Asus EAH4850 and unbelievable AutoTune result

[DRS]

I recently bought an Asus EAH4850 for my S939 X2 4200 (@2.6GHz) system. As
it's my first ATI card apologies in advance if this question has been asked
and answered.

The EAH4850 has a proprietary Asus cooler (a 90mm fan in a Zalman-type
copper heatsink), so it's effectively a 2-slot setup even though it doesn't
have rear exhaust vents like some Nvidia's do, but it runs very much cooler
than the 4850 reference card.

That being so, I turned Overdrive on and ran AutoTune (Asus CCC 8.5.22).
These are the results (default Asus 50% fan speed; A = ambient):

GPU 500 -> 790
Mem 993 -> 1073

idle 45C (A 19C) -> 50C (A 23C)
peak 63C (A 19C) -> 75C (A 23C)

If AutoTune is to be believed I can get a 58% performance boost from my GPU.
Now, I know enough about overclocking to know about stress testing and what
have you but still, as a first order question, can AutoTune be believed? Is
anyone else overclocking a 4850 to these sort of speeds?

 

[DRS]

This reviewer got his EAH4850 with a reference cooler up to 750 MHz.
Given the more capable TOP cooler on your card, 790 MHz is
reasonable. You can thank the 55 nm process.
http://www.overclock3d.net/reviews.php?/gpu_displays/asus_hd4850_512mb_pci-e_graphics_card_eah4850/3

My box nowhere says TOP. In fact, I can't find my card on Asus's au.asus
website! It looks like the EAH4850/HTDI/1G
(http://au.asus.com/products.aspx?l1=2&l2=8&l3=725&l4=0&model=2314&modelmenu=1)
but it has the same specs as the EAH4850/HTDI/512M
(http://au.asus.com/products.aspx?l1=2&l2=8&l3=725&l4=0&model=2328&modelmenu=2).
I'm wondering if Asus hasn't just decided to save itself a lot of grief and
provide all its 4850s with the "Glaciator" cooler.

A bit troubling is the proprietary Asus Autotune utility, though. The
EAH TOP edition should be running at 680 MHz in 3D:
http://www.asus.com/products.aspx?modelmenu=2&model=2329&l1=2&l2=8&l3=725&l4=0

The 500Mhz is probably my mistake. I used the CCC Default button to reset
the clock speeds and the GPU is at 625Mhz. Since my card is not explicitly
labelled as TOP this may well be correct, particularly since AutoTune
wouldn't take my memory past 1073 and the TOP's memory comes at 1050 and
should go higher. 790 then becomes a more reasonable but still impressive
26% boost. As it is, I should be able to run at TOP speeds without any
difficulty. I'm not interested in setting any records.

 

[Arno Wagner]

I recently bought an Asus EAH4850 for my S939 X2 4200 (@2.6GHz) system. As
it's my first ATI card apologies in advance if this question has been asked
and answered.

The EAH4850 has a proprietary Asus cooler (a 90mm fan in a Zalman-type
copper heatsink), so it's effectively a 2-slot setup even though it doesn't
have rear exhaust vents like some Nvidia's do, but it runs very much cooler
than the 4850 reference card.

That being so, I turned Overdrive on and ran AutoTune (Asus CCC 8.5.22).
These are the results (default Asus 50% fan speed; A = ambient):

GPU 500 -> 790
Mem 993 -> 1073

idle 45C (A 19C) -> 50C (A 23C)
peak 63C (A 19C) -> 75C (A 23C)

If AutoTune is to be believed I can get a 58% performance boost from my GPU.

Not quite. It is more like 33% and may be even less. Memory
peed is importand.

Now, I know enough about overclocking to know about stress testing
and what have you but still, as a first order question, can AutoTune
be believed? Is anyone else overclocking a 4850 to these sort of
speeds?

The problem here is that while the actual semicondictor can probably
stand 150C, the overall chip temperarure needs to be colder, because
of "hot spots" that do not show up in the chip temperature.
Overclocking makes the needed margin higher. In addition, chip
power consumption typically goes wuith the square of the clock
frequency.

Just to give you a sample calculation:

Assume the chip has a maximum temperature of 100C, and
a hot-spot limit of 150C, which it reaches at 110C
GPU temperature. That would mean 40C margin. You overclock
by a factor of 1.58, which is squared 2.49. This means
in the worst case you now need a distance between chip temperature
and hot-spot of 2.50 * 40C = 100C, i.e. a chip temperature below
50C. Going over that risks immediate catastrophic failure.

In addition, there is accelerated ageing, roughly at a factor
of 2 for every 10C. Say the GPU has 50 years life-expectancy
at 50C. With 90C and hot-spots at 130C, that gives you roughly
2.5 Months at full load before it dies.

Now assume your overclocking and your (better) heatsink.
75C gives a hot-spot temperature of about 175C,
which in turn results in alifetime of rougly 1 week
at full load.

Of course this is all simplified and the base numbers
are just examples, but it shows that hot semiconductors
can live very short.

Arno