28nm : Why Nvidia overclocks better??

In the new PCWorld article on FuryX, AMD cites 100mhz overclock as some big deal...1050 to 1150 on the core....doesn't seems much to me?
I think that 100 MHz might be what they were sure could be reliably hit, possibly a speed bin they had been considering all along. The variability in overclocking may be too high past that point to make it safe for AMD to give some kind of official indication that there is a guarantee.
 
In the new PCWorld article on FuryX, AMD cites 100mhz overclock as some big deal...1050 to 1150 on the core....doesn't seems much to me?

Nicely, AMD own numbers falls inline with what i asked earlier...1150mhz FuryX vs 1300mhz 980Ti, who will win?
I think they only did that to show off that you can actually OC it, it should have no problem going higher if you push it. I mean GCN ussually hits 1200+ without much trouble even on air.
 
http://www.legitreviews.com/amd-fiji-arrives-radeon-r9-fury-x-details_166515#OpQZsfoAmLt9ajjt.99

No overclocking of HBM1! Contrary to what was said earlier...??

At this clock speed it is capable of hitting 512GB/s of memory bandwidth. One interesting thing that we just learned this week is that AMD will not allow you to overclock the memory when the first Fiji cards are released. AMD feels that the memory technology is too new and there is more than enough bandwidth, so they are locking down the ability to overclock the memory in AMD Overdrive. This might change down the road, but for the time being only the core clock can be overclocked by end users.
 
I think that 100 MHz might be what they were sure could be reliably hit, possibly a speed bin they had been considering all along. The variability in overclocking may be too high past that point to make it safe for AMD to give some kind of official indication that there is a guarantee.

I dont see it as 'official' indication, or at least when come to overclocking, we don't expect official words from chip makers, only bragging rights.

Hence i expected AMD to show-off at least a +30% for this one-off statement, like Intel 5Ghz CPU brag. +100mhz (10%) just seemed so weak...after their brags about the AIO capabilities and number of power phases, no?

Or AMD is just going the wrong way for PR speak..?
 
I think that 100 MHz might be what they were sure could be reliably hit, possibly a speed bin they had been considering all along. The variability in overclocking may be too high past that point to make it safe for AMD to give some kind of official indication that there is a guarantee.
If they are that confident in 1150Mhz, why not have it as a second bios setting enabled by a switch on the card as with previous Radeon flagships?
 
Hearing that a 20-25% overclock on Fury-X with stock volts is relatively easy, if you loosen the power limit.
I'm going out on a limb here but I have a feeling that Fiji overclocks just as well as Maxwell, ~20-25% standard and 30-40% if you push it a bit

If they are that confident in 1150Mhz, why not have it as a second bios setting enabled by a switch on the card as with previous Radeon flagships?
Have a not so great memory and it is even worse currently.
I thought that the 2nd bios was typically for a thermal/voltage limit increase, not that it was defining any specific clockspeed..
 
I always assumed that the transistor count / die size ratio is the culprit :)

In latest chips, Fury has 8,9 billion transistors on 596mm2, while GM200 has "only" 8billion on 601mm2.
 
Strong first post, it's pretty obvious that it'd come down to overclocking for the best enthusiast card, nvidia would love to release a Titan XXX if Fury doesn't go wild with an overclocker's dream.
 
I always assumed that the transistor count / die size ratio is the culprit :)

In latest chips, Fury has 8,9 billion transistors on 596mm2, while GM200 has "only" 8billion on 601mm2.
It's definitely a correlation, but I'd hesitate to call causation. It's possible that AMD is using a denser standard cell library, but even that doesn't have to imply lower performance.
 
It's definitely a correlation, but I'd hesitate to call causation. It's possible that AMD is using a denser standard cell library, but even that doesn't have to imply lower performance.
However. Active power also depends on the amount of transistors switching. If all other things are equal, and AMD needs more transistors to accomplish the same task as Nvidia, then even if a denser cell library will help them fit within die size limits, it will still cost power.
For a hypothetical example, if Nvidia's caches need less transistors for the same amount of memory, Nvidia gains a power advantage, even if their fewer transistors take up the same space. Those Nvidia transistors may take more space because AMD is better at physical design, or there may be a fundamental tradeoff involved.
Anyway, I know nothing about this, except that things may not be as simple as they appear.
 
It may be more illustrative if we had a die shot for the chips we're comparing. Interface area, the mix of transistors, and the activity level of the units using them would matter as well.
Raw numbers alone may not help, particularly if the HBM interface frees up area.
Eyeballing some of the dies shots, it might be the case that the shader engines on GCN have less die area allocated versus what Kepler or Maxwell have for their equivalent subdivisions. Given how active they are, and how much hardware they pack, the mix may be more density-focused. I really need to find time to do more than guess, since first glances can be thrown off so easily. Having more hardware besides the units that are active the most means they have to be denser, even if the slower hardware using more conservative layouts makes the overall transistor/area count more close.
 
Die shots are a thing of the past. A bunch of big review websites should band together and have the latest and greatest dies xray'ed. Ryan?
 
Die shots are a thing of the past. A bunch of big review websites should band together and have the latest and greatest dies xray'ed. Ryan?
I don't even want to imagine how much it would cost to get Chipworks or such to go do something like that. Die shots are not easy, and the people who do them professionally are in the business of making a business out of them.
 
http://forums.overclockers.co.uk/showpost.php?p=28207511&postcount=1129

The admin at a UK online store have spoken about Fury poor stock overclocking. AMD may have locked down on the voltages(core & hbm)...I am not liking that i need to add core voltages to overclock past 1.1Ghz...

At this stage...i think Nvidia will not be dropping prices of 980Ti...it may even go up!?

Other sources are saying +1.2ghz on stock volts.
Gibbo also initially said that Tahiti was a not so great overclocker and then a week later said it was great.
 
Other sources are saying +1.2ghz on stock volts.
Gibbo also initially said that Tahiti was a not so great overclocker and then a week later said it was great.

Was he referring to Hawaii or Tahiti?
Tahiti was a good overclocker, at least the 7970 was.
Hawaii was not impressive when pitted up against Nvidia's.
 
And 8Pack later in the same thread mentioned that AMD ran several Fury cards at 1.2GHz with stock volts. There obviously is no guarantee you will get that but spread should be decent for 1.2Ghz capable cores. You must remember that Gibbo was comparing it with 980Ti G1 which overclocks to 1.4 -1.5 GHz. Next to it even 1.3GHz will look like bad overclock.Also he rans business and needs to sell 980Ti and Titans X still, so he will not be hyping new card with limited launch stock of about 100 cards as this would hurt his business.


Additionally AMD engineers ran FuryX at stock for several hours in gaming loads with passive cooling and didn't hit any thermal throttling. Near silent gaming card of this performane is a first in a long time.

Can't wait for proper reviews and decent stock of cards in UK (without price gouging of over 100 quid above RRP).
 
Last edited:
I was afraid of this, 980 Ti OC edition DESTROYS Fury X
http://forums.anandtech.com/showpost.php?p=37507613&postcount=187

But the liquid-cooled Fury X was made to be overclocked. “You’ll be able to overclock this thing like no tomorrow,” AMD CTO Joe Macri said at the card's unveiling. “This is an overclocker’s dream.”
Not enough rolleyes here.
Can someone take Joe to tasks for such extremely dishonest antics?
He is your CTO for chris' sakes! ....not some small time forums viral promoter.
Not going to trust AMD again, this and their Bulldozer debacle. When the going gets tough, AMD gets lying.
 
Back
Top