ATI Radeon HD 4000 Series discussion

[STaRGaZeR]

If I load up ati catalyst control center while my pc is just displaying 2d graphics, the core and memory clock of my HD 3870 is 300/800...just ask any 3870 owner, it's a video card version of cool n quiet. There are third party bios editors, that were announced in the news section here. I've used it. You can change 3 different settings, since there are 3 power planes. One is idle/2d, where there's gpu/ram speed, and a voltage adjustment for the gpu, there's a middle range clock speed, can't remember it's numbers because I didn't touch it, and there's a full 3d clock speed, 777/1125?. The voltages were 1.1ish, 1.2ish and 1.37...i think. I changed the idle/2d to 0.975, considering dropping it to the lowest, 0.9v. I've seen a slight drop of 1-2W in idle, so I think it did make work, it's just that at idle, the GPU at these settings is probably already using a minimum, it's other components on the card that are drawing power. Which is why i suggesting maybe they're dropping ram clocks now too, doubtful, it could be possible that they're not just undervolting, but shutting down some of the stream processors?

300/800? It should be 300/1126

[Bo_Fox]

I guess I'll be replacing my 8800GTX with a dual-core R770 if it is anything as ATI has promised (true dual-core performance, not just Crossfire).

[deathman20]

I guess I'll be replacing my 8800GTX with a dual-core R770 if it is anything as ATI has promised (true dual-core performance, not just Crossfire).

Ooh if thats the case... that might be my sign to get into dual cards.

[ryboto]

300/800? It should be 300/1126

I think I may have changed it in the bios...but that's what I was seeing. Oddly, I think my CCC is messed up since the hotfix, I can't use overdrive. If I set clocks, and use either "test custom clocks", "apply", or even "OK", it just resets the overdrive clocks to the default.

[Cooper]

I guess I'll be replacing my 8800GTX with a dual-core R770 if it is anything as ATI has promised (true dual-core performance, not just Crossfire).

ATI promised anything or you just read it on the net?

[alexio]

If I load up ati catalyst control center while my pc is just displaying 2d graphics, the core and memory clock of my HD 3870 is 300/800...just ask any 3870 owner, it's a video card version of cool n quiet.

Ofcourse I know ATI cards have different performance levels since the R520. I never said that they don't. You seem to be completely ignoring my post about deviders and some problems that come with some of them.

There are third party bios editors, that were announced in the news section here. I've used it. You can change 3 different settings, since there are 3 power planes. One is idle/2d, where there's gpu/ram speed, and a voltage adjustment for the gpu, there's a middle range clock speed, can't remember it's numbers because I didn't touch it, and there's a full 3d clock speed, 777/1125?. The voltages were 1.1ish, 1.2ish and 1.37...i think. I changed the idle/2d to 0.975, considering dropping it to the lowest, 0.9v. I've seen a slight drop of 1-2W in idle, so I think it did make work, it's just that at idle, the GPU at these settings is probably already using a minimum, it's other components on the card that are drawing power.

Let's just say 3-4W as your reading my be off by a little. You used a kill-a-watt kind of thing? Well they are for measuring in 1W accuracy But let's just say 3-4W.

You lowered the vGPU by roughly 10%. This means the new vGPU has a relative value of 0.9 (90%) to the old 1.1ish vGPU. Power draw scales roughly exponentially with voltage in these chips. So: 0.9 x 0.9 = 0.81. This means 3-4W is +-19% of the stock idle power draw. 3-4/19 * 100 = 16-21W, this would be the stock power draw of the R670 in 2D. Now add 3-4W for power regulation on the board and you can imagine the memory doesn't use a lot of power. This is normal because GDDR4 was especially designed to save energy in idle mode, just look-up the tech behind it, it's interesting stuff.

Which is why i suggesting maybe they're dropping ram clocks now too, doubtful, it could be possible that they're not just undervolting, but shutting down some of the stream processors?

Take a look at the tech papers on GDDR4. I think that the ICs that are used on todays GDDR4 graphics card use about 0.3W/IC in idle (at full speed). There's not much to gain there

And please don't tell me anything about BIOS modding, I'm sick of HEX

[Frank M]

Maybe I am the only one to notice but the 4850 is just an underclocked 4870!! Maybe we will see another 2900pro / XT type situation!!! I sure hope so!

Y hullo thar, Captain Obvious!

The same with 3800 series: RV670 in -50 and -70; the difference is clockspeeds,
RAM type and amount. I think everyone saw it, because it was to be expected.

[deathman20]

Y hullo thar, Captain Obvious!

The same with 3800 series: RV670 in -50 and -70; the difference is clockspeeds,
RAM type and amount. I think everyone saw it, because it was to be expected.

Indeed... 3870 is a biinned RV670 chip and the 3850 is just one that can't make the cut.

[ryboto]

Ofcourse I know ATI cards have different performance levels since the R520. I never said that they don't. You seem to be completely ignoring my post about deviders and some problems that come with some of them.

Let's just say 3-4W as your reading my be off by a little. You used a kill-a-watt kind of thing? Well they are for measuring in 1W accuracy But let's just say 3-4W.

Take a look at the tech papers on GDDR4. I think that the ICs that are used on todays GDDR4 graphics card use about 0.3W/IC in idle (at full speed). There's not much to gain there

I wasn't trying to argue with you, I just don't understand why you're saying about "deviders", and I never said anything against your statements, i don't know what you mean by"performance levels" either...the R500 series didn't have dynamic clock throttling, at least my x1950pro didn't. Either way, the 3870 idles 10W lower than my x1950pro, so it's doing something differently.

[STaRGaZeR]

Guys, the full stock settings for a reference design HD3870 are:

2D --> 300/1126 1.241v
Low3D --> 300/1126 1.241v
High3D --> 777/1126 1.327v

And usually lowering 2D voltage in BIOS ends with higher 2D voltage, dunno why. Power consumption is more or less the same, even if you use 3D voltage with 2D clocks the chip sucks +-2W. RV670 power consumption scales mainly with frequency, not with voltage. You can test it if you want, looking at temperatures and wattage to confirm it.

[alexio]

RV670 power consumption scales mainly with frequency, not with voltage.

That's quit impossible I think (due to the laws of physics). It's more likely that what you are trying to set isn't what you get.

[Rattle]

if ATI can reclaim in the performance crown with AA I will grab one up for sure.
Until then I'm happy with my 8800gts 512.

[STaRGaZeR]

That's quit impossible I think (due to the laws of physics). It's more likely that what you are trying to set isn't what you get.

I couldn't believe it myself, but testing revealed the truth. With 2D clocks fixed and varying voltage up and down the difference in power consumption was negligible. Do you think a DMM and a hard voltage mod don't work well together?

Now maintaining 2D voltage I upped clocks and did the same measurements, the difference was higher. Now combine high voltage and high clocks, you will end with final 3D PC.

And the laws of physics say:

X voltage, low MHz: - W
X voltage, high MHz (till it's stable): + W

Y MHz, low voltage: - W
Y MHz, high voltage: + W

BUT the relationship between the two is different in each chip and design. I'm only saying my tests conduced to that conclusion, that frequency is more important than voltage in RV670 talking about PC (with the chip in idle). Of course both have an effect, but the difference varying voltage (in 2D clocks) is negligible. I can be wrong of course.

[grimREEFER]

btw, is there a new direct x version or anything? or does the change in the naming scheme solely represent performance

[LordEC911]

AMD/ATi talked about these plans for notebooks some time ago, and the
next revision of 780G-mobile is supposed to support this. Lately nv also
decided that this is a good idea, so they started this too, but theirs
doesn't yet work either.

Just to clarify, Nvidia doesn't have their version "working" yet either....

Edit- Hmmm ninja edit?

[Frank M]

Just to clarify, Nvidia doesn't have their version "working" yet either....

I don't think I've said otherwise

Ed: woot, 1k!

[Dragy2k]

you have to love progress

[Tienar]

As we were able to find out, since few days the RV770 GPUs was in mass production and related products (expected) HD4x00 series will be in one to two months to buy.

http://www.ati-forum.de/allgemein/ne...tion/#post1589

[Miwo]

That's quit impossible I think (due to the laws of physics). It's more likely that what you are trying to set isn't what you get.

Stargazer is correct when he says the RV670's power consumption scales more with Freqency, not voltage. I've observed the same exact behavior. My card is hardmodded at 1.5v, and I change clocks in Rivatuner to take out 2D/3D switching voltages/speeds.

I've got a kill-a-watt meter in front of me, here are some quick
measurements at idle....

HD3870 idle draw scaling @ 1.5v core/mem : system consumption

150/1000 : 125w
200/1000 : 128w
400/1000 : 141w
600/1000 : 152w
800/1000 : 165w
950/1000 : 174w
950/1226 : 177w

I've found that the power draw from using Powerplay 2D Voltages and clock (297 MHz) isn't even worth the effort, not to mention some games switching clockspeeds. You'll get less consumption, even with higher core voltage if you just scale the clock way down. I just leave the core speed at 150 at desktop, then use 950 when I play games. Best of both worlds

[Origin_Unknown]

http://www.ati-forum.de/allgemein/ne...tion/#post1589

one or two months? i cant wait any more for new stuff

[NH|Delph1]

http://www.ati-forum.de/allgemein/ne...tion/#post1589

They're correct, but it's mainly RV770. The rest are a bit behind and won't arrive until Q3.

//Andreas

[Monkeywoman]

you guys seen this? straight from chiphell, "WARNING: The following information is absolutely non-sense and represents the view of someone completely insane...

Core: AMD/ATI RV770
Die size: ~250mm square
Production: TSMC 55nm
Silicon Revision: A11
Shader core: 160! x 5D (800! SP)
TMU: 32
ROP: 16/32 (?)
MC: 256-bit External, Ring bus
Frequency: 825~875 (xx70), 700~775 (xx50)

Use your imagination good and keep your mouth shut."

real? fake? mix to keep the competition guessing?

source, http://bbs.chiphell.com/viewthread.p...a=page%3D1

[[XC] gomeler]

Indeed... 3870 is a biinned RV670 chip and the 3850 is just one that can't make the cut.

My HD3850's aren't happy thinking they didn't make the A Team They crush most HD3870's, then again they are fed 1.65v The HD3850 I'd say is a more fun card to overclock as the "gains" are larger from stock to OC vs HD3870. One can only hope the HD4850 is just like this

[NH|Delph1]

you guys seen this? straight from chiphell, "WARNING: The following information is absolutely non-sense and represents the view of someone completely insane...

Core: AMD/ATI RV770
Die size: ~250mm square
Production: TSMC 55nm
Silicon Revision: A11
Shader core: 160! x 5D (800! SP)
TMU: 32
ROP: 16/32 (?)
MC: 256-bit External, Ring bus
Frequency: 825~875 (xx70), 700~775 (xx50)

Use your imagination good and keep your mouth shut."

real? fake? mix to keep the competition guessing?

source, http://bbs.chiphell.com/viewthread.p...extra=page%3D1

(Not shooting the messenger)

Feels like someone misinterpreted the source and thought they would increase the number of SP by 96*5 (480) instead of to 96*5 (480).

//Andreas

[jimmyz]

My HD3850's aren't happy thinking they didn't make the A Team They crush most HD3870's, then again they are fed 1.65v The HD3850 I'd say is a more fun card to overclock as the "gains" are larger from stock to OC vs HD3870. One can only hope the HD4850 is just like this

Yeah I laughed myself when I read that.... Mine have done 1ghz each @ 1.6 volts on water, haven't seen any better numbers from the 3870's I've played with.