Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Prince134

macrumors 6502
Original poster
Aug 17, 2010
338
153
This is to update to you guys that two HD7970 are in my machine and without external power. And it's silent! You only spend a fraction of money comparing to buying a new MacPRO, provided that performance is your most important decision factor.

Meet the new King of Mac:)
Dual X5690 3.46GHZ, dual HD7970, 128GB DDR3 1333mhz (8X16gb), two 500gb SAMSUNG EVO at 1.0TB raid0.

The SAMSUNG SATA Raid 0:

https://forums.macrumors.com/attachment.php?attachmentid=473007&stc=1&d=1400545178

The Raid 0 speed Test:

https://forums.macrumors.com/attachment.php?attachmentid=473011&stc=1&d=1400545178

Cable used for SATA power connector: (note that you need to cut one of the edge for the connector.)

https://forums.macrumors.com/attachment.php?attachmentid=473009&stc=1&d=1400545178


OK. The way I have done is using VBE7 (google the keyword) to lower both cards' voltage from default 1.17V to 0.956V, downclock only tiny bit to 925MHZ for GPU and 1300MHZ for memory. By doing this, HD7970 will consume less than 200W ( around 175W-200W). Second card in my mac Pro uses the remaining 2 SATA ports.

Luxmark OPENCL test with both D700s (as it reported on my osx 10.9.1) with CPU+GPU, the default scene to be 5146.

https://forums.macrumors.com/attachment.php?attachmentid=473023&stc=1&d=1400547538


----Critical information about combination of the 6+8 power----Updated on May 19 2014--

It's critical and this is why I decide to post more instructions for you guys who wants to do it. This makes the theory profound and even lower risk for Mac Pro 5.1.

I used HDsensor to get the reading. It also solved a myth that both 8+6 pins never draw the same currents. Experiments done with the lower clocked/volts HD7970 under Unigene Valley default demo. It reads 27W from PCIE + 28W from 6Pin + 70W from 8pin in total of 125W. (That's lower than what had been calculated). The readings show a necessary that power connecting from SATA port (calculated 55W typical capacity from SATA port) to be arranged to the 6pin on card, while power from the Mac Pro onboard boaster (75W typical capacity) need to be only connected to 8 pin.

As shown here:


https://forums.macrumors.com/attachment.php?attachmentid=473010&stc=1&d=1400546893


Thanks for Asgorath's post in this thread in #14,
https://forums.macrumors.com/threads/1514609/


Enjoy your new KING OF MAC! :D

There are bios posted here:
https://forums.macrumors.com/threads/1733650/

Please don't be too serious about the maxed Xeon 3.46Ghz, these two D700 upgrade are far more important than the CPU itself. In other words, to be as competent as nMP, and for openCL which is strongly supported by APPLE in FCP and ADOBE's PREMIER PRO. Honestly, I tried single CPU x5670 with dual D700s, the BruceX test is the same 23-24 sec. Hope you guys can update yours if you have done it. Any questions for the instruction are welcomed.
 

Attachments

  • IMG_0936.JPG
    IMG_0936.JPG
    1.7 MB · Views: 7,830
  • IMG_0938.JPG
    IMG_0938.JPG
    1.7 MB · Views: 2,763
  • IMG_0940.JPG
    IMG_0940.JPG
    1.7 MB · Views: 5,971
  • Screen Shot 2014-05-19 at 2.58.15 PM.png
    Screen Shot 2014-05-19 at 2.58.15 PM.png
    1 MB · Views: 3,069
  • Screen Shot 2014-05-10 at 8.56.58 PM.png
    Screen Shot 2014-05-10 at 8.56.58 PM.png
    1.1 MB · Views: 4,571
Last edited:
Sounds Sweet!

Congratulations on getting it on-line and with the standard power supply! I will have to try that on my Mac Pro.

BTW, are you running OS X 10.9.1 or 10.9.2?

I have been having problems with 10.9.2.
 
Congratulations on getting it on-line and with the standard power supply! I will have to try that on my Mac Pro.

BTW, are you running OS X 10.9.1 or 10.9.2?

I have been having problems with 10.9.2.

I am using 10.9.1. I think 10.9.2 should be working too. It just won't report D700s but 7970 instead.
 
Careful, stating that the emperor has no clothes will get you called names around here.

And never forget, you may have made a faster computer for less money, but it still doesn't have that panacea for all things wrong with the modern world, TB. (Not tuberculosis, Thunderbolt)

Congrats on your accomplishment, please keep everyone updated on the long term.

Can you link to the adapters you used?
 
Thanks for sharing! I read about this VBE7 program. Looks as though it will kill our boot screens (not critical, I have a spare Radeon 2600) is the only drawback. Hmmm... may have to try this :D
 
Bring down the voltage is a good idea when the card is running stable.

BUT: 2 SATA ports are not enought to handle the electric current. One SATA port with all 3 circuits can handle 4,5 ampere (1,5 ampere per circuit). So you can burn out your motherboard lanes or damage the SATA connector over the time. It is better to constuct a 3x SATA to 2x PEG adapter.

(on 200 watt the card have a current of ~5,5 Ampere)

http://www.allpinouts.org/index.php/Serial_ATA_(SATA,_Serial_Advanced_Technology_Attachment)
 
Last edited:
Bring down the voltage is a good idea when the card is running stable.

BUT: 2 SATA ports are not enought to handle the electric current. One SATA port with all 3 circuits can handle 4,5 ampere (1,5 ampere per circuit). So you can burn out your motherboard lanes or damage the SATA connector over the time. It is better to constuct a 3x SATA to 2x PEG adapter.

(on 200 watt the card have a current of ~5,5 Ampere)

http://www.allpinouts.org/index.php/Serial_ATA_(SATA,_Serial_Advanced_Technology_Attachment)

You are right, 4.5A per Sata provides ~55W (4.5X12V), with two you get ~110W. PCIE provides 75W, so The second card is roughly using the rated power to 185W. Because by lower the voltage from 1.17 to 0.956, it's around 205W. Now I down clocked GPU from 1Ghz to 925Mhz (down 7.5%), and memory from 1425 to 1300Mhz (down 8.7%). So I say the system is safe at around 185W consumption at load.

Well, you know what, after boot and in 2D program, the card is using 0.85v at 300mhz. Just give you an idea that most of the time the card is at far lower power consumption. As I said, I have raid0 for my SSD, I got only two Sata ports. And I think it's safe given the calculations above.

BTW, I haven't tried 0.89V, maybe some one can even make the GPU and memory working at say another 10% lower frequency. Why do you think new Macpro consume only 450W? I think because APPLE must have done the same thing as this for their D700s in nMP. I only paid $150 for a XFX 7970, why not try it myself and make the Mac Pro even more worth it for a longer time?
 
Last edited:
Thanks for sharing! I read about this VBE7 program. Looks as though it will kill our boot screens (not critical, I have a spare Radeon 2600) is the only drawback. Hmmm... may have to try this :D

Oh yeah, I know that since the VBE7 will erase the EFI part. Don't worry you can rebuild your EFI bios to have boot screen AFTER you down clocked and downvolted.
 
Last edited:
Hello Prince134,

can you explain a little bit more on how you made your MacPro 5.1 "silent"?
 
Just putting two cards into a cMP does not give you the equivalent power of a nMP, does it? In the nMP the D700's work in Crossfire in bootcamp. Also, in OSX they work in tandem on the nMP for software optimized for it. Do you get the same benefits in the cMP or are you just able to drive more displays?
 
Careful, stating that the emperor has no clothes will get you called names around here.

And never forget, you may have made a faster computer for less money, but it still doesn't have that panacea for all things wrong with the modern world, TB. (Not tuberculosis, Thunderbolt)

Congrats on your accomplishment, please keep everyone updated on the long term.

Can you link to the adapters you used?
Thanks for the reminding.

Obviously TB is not something I am worried about since I am not using it. and they are expensive. The applications of TB, let's face it, display, storage, daisy chain, are all minor convenience. To me at least. Since this Mac Pro can outperform nMP, what TB benefit can I look for? Storage? Don't the NAS and USB3 is working good enough?

The machine is running excellent. I think openCL is strongly supported by APPLE, so I am glad the two D700s shows power here. They are same or even more powerful than new Macpro's D700s because 925Mhz GPU and 1300Mhz memory are still higher than those of nMP.

I am using regular cable (molex), to the end connected to Sataport on mac, the SATA power connector has to cut one side on the edge. It's no big deal. Every time before close the case, I will double check the plugs are firm.
 
Last edited:
Hello Prince134,

can you explain a little bit more on how you made your MacPro 5.1 "silent"?

I say it "silent" because originally I used one HD7970 flashed with boot screen. I am satisfied with the openCL score it gave. When I got my second HD7970, I thought VBE7 can help me put it into my Mac Pro without external power. Now not only both cards double OPENCL scores but also more silent after flashed bios with lower voltage from 1.17V to 0.956V, which is to my surprise.

After down volts, both cards combined (yes combined!) has lower noise level than with only one that with original clock and voltage. It's almost inaudible in my environment. I can hear it only when it got kicked with openCL compute load in a longer period, but still not loud at all. For example, under Final cut pro 10.1 BruceX test, both cards' fan stays at regular speed. Before that, BruceX test will result in my fan speed at louder level. Because the lower wattage consumption from 250W to ~180sW, Max Temperature at load from 82c down to 72C. It's like double benefits.
 
Last edited:
I am asking because my Mac Pro was a 12 core X5660 (95W TDP per CPU) and the X5690 is 130W per CPU. My system was nowhere near silent event at idle the Mac Pro created a noise level that was very distracting to me.

Running Prime95 on all cores for 10 Minutes or more already created a noise level that reminded me of a vacuum cleaner. That was with a low end gfx card. I tried a lower clocked / under-volted 7970 as well but even one card after 5 Minutes of FCP rending created so much noise and heat in the system. I decided to try two 7850 instead but finally gave up because the Mac Pro case and fans are not designed to disperse that much heat within an acceptable noise level for me.

I even then tried it one more time and replaced all Fans with Noctua Ultra Low Noise versions with an external Fan controller.

So i guess silent is a fairly wide definition then. As long as you are happy with - it it's all good.

At the end I sold mine for a silent Mac Pro ;-)
 
Last edited:
Great experiment!
I'm hoping this is a hobby project and not vital for your day to day income.
Thanks!
 
Great experiment!
I'm hoping this is a hobby project and not vital for your day to day income.
Thanks!

Whether you call it hobby or not depending on what you mean it.:eek: My day to day income is not selling something I shared with here. So definitely not vital to me.

But I own two Mac Pros, one 12 core as you guys have seen here as max performer, King of Mac as I call it. The other is 6 Core. Both are my work horses. So its vital to me. But they are working excellent. What can I say.:D

The nMP is a beauty, I call it Queen of Mac! Still untouchable for me.:p
 
Just putting two cards into a cMP does not give you the equivalent power of a nMP, does it? In the nMP the D700's work in Crossfire in bootcamp. Also, in OSX they work in tandem on the nMP for software optimized for it. Do you get the same benefits in the cMP or are you just able to drive more displays?

Sorry forgot to answer your questions here.

Yes, it does. My Final Cut pro 10.1.1 BruceX test spent me 23-24 seconds. I still have a GTX680 4GB with me. When I put that single card it gave me 1 minutes and 42 seconds under same test. It's day and night...:mad:

Dual HD7970 in 2009-12 Mac Pro can be cross fired in windows too.

From the other thread, as we understand in nMP, it automatically use second card as openCL compute, as it's proved here with the 23 seconds BruceX test in 2009-12 Mac Pro. (single HD7970 BruceX test is 45 second) So it reiterate by itself that nMP in Mac OS does not cross-fired but allocated computing for its two D700s. nMP fastest D700s model got 21 Seconds BruceX test. Mine is 23. The 2 second difference is from the faster DRAM.

So answering your question, the two D700s in 2009-12 Mac Pro are fully utilised in tandam as are in nMPs.

The queen of Mac, nMP has its own merits. TB.

I saw the other thread about using mac laptop link to high computing TB box with GTX 780i. Yes that's good idea for laptop which can utilise higher grade GPU outside for gaming. But HEY, for Mac Pro. You've got D700s, what do you need TB for? I narrowed it down to storage and display. And that become minor.
 
Last edited:
Sorry forgot to answer your questions here.

Yes, it does. My Final Cut pro 10.1.1 BruceX test spent me 23-24 seconds. I still have a GTX680 4GB with me. When I put that single card it gave me 1 minutes and 42 seconds under same test. It's day and night...:mad:

Dual HD7970 in 2009-12 Mac Pro can be cross fired in windows too.

From the other thread, as we understand in nMP, it automatically use second card as openCL compute, as it's proved here with the 23 seconds BruceX test in 2009-12 Mac Pro. (single HD7970 BruceX test is 45 second) So it reiterate by itself that nMP in Mac OS does not cross-fired but allocated computing for its two D700s. nMP fastest D700s model got 21 Seconds BruceX test. Mine is 23. The 2 second difference is from the faster DRAM.

So answering your question, the two D700s in 2009-12 Mac Pro are fully utilised in tandam as are in nMPs.

Thanks for answering, much appreciated. I was skeptical but now believe. Congrats on your setup!
 
Careful, stating that the emperor has no clothes will get you called names around here.

And never forget, you may have made a faster computer for less money, but it still doesn't have that panacea for all things wrong with the modern world, TB. (Not tuberculosis, Thunderbolt)

Congrats on your accomplishment, please keep everyone updated on the long term.

Can you link to the adapters you used?

You are comparing buying a second hand cMP to a new nMP. Not fair.

Cause a brand new cMP upgraded to dual 3.46 and 2x7970 (when available from apple minus the CPU and GPU upgrade) would not be cheaper than the nMP

I agree right now you can buy a used cMP and upgrade it to out perform a nMP for around 50% $. Though that us new v old. I'd prefer the cMP myself right now. Next nMP update, not so sure.....
 
Luxmark screen shot

do you have any pictures or screenshots?

Here you go. the Luxmark screen shot 5,146 for the Sala Scene!:)
This is GPU + CPU score. Roughly deduction of 1,000 points to be the 2 GPU scores.
 

Attachments

  • Screen Shot 2014-05-10 at 8.56.58 PM.png
    Screen Shot 2014-05-10 at 8.56.58 PM.png
    1.1 MB · Views: 743
You may enjoy your "D700s" as long as you will stay on 10.9.1 ;)
In 10.9.2 and above you'll lose D700 engine along with about 10-15% of performance. It still will be on a very good level, at least in OpenCL.
 
You may enjoy your "D700s" as long as you will stay on 10.9.1 ;)
In 10.9.2 and above you'll lose D700 engine along with about 10-15% of performance. It still will be on a very good level, at least in OpenCL.

It does? Did not know that. Hmm, also means a person can miss out on 10.9.3 and the HiDPI settings for 4K.
 
You may enjoy your "D700s" as long as you will stay on 10.9.1 ;)
In 10.9.2 and above you'll lose D700 engine along with about 10-15% of performance. It still will be on a very good level, at least in OpenCL.

No, it won't. There were not a D700 engine or non D700 engine at all. They are AMD OpenCL engine. I only got different name on Luxmark from D700 to "Tahiti XT prototype". The scores are within 1% variation. So It's totally safe to upgrade, because it's still "D700s". Just like nMP's D700s, they are actually still "HD 7970s".
 
Last edited:
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.