• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Jen-Hsun Huang, NVIDIA co-founder, president, and CEO

If NVidia lied so blatently there would be a 20 page thread with the same half a dozen or so members milking it for all its worth in their crusade against the great evil that is NVidia. AMD get called out for doing it and all of those people remain completely quiet about it hoping that it quickly passes or throw their toys out of the pram at it being mentioned as in this case.
 
Last edited:
Silent_Scone said:
Well spoken, and the actual fact of the matter is the VRM on this card actually doesn't perform all that well when under exterted stress from what it uses at stock on-top of the fact there is little headroom on Fiji.

Essentially, being on stage and saying something is one thing when it is pretty much the exact opposite...It's not the end of the world, no...but it's something to add to the list of paper tiger hidden smoke.


When you look at the huge headroom Maxwell has from it's base and even boost clocks on various cards - it would have made more sense for NVIDIA to use this as a sell.
Click to expand...

I doubt the VRMs are a problem - though relatively cheap parts they are decently specced for junction temperature and amperage handling capabilities.
 
Rroff said:
I doubt the VRMs are a problem - though relatively cheap parts they are decently specced for junction temperature and amperage handling capabilities.
Click to expand...

I didn't say they were cheap parts. The Fury X uses the same grade of MOFSET as the 290x and tantalum caps to fit into the small footprint, the problem arises there, active cooling over this small area is the single biggest benefit to pushing Fiji. Once the cards are put under load the efficiency and noise go out the window as they become saturated and run very close to the tolerance limit hitting over 100c. With this in mind the capabilities of the VREG components are of no consequence. So at default operation we then circle back to the overlockers dream debate. Where is the headroom?
 
nashathedog said:
That's a different circumstance, AMD have pretty good scaling plus the example I'm talking about is my Fury Tri-x which at stock often beats the 980 and is considered by many too be faster than a 980 while having very little overclocking room unlike the 980 and losing out once both are overclocked.
Click to expand...

Don't you mean the Fury X ?, many say thats around the 980, wins some, loses some, and the Fury non X is only around the rebranded 290X.
 
Last edited:
LoadsaMoney said:
Don't you mean the Fury X ?, many say thats around the 980, wins some, loses some, and the Fury non X is only around the rebranded 290X.
Click to expand...

No not the Fury X, From what I've seen the Fury beats the 980 most the time when both are at stock clocks, The Fury is a good bit faster than Hawaii/Grenada but it's inconsistent and some titles show a 10 or more frame difference, some show just a few frames and in some it even loses.
There seems to be something software based holding Fiji back at the moment. When you look at the core quantity difference it should be doing a lot better when compared to Hawaii/Grenada but it's not and when you add overclocks it sucks to often see it lose out.
Hopefully it'll be solved at some point.
 
Silent_Scone said:
I didn't say they were cheap parts. The Fury X uses the same grade of MOFSET as the 290x and tantalum caps to fit into the small footprint, the problem arises there, active cooling over this small area is the single biggest benefit to pushing Fiji. Once the cards are put under load the efficiency and noise go out the window as they become saturated and run very close to the tolerance limit hitting over 100c. With this in mind the capabilities of the VREG components are of no consequence. So at default operation we then circle back to the overlockers dream debate. Where is the headroom?
Click to expand...

I said they are cheap parts (which they are - but not cheap rubbish cheap) - they are more than adequately specced for the application here though. Even with the temperature considerations their efficiency, etc. shouldn't be holding back overclocking at this level - sure they could be an issue at hitting extreme clocks if there was better capabilities elsewhere on the card.
 
nashathedog said:
No not the Fury X, From what I've seen the Fury beats the 980 most the time when both are at stock clocks, The Fury is a good bit faster than Hawaii/Grenada but it's inconsistent and some titles show a 10 or more frame difference, some show just a few frames and in some it even loses.
There seems to be something software based holding Fiji back at the moment. When you look at the core quantity difference it should be doing a lot better when compared to Hawaii/Grenada but it's not and when you add overclocks it sucks to often see it lose out.
Hopefully it'll be solved at some point.
Click to expand...


it may not be a software limitation. Increasing the number of cores doesn't give you linear scaling, other bottle necks raise their heads and interfere. Otherwise Nvidia would just reuse the same Fermi architecture. Fiji has the same 4 shader engines delivering work to more m=cores, Hawaii had the same 4 shader engines. If the limit is in the SE then the cores are effectively sitting idle. There was also no change in the number of ROPs which are linked to the shader engine, that could also limit the throughput.


Then there is a big debate on the effectiveness of HBM when you are not bandwidth limited, which most of the the time at most resolution you aren't. People rarely bother overlooking the GPU memory much because it doesn't increase performance much, because bandwidth is not a problem currently. HBM may be solving a problem that doesn't currently exist, but AMD had to go that route in order to ct power requirements. HBM may have some other cost when the bandwidth is not advantageous. I have seen some strange benchmarks where overclocking the Fiji memory resulted in big performance gains, yet there should be way less gains than with slower GDDR5 because Fiji should be less bandwidth limited.
 
Rroff said:
I said they are cheap parts (which they are - but not cheap rubbish cheap) - they are more than adequately specced for the application here though. Even with the temperature considerations their efficiency, etc. shouldn't be holding back overclocking at this level - sure they could be an issue at hitting extreme clocks if there was better capabilities elsewhere on the card.
Click to expand...

Not sure what you are saying as it doesn't really change what I've just told you. Yes it is an issue when overclocking. I am telling you one thing and you are trying to refute it by repeating that the VRM components are adequate
 
D.P. said:
it may not be a software limitation. Increasing the number of cores doesn't give you linear scaling, other bottle necks raise their heads and interfere. Otherwise Nvidia would just reuse the same Fermi architecture. Fiji has the same 4 shader engines delivering work to more m=cores, Hawaii had the same 4 shader engines. If the limit is in the SE then the cores are effectively sitting idle. There was also no change in the number of ROPs which are linked to the shader engine, that could also limit the throughput.


Then there is a big debate on the effectiveness of HBM when you are not bandwidth limited, which most of the the time at most resolution you aren't. People rarely bother overlooking the GPU memory much because it doesn't increase performance much, because bandwidth is not a problem currently. HBM may be solving a problem that doesn't currently exist, but AMD had to go that route in order to ct power requirements. HBM may have some other cost when the bandwidth is not advantageous. I have seen some strange benchmarks where overclocking the Fiji memory resulted in big performance gains, yet there should be way less gains than with slower GDDR5 because Fiji should be less bandwidth limited.
Click to expand...

I've read stuff that's mentioned the rops etc as being a possible issue so lets say I'm kinda hoping there's something that can be done software wise to improve things because I've also read stuff stating that to be the case, In truth it's most likely a bit of both holding it back.
As it stands things are not nearly as good as they could be and it needs solving.
 
TwsT said:
not sure if bait

"Lisa Su is the CEO and President of Advanced Micro Devices. She was appointed president and CEO in October 2014"
Click to expand...

It's just my humour, the last time she was talking at a conference the thread had a running joke that she didnt look very feminine.

you had to have been there I suppose :D
 
You must log in or register to reply here.
Back
Top Bottom