1kW PSU too weak for GTX 770?

[ianhuysman]

So about a week ago I put a GTX 770 in an older Dell computer, everything works fine but when I play some more demanding games (fallout 4, skyrim, gta 5,...) and everytime something happens in the game my system crashes. I tried reinstalling my video drivers (using DDU) but it has no effect. After doing some research I found out a GTX 770 requires 42amps. When I look at my PSU it says it can only deliver 18A on the +12V rail. I don't know if I'm reading it wrong because I don't know how these rails work in PSU's. I've put a picture of the label of my PSU below.

 

[jonnyguru]

Anything that's tripped at the supervisor IC is going to latch the PSU. Not cause a reboot.

 

[Karadjgne]

Look at the pinouts for the p5, p18 and p24 and compare them to pinouts for pcie 6 & 8 pin. In the p24, 8pin, 12v is 1,2,3 across the bottom, sensor 4,6 and ground 5,7,8. On p5 the 12v is 4,5,6 across the top, grounds 1,2,3 across the bottom. That's wrong pinouts. P18 is 12v 1,2,3 and ground is 4,5,6. OP is using p5 and p24, should be using p18 and p24. Looks like he has a 6pin eps being used as a 6pin pcie.

 

[william p]

PSU swap isn't going to be an option. The T7400 has power hungry FBDIMMS and needs a special MB connector to run that and 2 CPUs. It has a standard 24pin, and a 2nd 20 pin with all 12V, and ground wires. Each rail will have it's own color code. White, yellow, and Blue are 12V. Then you will find striped combinations of those colors. The 18A rating of the 12V. rails is misleading. I have 3 different Dell PSUs sitting here. All 3 have 3x 12V/18A rails. the 375W PSU says 30A max.on all 3 rails (360W), the 425W PSU says 401W max. on all 3 rails, and the 525W says 500W max. on all 3 rails. I've also seen single rail PSUs with the samoe output rating on the rails, one was 305W the other being a more efficient one was rated 255W. The lower rating is a selling point for Dells business customers who are buying 1000s of computers t once. But if Dell provided a 6 pin and 8 pin PCIe it should have enough power. Interestingly the 425W PSU has 2x 6pin PCIE, while the 525W has only one.

 

[turkey3_scratch]

Look at the pinouts for the p5, p18 and p24 and compare them to pinouts for pcie 6 & 8 pin. In the p24, 8pin, 12v is 1,2,3 across the bottom, sensor 4,6 and ground 5,7,8. On p5 the 12v is 4,5,6 across the top, grounds 1,2,3 across the bottom. That's wrong pinouts. P18 is 12v 1,2,3 and ground is 4,5,6. OP is using p5 and p24, should be using p18 and p24. Looks like he has a 6pin eps being used as a 6pin pcie.

If the pinouts were improper the computer would not boot, short-circuit protection would shut it down immediate. His problem has to be something else because he can already get into Windows, it's just during normal operation of the computer that it begins to cycle.

 

[william p]

You might look for swollen or leaking capcitors on the MB. There was a lot of that going on back then.

 

[jonnyguru]

If the pinouts were improper the computer would not boot, short-circuit protection would shut it down immediate. His problem has to be something else because he can already get into Windows, it's just during normal operation of the computer that it begins to cycle.

This. Seriously. You can't even power up a PC with incorrect pinout... never mind have it boot up and play games.

 

[william p]

You might try running Malwarebytes, and MBAR their anti rootkit scanner (it's very good) and whatever AV program you have.

 

[Karadjgne]

And bumble bees shouldn't be able to fly. So either the manual is printed wrong and the pinout for that p5 cable is printed upside down or there's a different issue. Could be that Op got his p5 and p18 mixed up. I don't know. But starting with the pins from all motherboard /gpu connectors and looking for pins pushed out of socket etc wouldn't be a terrible idea.
I've seen ppl with broken keyboard cables trip psu protections and cause similar issues.
That's a heavy duty server psu. For all I know, it's entirely possible that it truly does have 5 individual rails that do not join back to a single source and p5 happens to be on 1 rail that's not used by anything else op currently has connected and actually runs with that rail tripped, until it trips some other rail like what's powering the mains. It's an odd duck psu that I've not ever seen before having 2x 20 pin mains, 1 for basically the 12v rails and the other for the minor rails. I'm not a math genius, but I can tell when things do not add up right, so there is info somewhere that's either wrong or missing

 

[ianhuysman]

Look at the pinouts for the p5, p18 and p24 and compare them to pinouts for pcie 6 & 8 pin. In the p24, 8pin, 12v is 1,2,3 across the bottom, sensor 4,6 and ground 5,7,8. On p5 the 12v is 4,5,6 across the top, grounds 1,2,3 across the bottom. That's wrong pinouts. P18 is 12v 1,2,3 and ground is 4,5,6. OP is using p5 and p24, should be using p18 and p24. Looks like he has a 6pin eps being used as a 6pin pcie.

I double checked it before I installed my gpu, also my GPU shows a red light near the connector when there is something wrong with the connectors, they're both green so there shouldn't be any problems with that.

 

[ianhuysman]

I might make a sperate partition on my HDD and isntall a fresh copy of Windows 10 on it, if it doesn't crash anymore I know it will be software related

 

[jonnyguru]

And bumble bees shouldn't be able to fly.

But yet we know they do.

For all I know, it's entirely possible that it truly does have 5 individual rails that do not join back to a single source

Possible, but not probable.

Just like any ATX PSU with multiple +12V rails, +12V rails are not additive. And this PSU makes note of that by stating that the +12VA... +12VE shall not exceed (off screen). That tells you it's a single +12V rail that has the +12V split up to separate rails with an 18A OCP on each.

Your theory would make sense if there was a +12VSB, which a lot of modern servers actually have. But this label doesn't call out a +12VSB.

 

[photonboy]

The problem is I would have to buy at least an 800W power supply and it might nog even fit because it's Dell. Also I would need a PSU with a 20pin cpu power connector. So the only way I can test it is with the exact same PSU, wich would be a waste of money. But I guess I'll first start a new thread about my system crashing before I buy any new parts.

1. You can TEST if it's software easily if you have a spare hard drive. Just make a copy of Windows 10 (need 8GB+ USB stick). Follow instructions here: https://www.microsoft.com/en-ca/software-download/windows10
a) insert USB stick
b) download media creation tool and run it
c) wait for the download + USB format to finish
d) *do in-place UPGRADE now which MAY fix any software issues (just run "setup.exe" from the USB stick and follow instructions.. choose to keep apps/data if prompted though should be GREYED OUT. don't enter key if prompted just skip)

I've done that many times safely. If you want to SKIP that or it doesn't work then CONTINUE

e) add spare HDD/SSD (at least 60GB)
f) boot to USB stick (may need to change BIOS order) and follow W10 install instructions. Ignore key
g) install latest NVidia driver, and Steam + game or whatever is easiest to confirm it fails (Recommend testing with Unigine VALLEY to see if that crashes as that's the easiest test to repeat.)

2. 800W?
I'm sure you know by now, but I explained fairly clearly above how much power you need.

3. Not software (i.e. #1 Windows upgrade or clean install)?
Must SWAP either the graphics card (to one that needs same or MORE power) or the power supply.

SUMMARY:
That's your best bet for troubleshooting. If you start ANOTHER THREAD and concentrate on software your head will probably spin.

I can give other advice like run DDU to remove drivers, then reinstall video driver but I doubt that's the issue. When a PC crashes during any demanding game it tells me it's likely the PSU or graphics card.

 

[Magnus Thunderson]

one advantage to dell is there mostly white boxes i.e non proprietary the disadvantage is crappy parts so yes an upgrade to your PSU should be able to be done but one thing about power supply watt do matter but build quality matters more as some quality 600w have better performance the some cheep crappie 1000w

 

[Karadjgne]

I believe that's a Chicony OEM Dell psu, which means it's pretty good quality. Or has a fighting chance of being good quality at a minimum. Dell's biggest disadvantage is user upgrades, as stated, Dell is highly proprietary, you'll be hard pressed to find another 1kw psu with those connections. Other than what Amazon sells new for $120. Which happens to be the same psu. If it is Chicony, it's more than likely a better quality psu than a good portion of what's on the market right now. Delta is also OEM for some of Dell's psu's and those are better quality than a really large chunk of the market. Dell might be a proprietary Pita, but they do have very respectable pc's, monitors etc. I had a cheapo Dell optical mouse that simply would not die. It went from me, to my 2 daughters to my wannabe gamer son and back to me again. I wore holes in the bottom after the slides wore off and it still worked fine. I still have it, it's my backup mouse. Best $4 I ever spent on pc stuff.

Yeah, Dell is decent.

Oh, and there really are not any decent 600w psus, 550w yes, 650w yes, 600w? Not for what op needs, not even close.

 

[avenge]

Have you checked for bad capacitors Motherboard AND PSU ? You have to disassemble the PSU and probably have to clean it also, then take a good look at the capacitors.

https://www.google.bg/search?q=bad+capacitor&client=firefox-b-ab&dcr=0&source=lnms&tbm=isch&sa=X&ved=0ahUKEwiuuoic343WAhUGPRQKHf29B-0Q_AUICigB&biw=1321&bih=903

 

[ianhuysman]

So I installed a fresh copy of windows 10 on a second partition on my SSD, after installing the most recent drivers. And after playing some games it still crashed. But I guess I'll just leave it as it is because at the end of this year I'm planning to 'upgrade' to a 1600X system wich should be a lot better than the 10 year old xeons I currently have and it should be a lot more stable. I will try some stuff you all sugested, so I'll keep this post updated.

 

[photonboy]

So I installed a fresh copy of windows 10 on a second partition on my SSD, after installing the most recent drivers. And after playing some games it still crashed. But I guess I'll just leave it as it is because at the end of this year I'm planning to 'upgrade' to a 1600X system wich should be a lot better than the 10 year old xeons I currently have and it should be a lot more stable. I will try some stuff you all sugested, so I'll keep this post updated.

If you plan to upgrade, I suggest investing in a quality 650W EVGA G2 with ECO MODE (or similar). Even if it's not the problem you would then have a backup PSU (now the DELL).

Fairly certain it's the PSU or the graphics card. Only way really is to SWAP ONE.

R5-1600x is a good system. If you overclock which you should the R5-1600 will give the same performance AFAIK. It has a stock cooler but you may want better for the overclock and/or lower noise.

*RYZEN needs good memory bandwidth. I recommend a 2x8GB DDR4 3200MHz kit from G.Skill or Corsair, preferably an exact model known to work at full frequency (or at least 2933MHz) for the exact model MOTHERBOARD you choose.

GTX770 is about the same as a GTX1050Ti for most games so I wouldn't upgrade to anything less than a GTX1070.

Example build:

R5-1600
Cryorig AM4 cooler?
motherboard
250GB Samsung EVO SSD
2TB HDD
2x8GB DDR4 3200MHz memory

W10 Home or Pro 64-bit
EVGA G2 650W
case?
GTX1080 (Asus Strix or similar?)

 

[Karadjgne]

Etherium miners are goin hognuts over the gtx1070 atm, so prices are rediculous, you can get gtx1080's for lower prices than many models of the 1070.

Ryzen ram best suited is Patriot elite, g-skill Trident-Z or RipJaws V CL14 or FlareX, all of which use the Samsung B-die and there are many reports of ram in excess of 3600-4000MHz being attained. The Corsair LPX, g-skill RipJaws V CL16, RipJaws IV and several others are using SkHynix ic's which to date are having issues over 2666MHz, with some kits reaching 2933MHz. This isn't bad ram, or a failure on the ram or cpu, just a compatability issue that's slowly being resolved via bios updates and microcode from both vendors and AMD.