[THW] Review Samples vs Retail R9290x boards(large differences)

[ShintaiDK]

The 290 fix is even worse since the fanspeed goes above press sample, and the retail cards still perform worse as well:

 

[amenx]

You mean to say that the 6% performance dip is worth noting D:

Isnt that roughly the difference the 290x overtakes the 780?

 

[ShintaiDK]

Isnt that roughly the difference the 290x overtakes the 780?

One could also say the reason why you pay 150$ extra for the 290X over the 290 if translated into $.

 

[desprado]

Isnt that roughly the difference the 290x overtakes the 780?

now gtx 780 is 1% faster.

 

[R0H1T]

Isnt that roughly the difference the 290x overtakes the 780?

Hardly, the most important factor being resolution & then comes temp cause if the 290x isn't throttling then it's beating the 780 by 10% on average across most reviews.

 

[Abwx]

The 290 fix is even worse since the fanspeed goes above press sample, and the retail cards still perform worse as well:

No , it s wrong , the retail cards have lower fan speed
than reference and thus are less noisy as aknowledged
by hardware.fr who did tests retail cards to check THG
claims.

 

[ShintaiDK]

No , it s wrong , the retail cards have lower fan speed
than reference and thus are less noisy as aknowledged
by hardware.fr who did tests retail cards to check THG
claims.

Sure thing, before the fix:

Too bad AMDs lies and cheats got caught. But why change a perfectly good track record from the CPU division.

 

[3DVagabond]

The whole story is that they took retail cards whose
fan speeds where set out of spec by the manufacturers,
read fan running 6.5% slower than in ref cards , and then
presented it as fully specced cards not performing accordingly.

Well I stand corrected if thats the case. But AMD did seem to think it generally widespread enough to issue a driver fix.

In case people haven't read how they fixed the issue. They adjusted the fan speed by PWM. What they found was that the manufacturing tolerances of the fans aren't accurate enough to offer the fine granularity speed control needed to keep temps/clocks in balance. For example 40% could vary by a few hundred RPM. Instead they now actually set the fan speed. Seems like the engineers were being a bit to cutesy and the hardware wasn't up to the tech. Sometimes the simple solution is the best.

This doesn't exclude the error and except for the embarrassment and inconvenience for some users (a very very small number, from what it appears) I find it quite humorous when engineers *"out smart themselves". It's a far cry though from the "AMD is cheating!!!!" announcements that were first made about it here.


*Note: No actual engineers were harmed in the writing of this post.

 

[R0H1T]

Sure thing, before the fix:

Too bad AMDs lies and cheats got caught. But why change a perfectly good track record from the CPU division.

Oh hey look the press card was 5% faster than the retail version

The chart seems to suggest that the press card was hovering around ~950Mhz for the first half of the run & then throttled down closer to the Sapphire's ~900Mhz average speed throughout the 100 sec run, well that's assuming I haven't read the charts upside down

 

[Abwx]

Sure thing, before the fix:

Too bad AMDs lies and cheats got caught. But why change a perfectly good track record from the CPU division.

It is obvious from thoses graphs that once the retail card fan speed
is set close to the reference card the scores difference become
negligible , in fact proportional to the remaining fan speed difference.

So AMD is cheating because retail cards are not fan speed wise
within specs of the ref cards but still the GPU should perform
the same , worse being that according to your strange logic
increasing fan speed of the retail card from 2300 turns/min
up to slightly lower speed than ref card will result in much
higher noise that the latter even , in short the retail card
at 2300 tr/min is already as loud as a ref card at 2550....

 

[amenx]

In case people haven't read how they fixed the issue. They adjusted the fan speed by PWM. What they found was that the manufacturing tolerances of the fans aren't accurate enough to offer the fine granularity speed control needed to keep temps/clocks in balance. For example 40% could vary by a few hundred RPM. Instead they now actually set the fan speed. Seems like the engineers were being a bit to cutesy and the hardware wasn't up to the tech. Sometimes the simple solution is the best.

So these cards need fans with 'higher manufacturing tolerances and accurate enough to offer the fine granularity speed control needed to keep temps/clocks in balance'..... vs other cards that dont. K, got it. :thumbsup:

 

[Silverforce11]

It's still only affected cards from Tom's isn't it, fan error or HSF error, just RMA it.. but nah, Tom's have to milk it for publicity AFTER NV told them to do it. Joke.

Any end user buying these cards, if it performs worse than press samples (beyond benchmark variations), RMA it. End of story.

 

[3DVagabond]

So these cards need fans with 'higher manufacturing tolerances and accurate enough to offer the fine granularity speed control needed to keep temps/clocks in balance'..... vs other cards that dont. K, got it. :thumbsup:

Other cards boost functions don't operate the same way.

 

[Silverforce11]

Its amazing how much bad press AMD is taking yet the R290X is still sold out in most shops here... guess most people DGAF and go water cooling anyway because the EK plates are also sold out at the same stores..

Maybe AMD DGAF because they figured only certain people folk out $550 for reference cards..

 

[ShintaiDK]

Its amazing how much bad press AMD is taking yet the R290X is still sold out in most shops here... guess most people DGAF and go water cooling anyway because the EK plates are also sold out at the same stores..

Maybe AMD DGAF because they figured only certain people folk out $550 for reference cards..

Yet you can still buy the BF4 edition cards, even tho it was a limited 8000 unit run. So either people dont buy the bundle, or they dont sell so much.

 

[monstercameron]

Yet you can still buy the BF4 edition cards, even tho it was a limited 8000 unit run. So either people dont buy the bundle, or they dont sell so much.

or that have more than whatever was speculated/quoted...

 

[DarkKnightDude]

or that have more than whatever was speculated/quoted...

Wouldn't be too surprising. Companies make things sound like they're limited to sell when in reality they're probably not.

 

[Abwx]

In case people haven't read how they fixed the issue. They adjusted the fan speed by PWM. What they found was that the manufacturing tolerances of the fans aren't accurate enough to offer the fine granularity speed control needed to keep temps/clocks in balance. For example 40% could vary by a few hundred RPM. Instead they now actually set the fan speed. Seems like the engineers were being a bit to cutesy and the hardware wasn't up to the tech. Sometimes the simple solution is the best.

This doesn't exclude the error and except for the embarrassment and inconvenience for some users (a very very small number, from what it appears) I find it quite humorous when engineers *"out smart themselves". It's a far cry though from the "AMD is cheating!!!!" announcements that were first made about it here.


*Note: No actual engineers were harmed in the writing of this post.

In principle manufacturing tolerances are such that two
same fans are below 1% difference in speed at a given
supply voltage so one could use the voltage as reference
to set the speed but in this case it is obvious that the retail
cards have not exactly the same fans as the ref cards.

 

[SPBHM]

from PC persepctive

http://www.pcper.com/reviews/Graphi...rmance-Fan-Speeds-R9-290X-and-R9-290/Catalyst

 

[GaiaHunter]

As I've asked in the other thread.

What is the core voltage of all the samples?

I get the feeling that different core voltages might be the culprit and not all the nonsense of golden coolers.

I also get the feeling no one really wants to talk about different core voltages because NVIDIA has announced ages ago that they would start shipping cards with different voltages under the same name. I wonder which of those chips are sent to reviewers, the low voltage bins or the higher voltage bins.

 

[DarkKnightDude]

*snip*

Interesting, so, Tom's is just looking for hits as usual?

 

[Leadbox]

In principle manufacturing tolerances are such that two
same fans are below 1% difference in speed at a given
supply voltage so one could use the voltage as reference
to set the speed but in this case it is obvious that the retail
cards have not exactly the same fans as the ref cards.

Outside of Tom's 2 cards, any other instances of this reported?
I would like to have those 2 cards examined for tempering myself :sneaky:

 

[Abwx]

Outside of Tom's 2 cards, any other instances of this reported?
I would like to have those 2 cards examined for tempering myself :sneaky:

Hardware.fr made an article with measured retail cards ,
they found that fan speeds where 6.5% lower than ref
cards with the perfs being about the same as the ref
once speeds are about even.

They do all their tests in a room whose temp is maintained
between 26 and 26.9°C btw.

 

[Gikaseixas]

So Nvidia asked Tom to push this? Low blow

 

[3DVagabond]

So Nvidia asked Tom to push this? Low blow

Guru3D said they were told about it as well by nVidia in an NDA meeting. I would guess that nVidia told all their media partners about it. Why Tom's was the one who said they "discovered" it, who knows. Why Tom's was the only one who decided to include an effected sample in their reviews making it appear that it was a common problem with retail samples, we can only speculate.