5770, 5570 OCCT 3.1.0 frequency & efficiency curves

[mkchiu]

Data dump of the last ~6 weeks of my HD 5000 testing. I'll soon post a brief data set of HTPC-oriented power in the MSI VGA forum for their 5770, 5570 & 785G as well as instructions on forcing the Hawk into continuous 0.950V_GPU operation.



MSI R5770 Hawk & MSI R5570-MD1G
RBE clock tables

video BIOS, stock values
​

MSI R5770 Hawk
efficiency in OCCT 3.1.0

overclock, underclock, overvolt, undervolt, system power (DCW), temperature, noise
​

MSI R5770 Hawk
power controller, feedback circuit, partial schematic

passive component values only
​

MSI R5770 Hawk
power controller, other

http://ht4u.net/forum/index.php/topic,51178.msg463940.html#msg463940
more data--esoteric, photo shows pertinent power controller circuitry & includes feedback circuit components
​

MSI R5570-MD1G
efficiency in OCCT 3.1.0

overclock, underclock, overvolt, undervolt, system power (DCW), temperature, noise
V_GPU = 1.175V (set) --> 1.191V (load w/ OCCT 3.1.0) killed the front-end MOSFET (OnSemi NTD4909N) which sources ~75W max to GPU only, VRAM has separate front-end
​

MSI R5570-MD1G
power controller, feedback circuit

location of feedback resistors
​

Test system

SS-300SFD

system power (DCW) derived from prior data of power supply efficiency
​

P180 Mini
MSI 785GTM-E45
Phenom II-C2 X2 550, as X4 B50

f_CPU V_CPU  V_NB
1600  0.9000 1.1000
2000  0.9625 1.1000
2500  1.0625 1.1000
2900  1.1750 1.1375

Scythe Big Shuriken
Switftech MCX159-CU w/optional fan
gskill DDR2-800 CL4 @ 4-4-4-12-2
RE2 500
random Asus DVD-RW
keyboard & mouse KVM powered
1600x1200x32@60
​

 

[JayliN]

awesome thread. Exactly the info I was looking for.

EDIT: For my XFX 5770 HD-577X-ZNFC Reference Design, I can only get 745mhz stable at 1v_gpu, 32degrees idle, 70 degrees load, 22degree ambient

 

[mkchiu]

brief data set of HTPC-oriented power:
UVD-power summary of MSI 5770 Hawk, 5570, & 785GTM-E45; Hawk min voltage
http://forum-en.msi.com/index.php?topic=137595.0

For my XFX 5770 HD-577X-ZNFC Reference Design, I can only get 745mhz stable at 1v_gpu, 32degrees idle, 70 degrees load, 22degree ambient

The ZNFC looks like one of the eggcooler designs which first showed up around 2009 DEC. The Hawk was a roughly 2010 FEB 5770-card-variant. Also, the newegg photos of the ZNFC seem to show a ST (L6788A) instead of the uP6204 power controller on the Hawk.

So it's likely either 1) that particular unit has an earlier week GPU or 2) there are power-quality / component-design related stability issues at a lower than designed-for voltage like 1.000V.

 

[JayliN]

Disregard my previous results. Didn't mount my Accelero S1 correctly (spacers were too thick).

at 1V. 21c ambient:

Idle: 29c
OCCT GPU Max Temp: 45c
Left 4 Dead 2 Max Temp: 40c
Max stable without errors in OCCT: 775mhz
Max stable without driver restart (grey screen) in game: 760mhz

And yes, this card uses the ST L6788A controller and the egg shaped cooler (which I have replaced with the Accelero S1). I'm not trying to replicate your results, I just wanted to give a comparison to the reference design.

MSI has a new 5750 out with military class components like the hawk which I think would also be a great HTPC hard. It would be nice to see how efficient that card is. Newegg currently has it for $117 shipped after coupon and mail in rebate!

 

[JayliN]

just wanted to add an interesting link at silentpcreview forums.

http://www.silentpcreview.com/forums/viewtopic.php?t=58166

I bet those guys would love your data too.

 

[mkchiu]

I didn't expect a replication of results, just pointing out the most likely factors for the slight differences, all things being equal with stock cards--which I assumed.



I agree the current price on the new MSI 5750 is good.

              R5770 Hawk R5750-MD1G R5670-PMD1G R5570-MD1G
List          179.99     139.99     114.99      84.99
Discount       -5         -15        -6          0
Subtotal      174.99     124.99     108.99      84.99
Tax            14.44      10.31       8.99       7.01
S&H             0          0          6.98       5.99
Rebate in       1          1          1          1
Rebate return -15        -20        -10          0
Total         175.43     116.30     125.96      88.99

I saw the R5750-MD1G when shopping for yet another card. Yet, based on my power & performance estimates for the 5670 & 5750, neither was outstanding for HTPC. If I buy another HD5000, it will probably be a 5570, 5770, or possibly 5850. 5570 won for lowest power DXVA playback with sharpen complex 2 and minimal flickering due to GDDR3. 5770 for light gaming. 5850 for multi-monitor gaming.

The 5670 has higher clocks on GPU & RAM and GDDR5 vs the 5570, and should still result in significantly higher HTPC (DXVA) power even after tuning, and it will very likely exhibit GDDR5 delta f_VRAM tearing. Also, a 785G or newer seems to work fine in a 1080p HTPC application.

If AMD's Llano actually ships say mid-2011 and is 480 stream processors, it will likely end up the near-ideal HTPC setup. Southern Islands will likely have HD4000 performance/power (not including 4770), so a Northern Islands GPU-based system paired with a second daily-use Llano system is on my wish list.



I don't follow the spr forums much now. My silentpcreview forum account was deleted when I discovered and started describing the design defect(s) on my first 5770. There are more details on the card and defect(s) in the ht4u link above.

silentplummet's use of furmark was unfortunate. After the fiasco with my first HD5000, I had to quickly find a systematic stability test besides the commonly available furmark/kombustor. OCCT 3.1.0 provided relatively consistent and instantly obvious errors to the user (me), drew about 5% less PCIe 6-pin power than furmark, but drew the same system power as furmark, and raised card temp as much as furmark.

 

[techjunkie]

thanks a lot. I have been looking for this kinda thing for quite some time. I think wizzard and tpu should adapt this kind of testing

 

[JayliN]

snip.

I'm impressed with your knowledge and all the testing you have done. You should put all this in a blog or something.

 

[mkchiu]

think wizzard and tpu should adapt this kind of testing

It takes too much time fer every card. However, special cases such as new PCB layouts, new power controllers, and binned parts like Power Color's 5750 Go Green would be very interesting as baselines for comparing the (HD5000) family and manufacturing advances over time.

I'm impressed with your knowledge and all the testing you have done. You should put all this in a blog or something.

I know very little--only enough to be dangerous. What I've collected was the minimum needed to get my "job" done. I'm too lazy to blog and have other projects and laundry, the never-ending laundry.

 

[mkchiu]

MSI R5570-MD1G #2
efficiency in OCCT 3.1.0

overclock, underclock, overvolt, undervolt, system power (DCW), temperature, noise
A second card to replace the first unit I killed, and a better performer than 5570#1. It's about 110 Mhz better than stock or

5570#2     => 17%  =(760-650)/650,   --5570 is 650 MHz @ 1.05V stock
vs
5570#1     => 10%  =(715-650)/650 &
5770hawk#1 =>  7.4%=(913-850)/850 or --5770 is 850 MHz @ 1.125V stock
5770hawk#1 =>  4.3%=(913-875)/875.   --Hawk is 875 MHz @ 1.125V stock

I'm almost tempted to pull the heatsinks to look at manufacturing datecodes.
​