Dont understand...hd4850 10w idle @ 68c

quietnevbie · Post by **quietnevbie** » Mon Aug 18, 2008 9:54 am

ATT offers me [0,892-0,930-0,968-1,006-1,044-1,082-1,120-1,158]. (so in 0,038V steps)

If I load my BIOS with RBE, some clock info spots have a visible voltage setting of 1.046.

ATI could set this range to restrict overclocking, perhaps. I would imagine they don't care so much about potential custom undervolting. The only way for me to know what the real voltage steps are would be doing some stability testing by raising clockspeed and watching for errors.

My 2-3C temperature difference between default idle and custom idle is quite small when compared to your results.

(btw, it's clear that W1zzard is wrong about the 25MHz real memory speed steps... otherwise I wouldn't see corruption occuring during 471/472 change)

update:
the 50MHz core clock will cause spiky polygons in Furmark, so it seems that that 160MHz in the MSI BIOS has a reason. Lower than 160MHz core clock probably results in non-fatal artifacts in various situations.

ReelMonza · Post by **ReelMonza** » Mon Aug 18, 2008 11:11 am

quietnevbie wrote:(btw, it's clear that W1zzard is wrong about the 25MHz real memory speed steps... otherwise I wouldn't see corruption occuring during 471/472 change)

Well, maybe not :

PLL's clock frequency generation granularity. Hardware cannot derive any clock you wish, the clocks are derived from 27MHz oscillator by multiplying / dividing its' frequency (which is called reference frequency) using fixed set of integer dividers. So depending on used generator and some other conditions (e.g. target clock range) clock generation step can be up to 27MHz.

nafets · Post by **nafets** » Mon Aug 18, 2008 2:21 pm

quietnevbie wrote:ATT offers me [0,892-0,930-0,968-1,006-1,044-1,082-1,120-1,158]. (so in 0,038V steps)

If I load my BIOS with RBE, some clock info spots have a visible voltage setting of 1.046.

ATI could set this range to restrict overclocking, perhaps. I would imagine they don't care so much about potential custom undervolting. The only way for me to know what the real voltage steps are would be doing some stability testing by raising clockspeed and watching for errors.

My 2-3C temperature difference between default idle and custom idle is quite small when compared to your results.

(btw, it's clear that W1zzard is wrong about the 25MHz real memory speed steps... otherwise I wouldn't see corruption occuring during 471/472 change)

Thanks again for the details on the HD4850 voltages. I would assume that all the voltages listed are "real usable" settings, but from what you've seen the few lowest of which are not stable with your HD4850. Maybe with a Kill-A-Watt or DMM it would be possible to see the actual power usage differences (if any) in verifying that the different voltages set are actually working. But not everyone has either of those measuring devices.

Regarding the memory speed steps, the card is able to be set in 1MHz steps for the memory. It's just that using the "Sensors tab" in Gpu-Z to verify memory speeds will be inaccurate, as it only displays those speeds in 25MHz steppings. Whether this is from the PLL reporting in 25MHz steps or Gpu-Z only displaying in 25MHz steps, I am unsure of.

To see the actual memory speed just click the "Graphics Card tab" and look at the memory speed listed under the Driver Version. For me it has always been correct, to the nearest 1MHz...

thejamppa · Post by **thejamppa** » Fri Aug 22, 2008 3:45 am

I just installed HD 4850 + Acclero S1 rev + S-flex 800 RPM combo.

I honestly can say that the installation was hardest I've had:

1) There was 3 plastic spacers instead 4. So I made DIY from small plastic tube.

2) With 120mm fan on Antec Nine Hundred case, the plastic clip holding optional side fan is parctically 1mm or less away from 120mm fan I am using.

I am going to get framless twin tubo's soon to give slightly more space.

However:

GPU is 31-32 degree's C in idle. That's awesome.

To sumn installation:

Was pain in the ass. Thermaltape on ram sinks sucked. I had difficulties even I changed them but they're holding now. I've placed 2 small Sharkoon ram sinks on the small VRM's.

Its huge with 120mm fan and turbomodule with give slightly moreclearance in side ways.

However, performance is stellar and worth of possible problematic installation.

with HD 3850 256mb I got 9611 in 3D Mark 06. With HD 4850 I got 11508 marks. Nice jump.

quietnevbie · Post by **quietnevbie** » Fri Aug 22, 2008 9:10 am

I didn't bother with the plastic spacers at all. It doesn't look anyhow wrong without them.. but then again my temperatures are quite high, so I cannot advertise my way just yet.

How to get the ramsinks to stick on first try (or this is what I did):
1. Use ArctiClean to do magic to the surfaces, after removing the stock cooler.
2. Just put the ramsinks where they belong, and keep pressing some time.
3. Turn your oven to ~75C.
4. When oven has heated, put the card into the oven.
5. Keep the card in the oven for ~5minutes. (ramsinks up)
6. Take the card out from the oven, and use some plastic thing over your finger and press the ramsinks again some time. Beware: the card is hot.
7. Let it cool some time.
8. Put the card to the system.

MoJo · Post by **MoJo** » Fri Aug 22, 2008 9:28 am

I too recommend cleaning the RAM chips carefully before sticking on the ramsinks. I found they would stick without problems after cleaning. I used "isoclean", basically 99% alcohol computer/electronics cleaning solution in a spray can. No baking required afterwards

thejamppa · Post by **thejamppa** » Fri Aug 22, 2008 1:45 pm

I soaked ram sinks first about an hour in isoprophyil alcohol after peeling ram sinks. Then I would use a flat and hard object, like wooden spade or back side of plastic comb to remove the sticky stuff.

Then I applied arctic clean removal material and use Q-tip to remove smaller residues. After that i used Q-tip and surface purifier from Arctic Clean.

After that I cut 3m double sided thermal tape on suitable sized and attached them on ram sinks while pressing about 15 - 30 seconds hard.

3M thermal tape is pretty good. If contact surfaces are clean they will hold nicely.

MoJo · Post by **MoJo** » Thu Sep 04, 2008 10:06 am

A little update.

I good solution I found is to underclock the core to 500MHz, memory to 225MHz and drop the voltage as low as it will go. With the stock cooler and fan settings the 4870 then idles at 30C

Power consumption is just 6.5A in this mode - 35 to 40W which is not as low as some cards but still much better than current generation GeForces.

The problem with doing this via the BIOS was that it would flicker when changing to 3D mode. There are "low power 3D" and "high power 3D" settings. The card switches to "low power 3D" quite a bit in just general Windows usage, so the flickering is a problem. The solution is to simply set the low power 3D mode to be underclocked and undervolted as well.

In my testing it works fine. Video plays back OK, everything works as expected. The performance of this card is so good that even highly underclocked like this it still performs more than well enough for everything I can throw at it. Maybe if I was using 3D Studio or something it would be an issue, but I don't. You can always use software to force it into performance mode if you need to. And of course, if you run a game, it switches automatically.

rpsgc · Post by **rpsgc** » Mon Oct 06, 2008 1:08 am

I know this topic is old (and primarily about the HD4850) but I've got to post this...

http://www.bit-tech.net/hardware/2008/1 ... 870-1gb/11

As we saw with the HIS Radeon HD 4850 IceQ 4 TurboX last week, the Radeon HD 4870 1GB uses less power than the reference card (in this case the 4870 512MB). We were so surprised by these numbers that we re-ran our power consumption tests with the Radeon HD 4870 512MB, only to find that they were indeed accurate.

Sure, their methods are not the most accurate around but still...

It's a 18W difference at idle and 10W during load.

Kaleid · Post by **Kaleid** » Mon Oct 06, 2008 1:10 am

A small update from me too. RE: coil whine.

Seems the fault lay on the Zalman 850w unit which whined when loading harddrives heavily. I replaced it and after that the Sapphire HD4850 stopped whining as well. It is now as quiet as my old and now sold Gecube 3870.

MoJo · Post by **MoJo** » Mon Oct 06, 2008 9:18 am

rpsgc wrote:Sure, their methods are not the most accurate around but still...
It's a 18W difference at idle and 10W during load.

Most like it's just due to an updated BIOS and slightly lower voltage. See here:

http://forums.techpowerup.com/showthrea ... 64&t=67928

I am using that BIOS on a reference card and it works flawlessly. 45C at idle with the stock cooler, much better than the 85C it was idling at!

rpsgc · Post by **rpsgc** » Mon Oct 06, 2008 9:37 am

Good news nonetheless.

EDIT: Also, does that make the stock cooler quieter?

line · Post by **line** » Tue Oct 07, 2008 5:41 pm

ATI promises board partners much better PowerPlay in the next architecture (RV870).

http://www.nordichardware.com/news,8196.html

I hope that doesn't mean they give up on reviving it in the 4800 series.

MoJo · Post by **MoJo** » Wed Oct 08, 2008 10:35 am

The stock cooler is pretty quiet anyway, and I think the new BIOS is tweaked to make it even better. You can also use custom profiles in Catalyst to set it lower.

rpsgc · Post by **rpsgc** » Wed Oct 08, 2008 10:55 am

MoJo wrote:The stock cooler is pretty quiet anyway, and I think the new BIOS is tweaked to make it even better. You can also use custom profiles in Catalyst to set it lower.

Thanks. Now I'll just have to decide whether or not it's worth to buy an aftermarket cooler. As long as it's quieter than the noisiest component of my PC (HX520, 22dB according to SPCR) I'm good.

Now if only SPCR would review it *hopeful*

nutball · Post by **nutball** » Wed Oct 08, 2008 11:38 am

line wrote:I hope that doesn't mean they give up on reviving it in the 4800 series.

I'd give up hope if I were you, if you're referring to the 4850 and 4870. Read the relevant threads at B3D if you want a better perspective.

MoJo · Post by **MoJo** » Wed Oct 08, 2008 12:18 pm

The 4870 with the modded BIOS isn't bad at all. I don't have a 3870 to compare it to but the BIOS supports dropping the voltage and clock rates. I'll have to try and do some tests to measure how much power exactly is saved.

Kriz · Post by **Kriz** » Wed Oct 08, 2008 12:27 pm

I've had some recent adventures with a reference based MSI HD4870 512Mb with a modified bios from the techpowerup forum.

PC Configuration: Link

HD4870 (2D) idle clocks: 550MHz GPU / 225MHz RAM / 1.083v
HD4870 (3D) full clocks: 750MHz GPU / 900MHz RAM / 1.263v

Power testing with the Seasonic SS-300SFD 300W PSU...

Using onboard GPU: 41W Idle / 69W CPU Load (Prime95 Small FFTs)
Using MSI HD4870: 68W Idle / 96W CPU Load (Prime95 Small FFTs) / 180W Peak Load (playing DiRT)

= 27W consumption at idle.

When I first put this card in I was shocked to see the system idling at 108W! So, using a BIOS from this extremely helpful techpowerup forum thread, I have been able to save 40w at idle with this configuration.

MoJo · Post by **MoJo** » Wed Oct 08, 2008 2:39 pm

27W at idle is pretty good for high end card I think. Probably at least on a par with the 3870, if not a bit better from the looks of it.

Kriz - are you using the Fastmix bios? Correct me if I'm wrong but it does not lower the memory clock (to avoid flickering and issues with video decoding etc) but does lower GPU clock and voltage. With it my idle temps dropped from 85C to 45C.

Or are you using a different one?

Kriz · Post by **Kriz** » Wed Oct 08, 2008 8:01 pm

Yes you're right. I'm using the Fastmix bios contained within the 1083.zip file, and ATI Tray Tools to automatically switch between the 2D and 3D clocks I listed above.

From reviews over at xbitlabs and SPCR the 3870 appears to use somewhere between 18.7-20W AC at idle, so an extra 7-8.3W for the 4870 is a reasonable amount considering it's relative performance.

davemuk · Post by **davemuk** » Wed Oct 08, 2008 11:41 pm

Does anyone know how low you can get the 4850 to idle at ?

TIA,
Dave

MoJo · Post by **MoJo** » Thu Oct 09, 2008 3:06 am

Kriz wrote:Yes you're right. I'm using the Fastmix bios contained within the 1083.zip file, and ATI Tray Tools to automatically switch between the 2D and 3D clocks I listed above.

Ah, okay, I was doing that too but ran into a few problems. Video playback with the VRM9 renderer or with HD stuff often just didn't work, with the media player closing after a few seconds. MAME32/MAME+ would not work with DX9 either.

In the end I decided to not bother with it, for the extra hassle it causes.

Cistron · Post by **Cistron** » Thu Oct 09, 2008 5:43 am

Kriz wrote:= 27W consumption at idle.

It is more, because your calculation ignores the IPG idle consumption.

rpsgc · Post by **rpsgc** » Fri Oct 10, 2008 7:33 am

I'm sorry to go off-topic here but really I've got to get this off my chest.

Look at this:

Who in their right mind would say they are of the same size?

walle · Post by **walle** » Fri Oct 10, 2008 9:17 am

rpsgc wrote:Who in their right mind would say they are of the same size?

No one, though thatâ€™s not to say that the fellow human wouldnâ€™t be in his/her right mind, as such; just a wee bit biased thatâ€™s all.

rpsgc · Post by **rpsgc** » Sat Oct 11, 2008 7:44 am

So... what do you all think about those new Dark Knight graphics cards from Asus?

http://www.asus.com/news_show.aspx?id=13024

Pic1
Pic2

The Powercolor PCS+ has a similar cooler and it has some bad VRM cooling issues, could this one suffer from them as well?

This one has a bigger cooler but still...

Kriz · Post by **Kriz** » Sat Oct 11, 2008 12:01 pm

MoJo wrote:Ah, okay, I was doing that too but ran into a few problems. Video playback with the VRM9 renderer or with HD stuff often just didn't work, with the media player closing after a few seconds. MAME32/MAME+ would not work with DX9 either.

In the end I decided to not bother with it, for the extra hassle it causes.

I'll check that out during the next week and see if I have the same issues, but so far nothing yet.

Cistron wrote:
Kriz wrote:= 27W consumption at idle.
It is more, because your calculation ignores the IPG idle consumption.

Does the IGP use a significant amount when idling at the Vista desktop? Or more importantly, is the IGP completely disabled when a discrete GPU is used instead?

Well, I'm glad you mentioned this as I was checking over my numbers again and found a problem with the HD4870 results I posted. The BIOS and NIC speeds weren't consistent between the tests. So I've started again with the BIOS defaults, disabled ports I don't use and lowered the CPU voltage to 1.1V

Test 1: Without network cable connected.

38W / 65W = Onboard GPU - Idle/CPU Load (Prime95 Small FFTs)
40W / 67W = PCI S3 ViRGE - Idle/CPU Load (Prime95 Small FFTs)
67W / 94W = PCI-E HD4870 - Idle/CPU Load (Prime95 Small FFTs)

Test 2: Network cable connected and the NIC working at 1Gbit.

40-41W / 68W = Onboard GPU - Idle/CPU Load (Prime95 Small FFTs)
42W / 69W = PCI S3 ViRGE - Idle/CPU Load (Prime95 Small FFTs)
69W / 96-97W = PCI-E HD4870 - Idle/CPU Load (Prime95 Small FFTs)

Yes, I used an S3 ViRGE PCI card from 1995 to compare against the power use of the onboard GPU. Amazingly Vista didn't complain too much about it, but I did have to run all the GPUs with Aero off at 800x600x16bit just to be fair.

Test 1 shows us a 29W difference between the IGP and the HD4870 now, so 27W was definitely wrong. Test 1 also shows that a S3 ViRGE requires 2W more power than whatever the IGP is consuming. I think if the ViRGE had actually become warm while it was being tested then maybe I'd believe that. No, I think that there are probably some inefficiencies with this motherboard chipset design when using a discrete GPU instead of the onboard solution, or at least when using a card as ancient as this one.

Test 2 appears to confirm my original suspicions about the BIOS configurations and the effect the old network cable had by not always working at 1Gbit.

Something that still confuses me is the 27W difference between all the Idle/Prime95 results. I don't understand it. As far as I know, my PSU is not as efficient between 38W/65W as it is between 67W/94W. There should be a larger difference at the lower watts at least.

Alright, I've officially spent way too much time on this.

thejamppa · Post by **thejamppa** » Sat Oct 11, 2008 2:31 pm

Kriz wrote:
Alright, I've officially spent way too much time on this.

Not really, your post, well made me remember Felger Carbon, may he rest in peace. He was famous of making all kinds of tests and conducting experiements.

And there's not such thing as spending too much time on something you like. Nice test and good work ^^

MoJo · Post by **MoJo** » Sat Oct 11, 2008 3:15 pm

Kriz wrote:I'll check that out during the next week and see if I have the same issues, but so far nothing yet.

It wasn't every time, only sometimes. I never pinned down the exact circumstances.

One thing I noticed about ATI Tray Tools as well - it doesn't reset the clocks properly when leaving sleep mode.

Does the IGP use a significant amount when idling at the Vista desktop? Or more importantly, is the IGP completely disabled when a discrete GPU is used instead?

At the moment AIUI the IGP is completely disabled and powered down when a discrete card is used. Unfortunately, with Hybrid SLI the opposite is not true, the discreet card only idles but is not actually powered off.

38W / 65W = Onboard GPU - Idle/CPU Load (Prime95 Small FFTs)
40W / 67W = PCI S3 ViRGE - Idle/CPU Load (Prime95 Small FFTs)
67W / 94W = PCI-E HD4870 - Idle/CPU Load (Prime95 Small FFTs)

29W for a top end graphics card is a price worth paying if you want more than the IGP, IMHO. Close to what the 3870 managed too.

walle · Post by **walle** » Sat Oct 11, 2008 6:03 pm

rpsgc wrote:...has a similar cooler and it has some bad VRM cooling issues, could this one suffer from them as well? This one has a bigger cooler but still...

The size of the cooler doesnâ€™t matter here since it only handles the GPU heat and not the VRM, I have one HD 4870 myself and the GPU is not a problem and touch on wood; nor has the VRM been, up to this point, but, the VRM do run very hot.

If you want a cooler that also cools the VRM (aftermarket cooler) then (for example) Thermalrights T-rad could be an good choice, given you would use two fans, that way they would cover the entire card whilst blowing â€œcoolâ€