[SOLVED] DDR3 and +5V

chung_chang

Distinguished
Dec 28, 2006
9
0
18,510
I am building a home lab. While spec'ing out the power supply on outervision, one thing caught my eye... 28.5 A on +5V! After playing with my config, most of that +5V is coming from the 128GB DDR3 memory. Is DDR3 really powered from the +5V rail, or am I reading that wrong? I was considering EVGA 1000 G+, but it only has 24 A on the +5V. Here is my config: PSU CALCULATOR PART LIST
 
Solution
Hmm, jonny is the guru when it comes to power topics.

I can't find anything definitive without really doing a deep dive. Some of the DDR4 controllerschips I am seeing take 4.5-18v as inputs, so it might depend on the motherboard designer which they choose.

The signals to store data will be coming from the CPU integrated memory controller, which should run on 12V.

Hadn't considered standby. Always assumed the CPU was trickling power to keep it alive. So that could very well be the reason they have the power requirements go up on their calculator.

I wouldn't worry about it too much, plenty of systems like yours were built at the time, and I don't recall seeing any reports of 5V failure.

Might take a look at some OEM solutions and...
Doubtful, memory VRM is typically powered by the 12V rails. Those calculators generally can't be trusted for accurate results. All based on generalities and often worst case.

For example, early DDR3 would have consumed a lot more power than DDR3 available towards the end of its lifecycle. Depends on the number of chips on each stick as well. So they will have assumed the worst.

Each part manufacturer should list power requirements for each component. Doing it manually will get you the best results, then arbitrarily add something like 20% to the totals and buy something in that range.

Besides, if you are expecting a 1000W worst case scenario load, you should probably look at something even bigger. Though that does depend a lot on what you plan to have the system doing.
 
  • Like
Reactions: chung_chang
Thank you Eximo. I was googling which PSU rail powered DDR3 and my results were inconclusive. If they are indeed powered from +12V, then I'd feel much better. Even EVGA's most powerful PSU has 24A on the +5V. I guess that PSU calculator I used is just wrong -- adding/removing DDR3 only affected +5V rail, not the +12V rail.
 
Actually... DDR4 (not sure about DDR3) is powered by the +5V. And switches to +5VSB when you "shut down" which is why RGB RAM stays "lit up".

Maybe the same is true about DDR3? Can't say I know.
That makes sense. Rats! According to that PSU Calculator, configuring either DDR3 or DDR4 affected only +5V rail, not +12V rail. So maybe that tool is correct after all? I was hoping to to get a more definitive answer here, but so far it's 50/50. Both you and Eximo seem like very seasoned members here so I am still confused.

I did more homework and looked up the ram specs for my Kingston kvr16r11d4k4/64. According to Kingston, max operating power is 6.681 W per module. I will have 8 modules so total is under 55 W. Per the PSU Calculator, adding all 8 DDR3 changes the +5V rail by about 20 A. Converting to watt equals 100 W (5V * 20A), so it estimated almost twice the needed power. I understand DC-DC power loss from +5V to 1.5V for DDR3. And as Eximo explained, these calculators may be padding the results. So if I take a rough average, maybe a PSU with 24 A on +5V rail is enough in practice? This is assuming DDR3 is powered solely from +5V rail.

If no one else chime in, I guess I could find out the hard way -- buy the PSU and stress test the memory, and see if it trips. Would this be my next best course of action?
 
Hmm, jonny is the guru when it comes to power topics.

I can't find anything definitive without really doing a deep dive. Some of the DDR4 controllerschips I am seeing take 4.5-18v as inputs, so it might depend on the motherboard designer which they choose.

The signals to store data will be coming from the CPU integrated memory controller, which should run on 12V.

Hadn't considered standby. Always assumed the CPU was trickling power to keep it alive. So that could very well be the reason they have the power requirements go up on their calculator.

I wouldn't worry about it too much, plenty of systems like yours were built at the time, and I don't recall seeing any reports of 5V failure.

Might take a look at some OEM solutions and see what 5V rails they had in the beefy machines, just for some peace of mind.
 
  • Like
Reactions: chung_chang
Solution
A quick check on the Lenovo D30, which I am passingly familiar with. Included a 1100W power supply with 22A on the 5V rail. Could be configured with 128GB of memory 4A Standby. They also list an 8A Standby mode, not sure when that comes into play, also then says DC OFF mode. Possibly USB charging when the system is in standby?
 
  • Like
Reactions: chung_chang
A quick check on the Lenovo D30, which I am passingly familiar with. Included a 1100W power supply with 22A on the 5V rail. Could be configured with 128GB of memory 4A Standby. They also list an 8A Standby mode, not sure when that comes into play, also then says DC OFF mode. Possibly USB charging when the system is in standby?

Thanks for looking into this Eximo. Really appreciate it! I thought jonnyguru sounded familiar. I used to read his PSU recommendations years back...must have been over decade ago now. Always appreciated his solid PSU reviews.

I understand my homelab hardware (Ivy Bridge, DDR3...etc) came from a time when many PSU's had strong +5V rails. Nowadays, most of it has shifted to the +12V rail for modern systems. But if that Lenovo D30 could handle 16 sticks (512 GB) of ECC ram with 22A +5V rail, then I am not so worried with my 8 sticks (128 GB). Instead of waiting on B-stock EVGA 1000w G+, I went ahead and ordered a new Seasonic Focus GX-1000 with 25A +5V rail today. Won't ship until mid Sept but it should be enough for what I need. Thanks again to you and jonnyguru.

EDIT: Oh, I'm not so worried about +5vsb as my homelab would be on 24/7.
 
Last edited: