Radeon HD 4670: A perfect balance?

Ch0z3n · Post by **Ch0z3n** » Mon Jun 01, 2009 4:52 am

That gets done unofficially when a site pisses off the OEM and they don't get any review samples.

Hrmms, well SPCR isn't really a file sharing site, but I don't see why it couldn't be posted on one. Then people could flash their 4670s with gimped review sample firmware to save power =)

Ksanderash · Post by **Ksanderash** » Mon Jun 01, 2009 7:06 am

If only our hardware is capable, and that is ...alas and alack!

thejamppa · Post by **thejamppa** » Thu Jun 18, 2009 4:52 am

Asus' EAH4670 HDMI 512MB GDDR3 has working powerplay. Voltage drops to 0,90 V from 1,25v when idled. However its cooler is horrible, so that needs to changed.

jaylin · Post by **jaylin** » Thu Jun 18, 2009 6:52 am

I joined just to share this info:

My MSI r4670 512mb that I bought last october from newegg has working powerplay. This is not just a change in hwmonitor. The following are my observations measured with my kill-a-watt system w/o monitor:

2d mode undervolted from 1.15v->.90v - no noticeable difference in watts saved at idle but I get random blue screens.

2d mode undervolted from 1.15v->.95v - no more blue screens.

3d mode undervolted from 1.25->1.15v - Used to be able to hit 800+ core, now my max stable gpu core clock is 750. A savings of ~10watts at load running Left 4 Dead.

my next step is tweaking uvd volts to see if I can save some watts when watching movies.

Ch0z3n · Post by **Ch0z3n** » Thu Jun 18, 2009 8:55 am

how do you get blue screens without a monitor?

Ok, so what you did was take a card with working PowerPlay (voltage and frequency) and reduced the voltage settings on the PowerPlay profiles even further?

What about core/memory frequency in 2d?

jaylin · Post by **jaylin** » Thu Jun 18, 2009 9:25 pm

Ch0z3n wrote:how do you get blue screens without a monitor?

Ok, so what you did was take a card with working PowerPlay (voltage and frequency) and reduced the voltage settings on the PowerPlay profiles even further?

What about core/memory frequency in 2d?

what i meant by without monitor is that the power measurement is only of the system, not system+monitor.

Yes, I reduced the voltage settings on the powerplay profile using RBE.

The 2d clocks throttled to 165/250 out of the box. I did not change anything regarding that.

For comparison with the 4770 I just got today:
msi 4670 - 165/[email protected] - 101watts idle
diamond 4770 - 250/[email protected] - 125watts idle. +8 watts at 800mhz mem.

Same system specs. only difference is video card.

Nil Einne · Post by **Nil Einne** » Wed Jul 01, 2009 3:47 pm

MoJo wrote:I'm starting to think that the only way review sites can be sure of not being cheated by manufacturers is to buy the cards retail themselves.

Actually, there is one other option. Post the BIOS file so it can be compared with retail ones. Any chance of that, Mike?

Has there been any actual evidence to show the review card had some sort of special firmware? There was some suggestion that the 4670 review sample had reduced shaders, I can't remember the review now and am lazy to look at it again but were no benchmarks done? If they were, then if shaders were reduced this would show up in them. For this reason, I doubt AMD/ATI did this. It's definitely not something they did for most of their review samples.

ATI/AMD is not disabling features because they own shares in power companies so like to waste power. So unless there were some stability issues that Mike didn't noticed that forced ATI/AMD to change the power saving modes, I don't see any other way the bios could be an issue.

The bigger issue is cherry picked cards (or whatever) with the best result. It's a well accepted fact that manufacturers do this. This is why you trust overclocking results from professional reviewers with a grain of salt. For SPCR, given their focus, it seems logical you'd choose different things like lower power consumption and noise. I suspect this is far more likely to be the cause. Or perhaps just a mistaken in the testing.

As has been mentioned, there are obviously some issues that are resulting in many cards not having great idle. Potentially the inability for most cards to undervolt, which is almost definitely a result of changes from the reference design probably to save cost. Also there may be other circuitry differences. As mentioned somewhere (was it this thread?) the sad fact this is something few users will test or even be aware of so most manufacturers are not going to bother with it. A difference between 16W and 4W while not insignificant in power saving terms is only going to have a tiny effect on the noise level so that's not a compelling reason either.

MoJo · Post by **MoJo** » Thu Jul 02, 2009 2:58 am

Nil Einne wrote:Has there been any actual evidence to show the review card had some sort of special firmware? There was some suggestion that the 4670 review sample had reduced shaders, I can't remember the review now and am lazy to look at it again but were no benchmarks done? If they were, then if shaders were reduced this would show up in them. For this reason, I doubt AMD/ATI did this. It's definitely not something they did for most of their review samples.

I think it's less about disabling features and more about hand picking the best available sample and tweaking the BIOS so that it gets maximum power saving.

It's been known to happen with sites that overclock review samples. The manufacturer tests a few cards or chips to see which one overclocks the most, and then sends that out for review. IIRC Tom's Hardware got stung that way with Celeron CPU a few years back - their sample could reach over 3GHz (more than 100% overclock) but most retail chips could get nowhere near that.

So unless there were some stability issues that Mike didn't noticed that forced ATI/AMD to change the power saving modes, I don't see any other way the bios could be an issue.

The main issue is with GDDR5. On cards with GDDR5 you can't change the memory clock without the screen flickering, so they all have memory underclocking disabled.

Klusu · Post by **Klusu** » Thu Jul 02, 2009 10:49 am

The noise does not bother the manufacturers.
"A difference between 16W and 4W...tiny effect on the noise level"... There is no effect, because the fan spins at the same speed (usually).
I tried to fix the speed adjustment on Gigabyte 512MB. I found several missing transistors and resistors (they could "save" a couple more transistors, why didn't they?), but no success, perhaps I did not find the last resistor... I have a Sapphire 512 card also, which does adjust the fan. I can compare both cards, they are almost the same, but I did not find where the control signal comes from ( I think - from the GPU). (Btw, there is a trace to the place marked SW4.)
Both cards can go down to 0.9V GPU. The Gigabyte card goes down to 1.1V, The Sapphire card is always at 1.25V. Why? Because it is easier so. The manufacturer does not want to test the cards at lower voltages. There is no saving in hardware cost.
Some sort of special firmware? Apparently, the review card did go down to 0.9V. Do the mass production cards go down? Perhaps not.
The legendary 3W card probably drew a watt or two more. Nobody measured the card's draw directly. Rather the difference - the computer with the card versus the computer with integrated graphic only.

MoJo · Post by **MoJo** » Thu Jul 02, 2009 11:48 am

Klusu wrote:Rather the difference - the computer with the card versus the computer with integrated graphic only.

Someone needs to test with a Rage XL 8mb PCI card. They use less than 1W at idle and so provide a good benchmark against other cards, and take the on-board graphics out of the equation.

Perhaps SPCR should look into getting one off eBay.

Klusu · Post by **Klusu** » Thu Jul 09, 2009 5:22 am

I tried once more. I fixed fan control on Gigabyte card. I could not find the last missing part, I added my own part (a resistor between the control trace and +12V).
The picture I posted earlier:
http://www.bildites.lv/images/i1hsisrq0zsjq7nre3.jpg
Two places for transistors are marked. The left one is not necessary, if you use a 2 pin connector. The right one's base is the control signal, I added here 27kÎ© to +12V. The collector needs 10kÎ© to +12V (the place is between the two marked transistors) and 10kÎ© to the next (missing) transistor's base.
I wrote earlier: "My computer with this card at 165/250 0.9V draws from the wall 7.5W more than without it (just the integrated gf6100)"
The same card, 165/250 0.9V, this time without the fan (1W DC), another computer, 8.6W more than the integrated 690G.
I measured motherboard's draw from 12V, idle, these numbers should be more reliable:
1.4W just the 690G
Difference with the Gigabyte 512MB (no fan):
5.5W 100/125 0.9V (the lowest I can set)
5.8W 165/250 0.9V
6.1W 165/500 0.9V
7.1W 165/1000 0.9V (hangs)
5.8W 250/250 0.9V
5.9W 300/250 0.9V
7.0W 165/250 1.1V
7.3W 165/500 1.1V
Difference with Sapphire 512 (no fan):
6.4W 165/350 0.9V (300MHz memory-bad picture)
6.6W 165/500 0.9V
6.6W 250/500 0.9V
8.4W 250/500 1.1V
There is some draw from the 3.3V also, but the 12V part is the main part, GPU+memory.
More numbers, Sapphire. Bigger a little, probably because the card was hotter (for example, 750/1000 1.25 11.8W at 50Â°C,12.5W at 65Â°C):
6.7W 165/500 0.9
8.4W 165/500 1.1
9.5W 165/500 1.25
6.8W 300/500 0.9
8.8W 300/500 1.1
9.8W 300/500 1.25
7.8W 300/1000 0.9
9.8W 300/1000 1.1
11.0 300/1000 1.25
10.6 750/1000 1.1
11.9 750/1000 1.25

Nil Einne · Post by **Nil Einne** » Thu Jul 09, 2009 5:57 am

Ch0z3n wrote:I have kind of given up on the 'perfect card'. Sadly, no one seems to be able to come close. The 4770 would be awesome if they did some more work on the idle power consumption and maybe take off the 6-pin connector since it only uses ~60w... oh, and being able to buy them anywhere would be awesome.

Someone did do the 4830 without 6 pin, so perhaps they'll do it for the 4770 [http://www.techpowerup.com/91388/Triple ... ector.html]. There's also 9600s I believe.

The passives have been demoed http://www.tcmagazine.com/comments.php?shownews=26958
http://www.techpowerup.com/95987/PowerC ... cards.html
http://www.techpowerup.com/91388/Triple ... ector.html
but do have the 6 pin.

Nil Einne · Post by **Nil Einne** » Thu Jul 09, 2009 6:16 am

MoJo wrote:I think it's less about disabling features and more about hand picking the best available sample and tweaking the BIOS so that it gets maximum power saving.

I don't really see what there is to tweak in the bios though. Perhaps there are some settings we're not are of but it seems to me, the most likely thing is just the PowerPlay states. Perhaps the card was 100mhz GPU and 125mhz RAM on idle with the 0.9V but this is something you can do (or at least try) by modifying your bios (if you have 0.9V working). You don't need a special bios for that. If the 0.9V doesn't work on your card, then that's an issue as I mentioned, but it's not related to the bios.

In other words, the way I see it, we don't need the bios from SPCR (I'm not saying it shouldn't be provided simply that I doubt there's anything particularly special about this bios beyond well developed PowerPlay states). Ideally SPCR reviews should provide PowerPlay state information from the bios but that's about all we need. The big issue remains in the hardware IMHO i.e. cherry picked cards as I already mentioned (as well as the problem I mentioned of some cards not having 0.9V working). As I've stated, I especially doubt there are any issues with disabled shaders.

Indeed given the results Klusu's is getting I'm not even convinced there is a big difference between his card and the SPCR one. The difference could easily be within the margin of error as he noted.

BTW, Klusu, any chance we could get some pictures showing the components that were missing?

MoJo · Post by **MoJo** » Thu Jul 09, 2009 8:06 am

Nil Einne wrote:I don't really see what there is to tweak in the bios though. Perhaps there are some settings we're not are of but it seems to me, the most likely thing is just the PowerPlay states. Perhaps the card was 100mhz GPU and 125mhz RAM on idle with the 0.9V but this is something you can do (or at least try) by modifying your bios (if you have 0.9V working). You don't need a special bios for that. If the 0.9V doesn't work on your card, then that's an issue as I mentioned, but it's not related to the bios.

I imagine that was probably something to do with it. The normal BIOS has to be quite conservative because it has to work with every chip, but as with overclocking you sometimes find that a particular one will run okay well out of spec.

All they had to do was test a few cards to see how low they would go, pick the best one and modify it's BIOS to downclock as much as possible.

On top of that they could also pick better quality components for the review sample, such as more efficient voltage regulators and transistors.

Nil Einne · Post by **Nil Einne** » Fri Jul 10, 2009 10:50 am

Has anyone ever come across any evidence that there's any use testing the intermediate voltages with RBE if 0.9V doesn't seem to do anything? In my HIS iSilence4 4670, I've tried with 0.9V but it doesn't seem to have an effect, the temperatures (I believe idle although didn't actually wait for it to reach equilibrium, definitely under load) are the same (this is a passive card) and more significantly perhaps, the card has no problems even under stock settings (750, 850) with Furmark which suggests to me the voltage is not reduced (or if it is, not by much). Either that or I have a very good chip which can handle such a low voltage, but I think most would agree that's very unlikely.

Don't have a power plug watt measuring device so can't test power usage and don't dare try measure voltage off the card (don't even know how).

I'm just wondering if it's possible the way the card is designed if 0.9V is not available then it will use 1.25V but 1.1V or 1.0V or whatever works

Cheers

Nil Einne · Post by **Nil Einne** » Sat Jul 11, 2009 5:57 am

davidh44 wrote:I got an HIS 4670 Turbo IceQ based on reading reviews that it was very quiet. At idle, I find that it's too noisy for a quiet PC. I probably should've gotten the silent version with stock Zalman heatsink.

You may want to look at the Powerplay states and considering modifying them if you have no qualms about bios flashing. I read in one review/forum that the IceQ (think it was 48xx) Powerplay was very limited reduction so the card was noiser at idle then a card with stock HSF

Nil Einne · Post by **Nil Einne** » Sat Jul 11, 2009 10:59 am

Klusu wrote:I tried once more. I fixed fan control on Gigabyte card. I could not find the last missing part, I added my own part (a resistor between the control trace and +12V).
The picture I posted earlier:
http://www.bildites.lv/images/i1hsisrq0zsjq7nre3.jpg
Two places for transistors are marked. The left one is not necessary, if you use a 2 pin connector. The right one's base is the control signal, I added here 27kÎ© to +12V. The collector needs 10kÎ© to +12V (the place is between the two marked transistors) and 10kÎ© to the next (missing) transistor's base.
I wrote earlier: "My computer with this card at 165/250 0.9V draws from the wall 7.5W more than without it (just the integrated gf6100)"
The same card, 165/250 0.9V, this time without the fan (1W DC), another computer, 8.6W more than the integrated 690G.
I measured motherboard's draw from 12V, idle, these numbers should be more reliable:
1.4W just the 690G
Difference with the Gigabyte 512MB (no fan):
5.5W 100/125 0.9V (the lowest I can set)
5.8W 165/250 0.9V
6.1W 165/500 0.9V
7.1W 165/1000 0.9V (hangs)
5.8W 250/250 0.9V
5.9W 300/250 0.9V
7.0W 165/250 1.1V
7.3W 165/500 1.1V
Difference with Sapphire 512 (no fan):
6.4W 165/350 0.9V (300MHz memory-bad picture)
6.6W 165/500 0.9V
6.6W 250/500 0.9V
8.4W 250/500 1.1V
There is some draw from the 3.3V also, but the 12V part is the main part, GPU+memory.

Looking more closely at your results, the interesting thing to me is that the difference between 165 and 300 at 0.9V is minimal.

Did you test say 165 vs 300 at 1.25V or has anyone seen such a test? I'm just wondering whether from a pure power consumption POV, there's actually much reason to go so low...

kiwijunglist · Post by **kiwijunglist** » Mon Jul 13, 2009 2:21 am

I'm confused does the PowerColor 1GB HD4670 (The version with 2xDVI) support variable speed fan control using the 2 wire fan pin. Same state it does, same state it doesn't.

This is for HTPC rig, hooked up to 1920x1080P LCD TV. I want a quiet setup for hardware accelerated 1080i/1080P, doesn't matter if the fan is audible if i'm gaming (which would be very rare as i have a gaming pc for that). Does the fan spin up when doing hardware deinterlacing or heavy DXVA?

Which HD4670 support variable fan speed? I'm not talking just having the fan slider in CCC (which doesn't necessarily mean the fan is variable), I'm talking about the fan actually changing RPM with the slider/temperature.

SebRad · Post by **SebRad** » Mon Jul 13, 2009 4:40 am

Hi, if you have the same power color 1GB HD4670 that I do then yes the fan speed is varied in responce to GPU core temp. Being only a 2 wire fan there is no rpm monitoring though. In stock form it's not bad at idle, you can reduce the fan speed further in CCC or RivaTuner or the like and have it pretty quiet. Under heavy load the cooler is quite loud (by SPCR standards)
My card does not have any underclocking or undervolting in the BIOS. When I added some in the BIOS (for some reason RBE1.21 doesn't work but RBE1.20 does) I then got horrible flickering on my 2nd screen.
Hope this helps, Seb
Edit - fixed link to right thread

Klusu · Post by **Klusu** » Tue Jul 14, 2009 11:02 pm

Added some numbers to my previous post.

Nil Einne wrote:..any use testing the intermediate voltages ...

Probably no.

jaylin wrote: ...my next step is tweaking uvd volts to see if I can save some watts when watching movies.

I tried to reduce clock and volts. No picture.

kiwijunglist · Post by **kiwijunglist** » Sun Jul 19, 2009 12:11 am

@SebRad - Thanks

So which one do you reckon i should get. This is for HTPC Silverstone LC16-M Case running 1920x1080 LCDTV.

Powercolor HD4670 512mb - NZ$149
Powercolor HD4670 1024mb - NZ$159
ASUS EAH4670 512mb - NZ$163
HIS HD4670 IceQ 1024mb - NZ$165

Prefer to run stock without modding bios. Mostly for 1080P/1080i HDTV/MKV viewing with mediaportal

thejamppa · Post by **thejamppa** » Sun Jul 19, 2009 2:43 am

The Asus one has fixed fan speedo f 2000 rpm's and its loud, had horrible sound quality and tonality. You definately want to change that so maybe your best bet it is His? But Asus card drops core voltage from 1,25v to 0,9v. I don't know about HIS'

kiwijunglist · Post by **kiwijunglist** » Sun Jul 19, 2009 10:28 pm

HIS IceQ Pro = 24.3 dbA @ Idle -> 29 dbA @ Full load
Powercolor = 21 dbA @ Idle -> 24.2 dbA @ Full Load

1kbyte · Post by **1kbyte** » Sun Aug 02, 2009 9:19 am

My Biostar 4670 was burnt 2 month ago, caused by Zalman VNF100. I check throughly on both installation and manual after this "incident" but cannot find anything wrong

.... anyone had same experience?

thejamppa · Post by **thejamppa** » Sun Aug 02, 2009 9:36 am

1kbyte wrote:My Biostar 4670 was burnt 2 month ago, caused by Zalman VNF100. I check throughly on both installation and manual after this "incident" but cannot find anything wrong .... anyone had same experience?

My HD 4670 has run perfectly for about 4 month's now. No problems what so ever. Maybe something short circuited? I had to file off bit Zalman's bottom so it wouldn't touch one of the capacitors.

davidh44 · Post by **davidh44** » Mon Aug 03, 2009 8:44 pm

thejamppa wrote:The Asus one has fixed fan speedo f 2000 rpm's and its loud, had horrible sound quality and tonality. You definately want to change that so maybe your best bet it is His? But Asus card drops core voltage from 1,25v to 0,9v. I don't know about HIS'

HIS does drop to 0.9v at idle, and the fan is quiet. But not quiet enough for my taste. I tried to use RBE to adjust fan speed, but the "fan settings" button is greyed out. Does that mean I'm out of luck?

thejamppa · Post by **thejamppa** » Tue Aug 04, 2009 1:54 am

davidh44 wrote:
thejamppa wrote:The Asus one has fixed fan speedo f 2000 rpm's and its loud, had horrible sound quality and tonality. You definately want to change that so maybe your best bet it is His? But Asus card drops core voltage from 1,25v to 0,9v. I don't know about HIS'
HIS does drop to 0.9v at idle, and the fan is quiet. But not quiet enough for my taste. I tried to use RBE to adjust fan speed, but the "fan settings" button is greyed out. Does that mean I'm out of luck?

Most likely. But try rivaturner and speedfan also just to verify there's no real fan control.

Ksanderash · Post by **Ksanderash** » Wed Aug 12, 2009 4:44 am

Klusu
Thank you for testing.

...

And you helped me to fix native thermo-control scheme ) I added two 3904 transistors and two resistances. There are two independent schemes, for 4 pin PWM and a classical 2 pin FAN.

macros · Post by **macros** » Tue Sep 08, 2009 7:20 pm

Hi All,

I'm looking at buying a discrete graphics card for my HTPC, and am wondering if the Gigabyte HD4670 1GB GDDR3 has fan speed control?

If not, any one have any experience with how noisy the fan is?

I'm sticking it in a Silverstone LC16M case, which will be at least 2m from the closest person. Last thing I want is a whine coming from the case while watching a movie!

Cheers, Cameron

thejamppa · Post by **thejamppa** » Tue Sep 08, 2009 11:07 pm

Hi Macros!

What I've heard, Gigabyte's fan is fixed speed. You better avoid that or get something like His iSilence 4 series passively cooled 4670. Or then you can change the stock cooler into passive cooler like Zalman VFN-100 HP which fits and is capable of cooling HD 4670 in case with good ventilation

Radeon HD 4670: A perfect balance?

Quick question

Re: Quick question

Gigabyte HD4670 1GB GDDR3 has fan speed control?