Radeon HD 4670: A perfect balance?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
If you wish to use an Accelero S2 with the Palit 4670 you will have to remove the VGA socket and the associated ribbon cable. I was considering getting one but I'm using the VGA so didn't. (This is based on printing this plan and comparing it directly to the card). You *might* also have to reduce the height of one of the ramsinks to fit it under one of the heatpipes.
I disconnected the fan in the end and stuck an 80mm panaflo directly to the heatsink with double sided tape, it increases the height by another inch though.
qpegasus, if your card sounds like a vacuum maybe the fan on it is faulty? Before I changed mine it was not silent, but it was quiet at low speeds, 23% was best due to a whine at 20%. (I know this is subjective and we all perceive noise differently)
I disconnected the fan in the end and stuck an 80mm panaflo directly to the heatsink with double sided tape, it increases the height by another inch though.
qpegasus, if your card sounds like a vacuum maybe the fan on it is faulty? Before I changed mine it was not silent, but it was quiet at low speeds, 23% was best due to a whine at 20%. (I know this is subjective and we all perceive noise differently)
Ive confirmed that it is the 4670 fan by disconnecting it and booting up. Ironically, I could hear a vacuum running in my neighbours house, which I couldnt before.
Its not quiet by anyones standards, so maybe it is the fan or the fan voltage controller at fault. Perhaps even a different model with a different heatsink/fan from yours?
Is it possible/safe to hook the fan to one of the mobo fan connectors and control the voltage from there?
I also have a spare Nexus 120mm (much too big methinks) or an old Akasa 80mm I could try rigging it up with.
Its not quiet by anyones standards, so maybe it is the fan or the fan voltage controller at fault. Perhaps even a different model with a different heatsink/fan from yours?
Is it possible/safe to hook the fan to one of the mobo fan connectors and control the voltage from there?
I also have a spare Nexus 120mm (much too big methinks) or an old Akasa 80mm I could try rigging it up with.
I think Palit only do one 4670, but I could be wrong - Mine has a double height fake copper cooler with a small 2-wire fan and VGA, DVI, HDMI and Displayport on a double height backplate.
I run an 80mm Panaflo off the VGA card fan header, so it should work the other way round. The only problem is that although the pin spacing for the fan connector is the same, the plastic housing is keyed differently so it might not physically fit on a mobo header. I got around this by removing the plastic shroud around the pins on the 4670 PCB and just using friction to hold the new fan connector on.
Have you given up on the idea of an Accelero?
I run an 80mm Panaflo off the VGA card fan header, so it should work the other way round. The only problem is that although the pin spacing for the fan connector is the same, the plastic housing is keyed differently so it might not physically fit on a mobo header. I got around this by removing the plastic shroud around the pins on the 4670 PCB and just using friction to hold the new fan connector on.
Have you given up on the idea of an Accelero?
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
Yeah, exactly so. And the checking point is the positive lead of any of the three capacitors that are sitting near the two ferrite coils. See the pic for details:maf718 wrote:@ Ksanderash : I have a 4670 and a digital multimeter, is this all I would need to read the voltage from the card, and if so where are the measuring points?
Thanks.
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
Undervolting mod for my Force3D 4670. If you have absolutely(!) similar board's design around the uP6201BQ chip, then it may do for you too, most probably.
...
And, hm... I also found that FB signal goes to a chip that is missing on my card... Here is IXBT's(X-bit labs) reference card:
So, in my opinion, there is a strong probability that if your card misses this chip too, you don't have VID shifting on the fly as the reference card has! Therefore no 0.9V in idle, and no SPCR reviewed 4W ...this is my pessimistic theory.
...
And, hm... I also found that FB signal goes to a chip that is missing on my card... Here is IXBT's(X-bit labs) reference card:
So, in my opinion, there is a strong probability that if your card misses this chip too, you don't have VID shifting on the fly as the reference card has! Therefore no 0.9V in idle, and no SPCR reviewed 4W ...this is my pessimistic theory.
Voltage results for my Palit 4670:
Idle ~ 1.248v and load (furmark) ~ 1.3v
The card has a slightly different design to the one you pictured but this was measured off the 3 capacitors near the crossfire connectors, I am pretty sure that is right.
According to the bios I pulled off the card and viewed with RBE the figures should be 0.9v idle and 1.25v load. I guess this means that my card is also missing the chip that controls voltage switching on the fly. Video card manufacturers really are cost cutting bastards.
In fact this mirrors exactly the situation I observed at the launch of the HD 4830; all the reference cards that were reviewed had incredibly low power consumption at idle; all the retail cards (with reduced component count) that were reviewed had higher power consumption at idle.
Ksanderash, I fear your theory is correct, I wonder if there are any 4670 cards that meet the reference design?
Idle ~ 1.248v and load (furmark) ~ 1.3v
The card has a slightly different design to the one you pictured but this was measured off the 3 capacitors near the crossfire connectors, I am pretty sure that is right.
According to the bios I pulled off the card and viewed with RBE the figures should be 0.9v idle and 1.25v load. I guess this means that my card is also missing the chip that controls voltage switching on the fly. Video card manufacturers really are cost cutting bastards.
In fact this mirrors exactly the situation I observed at the launch of the HD 4830; all the reference cards that were reviewed had incredibly low power consumption at idle; all the retail cards (with reduced component count) that were reviewed had higher power consumption at idle.
Ksanderash, I fear your theory is correct, I wonder if there are any 4670 cards that meet the reference design?
I run it with a passive AC S1. The only airflow whatsoever in the entire case is a single 120mm 1200rpm Scythe S-Flex running at 800rpm. No direct airflow whatsoever. I can play COD4 all day long and it never hicups.lechuck wrote:So 8800GT -> more than 30W idle + no or very little airflow = cooking
30w isnt as much as you think.
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
Damn, right these figures I have here enjoying with my crappy Force3D The 50mV difference between 2D idle and 3D load. If you'll give me a hi-res picture of the uP6201BQ's surroundings I can say with confidence which one resistor you can touch for undervolting purpose. I'm already testing the card under 1.20/1.25V, having an evident distinction from the stock temps. But I had to lessen also GPU clock to 700MHz for stability (it is not a big deal, I think, cause the bottle neck on 4670 is DRAM bus width). UPD: 1.10/1.15V stable at 650MHz.maf718 wrote:Voltage results for my Palit 4670:
Idle ~ 1.248v and load (furmark) ~ 1.3v
Ksanderash, I fear your theory is correct, I wonder if there are any 4670 cards that meet the reference design?
Saw no any other well-known 4670 cards that have that mysterious chip besides the reference one 1.2V vs 0.9V -- that's disgrace when your are for SPCR's main doctrine.
Aris
I'm with lechuck here. I'm feeling anxious when something is heating up (even if it is very very quiet process ) in my case higher than can bear the hand.
-
- Posts: 1406
- Joined: Tue Feb 13, 2007 7:28 pm
- Location: USA
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
I'm not shure if there is 60, and not higher... But, well, on the other hand we are all constantly upgrading so it may not to be an issue. And a TYVM goes to engineers for reducing number of or even completely removing Al-caps from design In the case of something 80-95C it may last much longer.jessekopelman wrote: 60C feels very hot to bare skin, maybe unbearable, yet that is plenty fine for a video card . . .
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
3mm aluminium plate will be enough for this purpose. It is easier to drill through too. So get down to work And think about how to fix the heatsink base to prevent it sliding on the GPU cristal. Maybe it wil be 4 additional dummy screws in plate, just to limit the heatsink base sliding in a horisontal plane.Matija wrote:I've been thinking...
What if I took a metal plate (aluminium or steel), drilled holes in it, and improvised a mounting mechanism for the HR-03 to replace its default fastener? I'd just need some nuts and bolts.
Would it be secure enough?
Aluminium? Steel? 2mm? 3mm?
Or you can take two aluminium strips, screw them on default mounting bracket flankly (from each of two side), and then drill trough in these two aluminium strips the 4 openings you need.
ive had video cards in the past run over 100c at load for years without issue. temp is not what you should be concerned with when it comes to graphics cards, long as your not getting any artifacts on screen its working fine. trust me it will let you know when its having problems. you will get graphical errors on screen way before your video card overheats.jessekopelman wrote:60C feels very hot to bare skin, maybe unbearable, yet that is plenty fine for a video card . . .Ksanderash wrote: Aris
I'm with lechuck here. I'm feeling anxious when something is heating up (even if it is very very quiet process ) in my case higher than can bear the hand.
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
Think also about the glass-cloth laminated textolite (I mean the material used for motherboard, videocard, etc. production) It is as strong as the metal is, and can be easily cut.Matija wrote: So all that's left now is to find an aluminium plate, 50x50x3mm...
...
Did a DRAM voltmod. From 1.85V to 1.65V (no artefacts in games). But see no noticeable decrease of temperature. Still can't hold on the finger on chips when load.
P.S. Oops! In Tomb Raider 1.65V give corruption, have reverted to 1.80V
Last edited by Ksanderash on Sun Dec 07, 2008 5:39 pm, edited 1 time in total.
Aris: Can we stop now with this in this thread, please? Yes, it's true that VGA can handle up to 100c, but the problem is that
you must take this heat out of the PC... and by having so hot VGA, then also MB will heat up, HDD will feel the heat, etc...
No good for my setup. So runninig passive is very thin balance, where every W/degC counts.
My VGA uses 17/18W in idle - and I'd change it only to gain 10W or so...
And if there is a holy grail VGA card out there (that needs only 4W) and this thread 'discovers' it, then we all win and
could dance together in the sunset.
If anybody has an idea where could we buy ATI reference cards, would be great!
Matija: Do you have revA or first rev. of HR-03?
you must take this heat out of the PC... and by having so hot VGA, then also MB will heat up, HDD will feel the heat, etc...
No good for my setup. So runninig passive is very thin balance, where every W/degC counts.
My VGA uses 17/18W in idle - and I'd change it only to gain 10W or so...
And if there is a holy grail VGA card out there (that needs only 4W) and this thread 'discovers' it, then we all win and
could dance together in the sunset.
If anybody has an idea where could we buy ATI reference cards, would be great!
Matija: Do you have revA or first rev. of HR-03?
This is a common misconception. A video card that runs hot does not output any more heat than the exact same video card with a heat sink that runs cool.lechuck wrote: Yes, it's true that VGA can handle up to 100c, but the problem is that you must take this heat out of the PC...
The same amount of heat is outputted into the case in both instances. One just does it a bit more efficient. But the overall impact on case temps is the same with either. The only difference is the temp on the video card, but as long as the video card isn't overheating and causing on screen artifacts this isn't a problem.
Aris: We are not talking about the same issue. It's true what you're saying, but if VGA generates less heat in the first place,
than you have less heat to worry about.
If VGA needs about 20W in idle, the temp on HS in 100% passive system will be about 60 degC.
If VGA needs about 30W in idle, the temp on HS in 100% passive system will be more than 70 degC.
So if VGA needs less then 10W in idle, then we have even less impact on system...
than you have less heat to worry about.
If VGA needs about 20W in idle, the temp on HS in 100% passive system will be about 60 degC.
If VGA needs about 30W in idle, the temp on HS in 100% passive system will be more than 70 degC.
So if VGA needs less then 10W in idle, then we have even less impact on system...
The OP said nothing about a completely passive system, he merely started a thread discussing this video card and whether or not its a "perfect balance".lechuck wrote:...the temp on HS in 100% passive system... .
Perhaps you should start your own thread if you wish to discuss VGA options for a completely passive system.
Thanks, but I'm not planning on voltmodding the card. Btw I do not have uP6201BQ chip on my board, but there is one similar with different name, as well as different pcb etc. If a picture of the Palit card is any use to you for reference I can upload.Ksanderash wrote:Damn, right these figures I have here enjoying with my crappy Force3D The 50mV difference between 2D idle and 3D load. If you'll give me a hi-res picture of the uP6201BQ's surroundings I can say with confidence which one resistor you can touch for undervolting purpose. I'm already testing the card under 1.20/1.25V, having an evident distinction from the stock temps. But I had to lessen also GPU clock to 700MHz for stability (it is not a big deal, I think, cause the bottle neck on 4670 is DRAM bus width). UPD: 1.10/1.15V stable at 650MHz.maf718 wrote:Voltage results for my Palit 4670:
Idle ~ 1.248v and load (furmark) ~ 1.3v
What strikes me as ridiculous is the way the bios is set to supply 0.9v at idle to the gpu when the card cannot physically supply this voltage. It does clock down to 165mhz as evidenced by gpuz etc but the voltage remains high.
The memory on this card runs at full speed at idle, I wonder how much power I could save by lowering this clock and whether this is even possible for this card?
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
maf718
And what is there on Palit version instead of the uP6201BQ PWM-contr.? APW7098 I guess?
I'm feeling defrauded and humiliated too, trying to get lower consumption when it must be carried out by design. I will watch the market if any new 4670 reincarnation will have a resemblance with reference.
And what is there on Palit version instead of the uP6201BQ PWM-contr.? APW7098 I guess?
Hey, RBE gives you full control over the frequencies, if not voltages. I can't now measure amount of saved power, my G31-M7 is still at warranty service, but you can edit clocks by yourself. Mine has a common 165/250MHz in idle.The memory on this card runs at full speed at idle, I wonder how much power I could save by lowering this clock and whether this is even possible for this card?
I'm feeling defrauded and humiliated too, trying to get lower consumption when it must be carried out by design. I will watch the market if any new 4670 reincarnation will have a resemblance with reference.
Yay! I got two 50x50x1.5 aluminum plates, for free. The only problem is that I don't have time to mess around with the card until next weekend :/
BTW, my Sapphire doesn't play well with RBE, the whole fan tab is greyed out. Would it be safe to play with clocks and voltages? I only see a downclock to 300/1000 (from 750/1000).
BTW, my Sapphire doesn't play well with RBE, the whole fan tab is greyed out. Would it be safe to play with clocks and voltages? I only see a downclock to 300/1000 (from 750/1000).
Ksanderash: The chip in question is marked:
CH=AG
XOM
I may play around with the memory clocks in RBE and flash the card, but if all I will save is 1 or 2 watts it does not seem worth the trouble.
Matija: My Palit card also has the fan tab greyed out in RBE, even though fan control works in CCC etc.
CH=AG
XOM
I may play around with the memory clocks in RBE and flash the card, but if all I will save is 1 or 2 watts it does not seem worth the trouble.
Matija: My Palit card also has the fan tab greyed out in RBE, even though fan control works in CCC etc.
-
- Posts: 353
- Joined: Sat Oct 18, 2008 6:30 am
- Location: Moldova, exUSSR
There are different 4670 firmwares, but all of them can be modified by the latest RBE. If you are not sure about the happy end, then make a backup copy of your stock firmware, prepare bootable DOS environment with latest version of Atiflash and the autoexec.bat, containing this: atiflash -p 0 firmware.rom -fs -fp. If anything goes wrong, this will flash the right firmware back.Matija wrote: BTW, my Sapphire doesn't play well with RBE, the whole fan tab is greyed out. Would it be safe to play with clocks and voltages? I only see a downclock to 300/1000 (from 750/1000).
Which brands step down voltage?
I just did an experiement, and found out my 3450 is drawing 20+ watts at idle, which is completely unacceptable for such a lousy card.
Anyone know if all versions of a brand will volt down? (HIS, and MSI specifically). I'm thinking about getting one of these:
http://www.newegg.com/Product/Product.a ... 6814161253
http://www.newegg.com/Product/Product.a ... 6814161262
http://www.newegg.com/Product/Product.a ... 6814127384
http://www.newegg.com/Product/Product.a ... 6814127383
http://www.newegg.com/Product/Product.a ... 6814161252
I know most are more powerful than I need, but I feel burned by the 3450's performance on full-screen things like Flash or AVC HD stuff, and don't want to risk the 4350/4550 route again, especially when the price difference is minimal, and the power draw might be even higher at idle on those cards - true? (cheap, less incentive to add the PowerPlay circuitry)
-Dan
Anyone know if all versions of a brand will volt down? (HIS, and MSI specifically). I'm thinking about getting one of these:
http://www.newegg.com/Product/Product.a ... 6814161253
http://www.newegg.com/Product/Product.a ... 6814161262
http://www.newegg.com/Product/Product.a ... 6814127384
http://www.newegg.com/Product/Product.a ... 6814127383
http://www.newegg.com/Product/Product.a ... 6814161252
I know most are more powerful than I need, but I feel burned by the 3450's performance on full-screen things like Flash or AVC HD stuff, and don't want to risk the 4350/4550 route again, especially when the price difference is minimal, and the power draw might be even higher at idle on those cards - true? (cheap, less incentive to add the PowerPlay circuitry)
-Dan
Re: Which brands step down voltage?
I was wondering one thing... is it possible to undervolt/underclock these cards in the same way as cpus?I know most are more powerful than I need, but I feel burned by the 3450's performance on full-screen things like Flash or AVC HD stuff, and don't want to risk the 4350/4550 route again, especially when the price difference is minimal, and the power draw might be even higher at idle on those cards - true? (cheap, less incentive to add the PowerPlay circuitry)