Redefining Budget Gaming Graphics: ATI's HD 4670
Thanks, I get somewhat confused by the numbering of video cards. One advantage right now is that you can buy a passively cooled 3850 - is anyone selling passively cooled 4670s or do you have to add an aftermarket heat sink?Vicotnik wrote:4670 most of the time it seems, but the 3850 could be faster in some situations since it has more memory bandwidth.wayner wrote:Which card is faster - the HD 3850 or HD 4670?
I couldn't wait any longer for the passive one. So got the Asus one, The heatskink is large enough so might work passive, i'll will definately give it a try.
Maybe the HIS version would be even better as it has a much larger heatsink that extends all the way to the exit vent, then positive preasure might be enough to cool it?
Maybe the HIS version would be even better as it has a much larger heatsink that extends all the way to the exit vent, then positive preasure might be enough to cool it?
I think it's just that it can do HDMI with audio and encryption etc, as opposed to unencrypted DVI over HDMI.
Apart from the benefit of having digital audio on the same cable, in theory encryption could one day be important if manufacturers start using DRM to require it before a movie will play back. Personally I don't care about that because DRM = no sale.
Apart from the benefit of having digital audio on the same cable, in theory encryption could one day be important if manufacturers start using DRM to require it before a movie will play back. Personally I don't care about that because DRM = no sale.
The debate about that is somewhat less of a black/white issue (nvidia drivers were 30% of all vista crashes )Aris wrote:8800GT uses around the same amount of power, but gives you MUCH better performance, and only costs around $30 more. Only thing the 4670 has is a shorter PCB Form Factor. But as long as your case can support longer video cards, the 512mb 8800gt is better in every aspect. Also nvidia drivers are still superior to ati.
what is clear is dat the 8800gt has a much higher tdp (105w as stated by nvidia) and it's htpc related features are not as good at the 4670 (or nvidia's own newer products) no vc-1, not as efficient and as far as sound over hdmi is concerned nowhere near modern cards.
-
- Posts: 4
- Joined: Wed Nov 05, 2008 7:38 am
- Location: www
-
- Posts: 4
- Joined: Wed Nov 05, 2008 7:38 am
- Location: www
How can I tell that http://www.gainward.com/product/product ... cts_id=189Vicotnik wrote:If the Gainward card follows the reference design, then yes the S1 rev2 will fit.
IceQ 4670
I have an IceQ 4670. The fan is uncontrollable, but at least the noise level is quiet to begin with.
The fan control in CCC8.10 is disabled on my screen and I double checked with a multimeter, the voltage stays at 12v. The fan only uses a 2 pin connector.
The fan control in CCC8.10 is disabled on my screen and I double checked with a multimeter, the voltage stays at 12v. The fan only uses a 2 pin connector.
Just one note on the article. Since this is an ATI gfx with UVD2 I feel it would have been interesting to test the powerdraw of this card paired with say a tested 740G mobo to be able to compare the numbers to the 780G mobos playing videos to see if any of the solutions would have an edge in a HTPC (Given the feature sets, that wouldn't be a far fetched comparison).
The best we can do now is to compare relative draw, since the gfx is tested on an Intel platform and the efficiency of the cpu is a factor which is much less satisfactory.
The best we can do now is to compare relative draw, since the gfx is tested on an Intel platform and the efficiency of the cpu is a factor which is much less satisfactory.
-
- Posts: 1406
- Joined: Tue Feb 13, 2007 7:28 pm
- Location: USA
The problem is that there is a greater variation in the competing motherboards or HD4670 cards from different manufacturers than the theoretical power savings of a 740G vs a 780G. What I am saying is that chipset to chipset you are talking less than 5W, but it is easy to find two MB with the same chipset (either 740G or 780G) from competing manufacturers that have a > 5W variance due to VRM design or other considerations. At the same time, the best case for an HD4670 seems an idle draw of 3W, but there seem to be examples that idle at up to 8W. I think the simple answer is that 780G provides at least potentially lower power, but only by low single digit Watts. Not enough difference, IMO, if you could actually use the HD4670 features like 8 channel LPCM or better 3D acceleration.Tobias wrote:I feel it would have been interesting to test the powerdraw of this card paired with say a tested 740G mobo to be able to compare the numbers to the 780G mobos playing videos to see if any of the solutions would have an edge in a HTPC
-
- Posts: 10
- Joined: Sat Nov 29, 2008 11:31 pm
- Location: Finland
It seems that the only card capable of 3-4W idle power was the ATI reference card.
Samsung couple of years ago produced first batch of better quality monitors for reviewers and later they added more lcd factories into producing the 'same' monitor but with lower quality. One way of preventing this is rewrite the review and select the worst specimen from the shop. That would be the best way to force the companies to rise the quality.
Because I bought the 4670 after reading the SPCR review of idle power, and we haven't found one card that is commercially sold to be capable of that low power, I think it would be goog to add this information to end of the review. If some maker can actually achive that low power, it would be good advertising from them to give you a new card to review. Before that, 4W idle power is just a nice story.
Samsung couple of years ago produced first batch of better quality monitors for reviewers and later they added more lcd factories into producing the 'same' monitor but with lower quality. One way of preventing this is rewrite the review and select the worst specimen from the shop. That would be the best way to force the companies to rise the quality.
Because I bought the 4670 after reading the SPCR review of idle power, and we haven't found one card that is commercially sold to be capable of that low power, I think it would be goog to add this information to end of the review. If some maker can actually achive that low power, it would be good advertising from them to give you a new card to review. Before that, 4W idle power is just a nice story.
-
- Posts: 1
- Joined: Tue Aug 28, 2007 5:38 pm
power consumption measuring
thread necro. woot.
It's possible to measure absolute gfx card power consumption with decent accuracy if you have a second removable gfx card to test with. e.g. an old PCI card.
Measure power draw with GFX card under test + PCI VGA
Measure with just the GFX card under test
Measure with just the PCI VGA card.
Power draw of PCI card by itself = power with both cards - power with card under test.
Power draw of card under test = power with both cards - power with just the PCI card.
I've done this myself and gotten what look like sensible results for my old Althon64 that I use for games. e.g. +8W at the wall for an AGP GF6200a in VGA textmode (idle). +5-6W for a PCI MGA 2064. I don't seem to have measured the power delta for my current AGP 7600GT (which is an awesome card for an old A64). System specs: Enermax EG465P-VE of unknown efficiency. K8V (KT800 chipset) with a 2.2GHz S754 Athlon64 (Newcastle core). Total power draw ~90W (including drives and fans).
If you don't have a PCI VGA card, just boot your machine headless (no VGA). If your OS doesn't boot properly without a VGA card, then try GNU/Linux, or something else that doesn't suck. You can log in over the network to make sure it booted up ok, and run CPUburn to generate some load for power testing, for example.
Anyway, has nobody ever thought of this? Because every article I see says how they aren't able to measure absolute power draws for anything. I know this isn't perfectly accurate because it counts the extra power draw from the PCI bus being active in with the PCI card's total, and the extra mobo power from the PCIe lanes being on in with the PCIe card's total. If your secondary card was also PCIe, that might help a little bit. (Any PCIe card will work in a PCIe x1 slot, even if you have to physically saw off the back of the slot...)
It's possible to measure absolute gfx card power consumption with decent accuracy if you have a second removable gfx card to test with. e.g. an old PCI card.
Measure power draw with GFX card under test + PCI VGA
Measure with just the GFX card under test
Measure with just the PCI VGA card.
Power draw of PCI card by itself = power with both cards - power with card under test.
Power draw of card under test = power with both cards - power with just the PCI card.
I've done this myself and gotten what look like sensible results for my old Althon64 that I use for games. e.g. +8W at the wall for an AGP GF6200a in VGA textmode (idle). +5-6W for a PCI MGA 2064. I don't seem to have measured the power delta for my current AGP 7600GT (which is an awesome card for an old A64). System specs: Enermax EG465P-VE of unknown efficiency. K8V (KT800 chipset) with a 2.2GHz S754 Athlon64 (Newcastle core). Total power draw ~90W (including drives and fans).
If you don't have a PCI VGA card, just boot your machine headless (no VGA). If your OS doesn't boot properly without a VGA card, then try GNU/Linux, or something else that doesn't suck. You can log in over the network to make sure it booted up ok, and run CPUburn to generate some load for power testing, for example.
Anyway, has nobody ever thought of this? Because every article I see says how they aren't able to measure absolute power draws for anything. I know this isn't perfectly accurate because it counts the extra power draw from the PCI bus being active in with the PCI card's total, and the extra mobo power from the PCIe lanes being on in with the PCIe card's total. If your secondary card was also PCIe, that might help a little bit. (Any PCIe card will work in a PCIe x1 slot, even if you have to physically saw off the back of the slot...)
The Rage Pro PCI is an ideal card to use for power testing, as it uses <1W at idle so you can basically discount it from measurements.
Using Linux headless is a nice idea but many motherboards will not boot at all without a graphics card. Also, if your main OS is Windows, there could be differences in power draw at idle compared with Linux so it's best to measure everything in Windows.
Using Linux headless is a nice idea but many motherboards will not boot at all without a graphics card. Also, if your main OS is Windows, there could be differences in power draw at idle compared with Linux so it's best to measure everything in Windows.
Passive cooler for 4670
I'd be interested to know which passive coolers or near silent active coolers work with this card. I've bought an asus hd4670 and want to get rid of the noisy fan. If anyone has any tips???
Re: Passive cooler for 4670
The Arctic Cooling Accelero S2 is probably your best bet for a passive cooler. (As mentioned above the S1 can also be used but might be overkill). It's only about a tenner at Scan and similar places and is easily efficient enough to cool a 4670.Alex J wrote:I'd be interested to know which passive coolers or near silent active coolers work with this card. I've bought an asus hd4670 and want to get rid of the noisy fan. If anyone has any tips???
They don't list the 4670 as compatible on the product page on their website but I think that is because most manufacturers ship their own board design rather than using the reference design. You could check your Asus card against this diagram to see if it fits, but it should. If you print it at 100% size you can even make a scale template.
I have fitted an Arctic Cooling Accelero S2 to the Asus HD4670. Fits comfortably inside a P182 case and easy to install.Alex J wrote:
I'd be interested to know which passive coolers or near silent active coolers work with this card. I've bought an asus hd4670 and want to get rid of the noisy fan. If anyone has any tips???
My only tip is don't try to twist the heatsink off - there are some small components on the card that are easy to break off if the heatsink suddenly releases. Instead use a hairdryer to gently warm the heatsink (after you've taken the fan off) until it is fairly warm to touch, but not hot. Then lift the heatsink straight off using a rocking motion. Will come off very easily - if not then heat it a little more.
I measured the GPU temperature under load from CPUBurn and Furmark before and after. For my system it was 69 degC with the stock fan, 66 degC with the Accelero S2. The Accelero felt only mildly warm to touch at this temperature. At idle the temperatures were 33 degC with the fan and 37 deg C with the S2.
Works beautifully - cool temperatures and no noise!