other game will have far less power draw.

Edit: Looks like Batman: Arkham Knight uses about 10 watts less power but, this is at 4k resolution, which I find draws more power than lower resolutions.
 
Last edited:
ManofGod said:
Well, I am pulling 350 Watts from the wall well playing Crysis 3 and that is pretty good. (Ryzen R7 1700X at 1.35v and 3.8 Ghz, Vega 56 at Turbo settings through the the Wattman control panel.) Crysis 3 will typically pull 50 to 100 Watts more power than almost any other game since it makes really good use of all available hardware resources. This is all without me even doing any tweaking or undervolting on the videocard yet. :D :)

That power draw is typical of any high end gaming system, at least as far as Crysis 3 goes. Any other game will have far less power draw.

Edit: Looks like Batman: Arkham Knight uses about 10 watts less power but, this is at 4k resolution, which I find draws more power than lower resolutions.
Click to expand...


So your card is pulling over 200 watts, to give just above Fury X performance, what does that mean...... 27% drop in power consumption for 12% increase in performance, purely showing the architecture wasn't changed to help with power efficiency.
 
razor1 said:
So your card is pulling over 200 watts, to give just above Fury X performance, what does that mean...... 27% drop in power consumption for 12% increase in performance, purely showing the architecture wasn't changed to help with power efficiency.
Click to expand...

Where the hell do you get this carp?
 
ManofGod said:
Where the hell do you get this carp?
Click to expand...


Whats your CPU put 125 watts max, rest of the system 25 watts, unless ya got multiply 8 gb or 10gb HD's for porn, that leaves what for the graphics card 200 watts? There ya go. The rest of it is just %'s based on how the Vega 56 performs on benchmarks vs Fiji's benchmarks, not to hard to find out.
 
razor1 said:
Whats your CPU put 125 watts max, rest of the system 25 watts, unless ya got multiply 8 gb or 10gb HD's for porn, that leaves what for the graphics card 200 watts? There ya go. The rest of it is just %'s based on how the Vega 56 performs on benchmarks vs Fiji's benchmarks, not to hard to find out.
Click to expand...

he was using a vanilla Fury.. so for him the upgrade goes well from Fury to Vega56.
 
razor1 said:
Things to drop core voltages and keep frequency up, that is what nV did with Maxwell and did more with Pascal. To quantify how much they did with Pascal we don't have the tools to do it, for Maxwell we do, they were able to keep the same voltage as Kepler but their clocks went through the roof.
Click to expand...

Not to take away from your overall point, but Pascal and Maxwell is a Vega - Fiji situation itself. Clock for clock, core for core, they're near identical. Maxwell itself even clocked insanely, they didn't really get that much more speed out of Pascal.

It just happens that Maxwell was an incredible arch to begin with.
 
Ramar said:
Not to take away from your overall point, but Pascal and Maxwell is a Vega - Fiji situation itself. Clock for clock, core for core, they're near identical. Maxwell itself even clocked insanely, they didn't really get that much more speed out of Pascal.

It just happens that Maxwell was an incredible arch to begin with.
Click to expand...

Yeah there has been a few clock for clock tests, which show the same thing, but they didn't get more clocks out of Pascal by making their pipes longer. They got 500 + mhz more than Maxwell did while dropping power consumption. The node can only do one of those on a significant basis, that is dropping power consumption, it can't really help getting more clocks, that is a design choice. If we see both of them and that is what we saw with Maxwell and also Pascal, that can't be done by node only. In the past it was done because nodes were capable of doing both, but not anymore with leakage the way it is. Architecture must be designed to reduced leakage amounts or actually chances of increased leakage. That is when they can drop core voltage and increase frequency at the same time.

There were too many changes to directly associate tflop = tflops comparison. At least for this discussion, core counts are way different in Pascal from same tier cards. And by NV's own admission the changes they did with Maxwell, they did more of the same with Pascal, that is why they were able to get those extra Mhz out of it.

Maxwell was and is a great architecture, it was pretty much perfect in almost all regards. Pascal just took that and made it better in only one aspect, clocks and voltage.

Looking up some number the only one that we can do a direct comparison to are these two cards the gtx 1050ti and the gtx 950, same cuda counts same bus size, so virtually the same chips. The 1050ti gets 30% more performance and uses 30% less power, that breaks the node changes by quite a bit. Factoring in the memory differences, the 1050ti has more memory chips (2gb more) so that would increase power consumption by 5-10watts, we are looking at around 40% difference in power. Even at 30% 30% it breaks the node at 30% 40% is breaking it by a large margin.
 
Last edited:
Funny enough, I haven't found any clock-for-clock 950 to 1050ti comparisons. If I knew I could get enough YouTube views to pay for it, I'd order them and test them myself.

It is crazy how low they got the voltages on the 1080 and 1070. 1.05 vs 1.2 on the 980ti, comparing boost volts to boost volts. 1070/1080 can idle as low as .6V! I really wish AMD could compete with that.
 
Ramar said:
Oh shit, thanks. Sorry to not want to waste thirty+ minutes watching those videos, but did he measure power consumption? FPS is near identical, I'd be really interested to see what the actual power savings are at that point.
Click to expand...


Sadly he didnt measure power, just compared FPS overall in games.
 
Dayman said:
Sadly he didnt measure power, just compared FPS overall in games.
Click to expand...


Well the main change for Pascal was voltage, and everyone that tested clock for clock on pascal vs maxwell, missed it :). What did nV say when they released Maxwell, perf/watt is the new metric for performance, cause if ya have that, ya got performance automatically.
 
Ramar said:
Oh shit, thanks. Sorry to not want to waste thirty+ minutes watching those videos, but did he measure power consumption? FPS is near identical, I'd be really interested to see what the actual power savings are at that point.
Click to expand...


No one has done that, I got the performance and wattage figures from stock cards from Techpowerup review of the 1050ti.

Added note 1050ti uses Samsung's 14nm too which is well a bit weaker then TSMC's 16nm node, thus why for the 1050ti and 1050 nV decided to lock voltages at a maximum on those cards which is actually quite lower than what TSMC's 16nm nV chips are capable of.
 
So, when are we going to move this thread to the Nvidia section since that is all this thread is about?
 
IdiotInCharge said:
You're pro-AMD and anti-Nvidia, we get it, talking about GPU efficiency when it's not totally AMD triggers you. Thankfully, you can just click 'Unwatch Thread' at the top and move on!
Click to expand...

Nope, I am Pro computer hardware with Pro Amd for my personal builds only. I know computer hardware quote well and have no issues with facts. This is an Nvidia thread now, just the way things are. Lots of things bother me but talking about facts do not, like the fact that this is an Nvidia thread now.

Edit: Oh, and check the forum heading you are in.......
 
_mockingbird said:
There's a name of the website at the bottom of the image. I know it's terribly exhausting to go to the website, but you have to do some work.
You are shifting the goal post.
Earlier, you asked, "Where is the proof that each GCN design only goal is tied to using more power?".
So, now you have it.
Click to expand...
I told you about RX 480 and it is better at power consumption, you ignore it . There even some demo of other GCN cards that do very well there lower clocked and behave pretty decent in certain areas. You refuse to link to the article for the very same reason that you never want any details you just want to post nonsense as your first post proves this.
Chimpee said:
Really, because clock for clock, RX Vega 64 and Fury X perform similar that suggest there is no architectural change (shown in [H] comparison), and which tells me the power saving is most likely result from going to 28nm to 15nm rather than any architectural changes.
Feel free to post the video from AMD youtube channel, because I am not doing the work for you.
Edit, meant to type 14nm instead of 15nm.
Click to expand...

Odd that Shintai is the only one person that gets this.
You confuse things which are well known GF does not have a true 14nm process it is called that however....


Shintai said:
12LP is a renamed 14LP+ process and there are no area savings ro shrinkage. In reality its a 16nm or so node.
Just as 12 FFN is a 4th generation 16 FF that's really a 20nm FF node.
Hence why "12LP" can get 15% reduction over "16FF".
Click to expand...

So the node scaling is in reality a lot smaller then you think. Already posted that in the reply to Araxie.

GCN is a hardware feature set the current AMD gpu supports. It however does not dictate power usage in any way shape or form. There are APU which use GCN cores they perform pretty well.
Yelling and screaming that GCN is to blame is just another way of saying you have no clue....
 
Pieter3dnow said:
I told you about RX 480 and it is better at power consumption, you ignore it . There even some demo of other GCN cards that do very well there lower clocked and behave pretty decent in certain areas. You refuse to link to the article for the very same reason that you never want any details you just want to post nonsense as your first post proves this.
Click to expand...


IT IS NOT BETTER, its worse then last generation's midrange in comparison to nV. Both nV and AMD got either a 60% drop in power consumption or 40% increase in power savings because of the node.

AMD got that nV surpassed that! Do you understand now why GCN has been losing ground ever generation? GCN as an architecture is holding them back, nV has been modifying their architecture to get better perf/watt.

Odd that Shintai is the only one person that gets this.
You confuse things which are well known GF does not have a true 14nm process it is called that however....
Click to expand...

What are you talking about GF's is a real 14nm process, its actually very similar to Intel's 1st generation 14nm process just a little less dense. Same type of fin layouts. Intel's 1st gen 14nm process didn't scale all that well too pretty much maxed out at 4.5 ghz.

Try again.

So the node scaling is in reality a lot smaller then you think. Already posted that in the reply to Araxie.

GCN is a hardware feature set the current AMD gpu supports. It however does not dictate power usage in any way shape or form. There are APU which use GCN cores they perform pretty well.
Yelling and screaming that GCN is to blame is just another way of saying you have no clue....
Click to expand...


Look when chips are designed they are designed for a certain tdp vs performance vs voltage vs frequency. If they can hit those targets they usually do well, if they don't they become space heaters or worse Chernobyl. Now We already know Vega is very close to its voltage limits. Do you agree or disagree? Should agree that is a fact. Now do you think AMD planned for that? If they planned for that why would they plan for that? If they didn't plan for that, then they messed up somewhere. Now how many generation of chips has GCN pushed voltage to its limits. Lets see, Fiji, Polaris, Vega. Wow last 3 gens are GCN, what did I say GCN was good for 2 generations, after that it was out of its league.
 
Last edited:
Status
Not open for further replies.
Back
Top