Throttlestop overclocking Desktop PCs

[Retrorockit]

Now a 4GHz QX6800 without SSE4 has some real limitations!

 

[Goldbricker]

Hi, Retrorockit let's move the discussion over here. My Dell E520 system DOES have a QX6800:

So, as a safe first step I could set the Voltage higher say 1.45 and multiplier 13 that seems sensible?

One thing that confuses is that in TS I set the voltage to 1.45 and Multiplier to 13 but I don't see any increase in core speed. Am I missing a step? Previously I found that rebooting applied the change but this seems a bit risky.

 

[lexluthermiester]

So, as a safe first step I could set the Voltage higher say 1.45

1.4v would be a wiser step, even for the 65nm process Kentsfield's.

BTW, welcome to TPU!

 

[unclewebb]

One thing that confuses is that in TS

Post a screenshot of ThrottleStop so I can see your settings. After setting your voltage and multiplier did you press the Turn On button?

 

[Retrorockit]

Glad to see you have The G0 Version of the QX6800. Those MB don't make a lot of Voltage stock. If you put in a setting it can't produce you just get a lower Voltage. Heatsinking the VRM Mosfets is worth about .05V. Thin copper flags glued to the vertically mounted MOSFETs works. You have to be careful of the ones near the cooling shroud that they aren't too high, and that air can get to the ones behind the mounting post for the cooler.
In a couple places Enzotech MOS1 copper sinks which can be bent like copper wires work well.
The real increase in Voltage comes when you power the Fan off of a Molex. This is worth .1V. It allows your voltage to start .05V. lower, and go .05V higher. The starting .05V. lower for the same clock speed is worth doing and helps with Lex's concern of high Voltages. A small fan on the NB wired to the +,- terminals of the fan header will avoid a fan fail BIOS error that must be cleared manually. Don't use superglue, it off gases when heated and will ruin the optical drives. I use the 2 part Arctic epoxy that's made for this. Use the blue and yellow leads from the MB fan header for RPM, and PWM to control the fan. If it crashes when the fan speeds up then you need this.
Lex is a more by the book workstation user than I am. When these were $1000 chips caution was well advised. My approach is to keep the temperature within reason ,and have some fun with them. You should be able to run games and GPU benchmarks at 3.73Ghz, and CPUZ runs at 4Ghz like the one in my sig. The AFC 1512DG fan is cheap and just about a drop in for those. It's 150x50mm and 1.8A. 260CFM The Dell BTX MB header can handle it. But Molex power will be needed to get to 3.73GHz.
For cooling I always lap my heatsink and CPU. Worth about 3*C.
I'm assuming you have the D9729 heat pipe cooler that fits those. It's an old school tight fin cooler and might benefit from a Nidec M35105-57 fan stuck to the back of it. This is an old Dell fan that can be found cheap. It just needs 12V. to run. It has a thermistor control built in.
The T9303 cooler has different heatpipe layout but fits otherwise and has a wider fin spacing and cools better. Probably doesn't need the Nidec.
Bad capacitors were common on those systems, and that's what took mine down. So look them over before you invest too much effort into this.
The big fan powered off of Molex mod will probably get you to 3.73Ghz. I had done the MOSFET coolers first so can't be sure. BTX MOSFET cooling is very good to start with.
Back in 2015 when I was doing that I ran the GFB1212VHG 2 motor fan which is not as good as the AFC1512 mod.
I have a recapped MB, and some more extreme cooling mods to try on an E520 to push it a little further. But I've been busy with other things lately. When you get this up to speed the guys at the TPU Nostalgic Hardware Club would probably get a kick out of it.
I had an aftermarket single rail PSU in mine EVGA Supernova B2 which has enough 5V. power for those older computers. I mention this because at 3.73GHZ you might be drawing too much power for the multi rail Dell PSU. It has a 215W limit on each rail.

 

[Goldbricker]

Post a screenshot of ThrottleStop so I can see your settings. After setting your voltage and multiplier did you press the Turn On button?

I tried running UNLCK then dropping the Voltage from 14.5 -> 14 then pressing "Turn on". I should see the VID change in the top left and also on CPU-Z I presume.

 

[unclewebb]

@Goldbricker - You have SpeedStep disabled. The multiplier and VID voltage requests only get sent to the CPU when SpeedStep is enabled. On older CPUs, it is a good idea to enable SpeedStep in the BIOS. Some Dell computers can get locked to the 6 multiplier when SpeedStep is disabled in the BIOS.

 

[Retrorockit]

I do recall having to reboot to get my OC recognized by the benchmarks. I just used sleep mode and lived with it. Unclewebb may have a more technically correct solution. I also ran Prime 95x3 "workers" which produced about an 80-85% CPU load that I thought was realistic.
Yours has a full 266FSB, mine only did 265, and got pushed down the charts by other E520s with the same settings due to that.
This was probably one of the hardest Dells to OC. No pinmods to 333fsb, Set FSB did nothing at the time, and there was no resource to resolve the issues I found along the way. It also turned out to be one of the most rewarding. 3.73Ghz was good for 2nd place at CPUz in 2015, But I raised that to 4GHz just because that was the next TS setting. It held up for a couple years. But others have gone 3.73 on these since then. Not bad for something that originally had Pentium4 and PentiumD CPUs.
It also help to have a high performance user profile set in Windows.

 

[lexluthermiester]

Some Dell computers can get locked to the 6 multiplier when SpeedStep is disabled in the BIOS.

That generally only applies to laptops though.

 

[Retrorockit]

There are 2 Dell part# for the Delta AFC1512DG fan. NC466 which usually comes in a plastic housing that is useful for installation in a T3500, and DG168 which usually comes with a chrome grille which is nice for the T3400 and other BTX computers. It takes some work to get it into the mid tower BTXs, but drops right into the Micro BTX Optiplexes and E520 computers.

 

[OEMOC]

I just wanted to thank you all for the help you gave me months ago when I was looking into the t3500, and it's overclocking abilities. Between nabbing the t3500 for almost free, and finding a w3680 being thrown out for free I now have a budget gaming rig to let a great friend of mine borrow.

Specs:
w3680 @3.7x ghz w/ 150w tdp set through throttlestop
3x 4gb @1333mhz
r9 380x (using a 6 to 8pin adapter)
Stock dell 5xxW PSU

I have a 92mm fan on the CPU heatsink which keeps the CPU in the low 60s under stress testing. Between having no heatsink on the VRM, no additional NB cooling, using a 6 to 8 pin power adapter, and running the stock 5xxw Dell PSU I'm not looking to push the CPU OC any more than 3.7x ghz.

This ought to be a night and day upgrade for the guy as he's coming from an ivy bridge gaming laptop that requires CS:GO, Witcher 3, and Read Dead Redemption 2 to be run at sub 1080p at lowest settings:
i7-3xxxQM (4c8T)
8gb 1600mhz ddr3
860m 2gb (equivalent-ish to 750ti)

 

[lexluthermiester]

r9 380x (using a 6 to 8pin adapter)
Stock dell 5xxW PSU

That'll be ok. Fairly rock solid actually.

This ought to be a night and day upgrade for the guy as he's coming from an ivy bridge gaming laptop that requires CS:GO, Witcher 3, and Read Dead Redemption 2 to be run at sub 1080p at lowest settings:
i7-3xxxQM (4c8T)
8gb 1600mhz ddr3
860m 2gb (equivalent-ish to 750ti)

Yup, that'll be one hell of an upgrade!

 

[Retrorockit]

Nice to know the 190W GPU is running on the Dell PSU.
When you can use the Dell PSU it let's you put the money into a GPU or RAM upgrade. Their overload protection is first rate, so no reason to "upgrade" that part.

 

[OEMOC]

I had thoughts of running a 6 to 8 pin adapter AND a molex to 8 pin so I could throw a 290x (275w TDP) in there, but after doing some plugging that into a PSU calculator I decided it wasn't worth the risk.

If he ends up putting in parts that can be transfered into his own rig later (2.5in SSD, PSU, GPU) I'll definitely be keeping the stock dell PSU as those things are rock solid.

 

[Retrorockit]

That Dell PSU has 3x 216W rails (12v X18a=216W). So potentially about 650W@ 12V. available.
The 12v. wiring will be white/yellow/ blue color codes for each rail. If you get the entire blue rail powering the GPU, and 75W from another rail through the MB header there should be enough power. Moving the HDD/SSD into the floppy bays usually helps with this.
The UL rated PSU is very unlikely to catch fire, or damage components if it's overloaded. It should simply shut down safely.
But looking at the GPU hierarchy chart for legacy GPUs the 380X and 290X are not much different in performance. The 290X is a badge engineered HD series card, and the 380X is newer tech.
It might be interesting to see if the PSU can handle it, and then look for a newer card of similar power later on.
2 things I don't hesitate to add to old computers are PSU, and GPU upgrades because they can move forward to newer system when the time comes.
On Dells you have to look at the 5V. rail on the original PSU and make sure there is enough power there on the new PSU. EVGA is pretty good about this. But budget can be a very real consideration so what you've already done is quite respectable.

 

[OEMOC]

So my take away is to plug in the Vega 64 I have laying around into the T3500, remove the Vega's power limits and let it suck down all the watts?

Just kidding. I also went with the 380x because it doesn't give off as much heat as the 290x. He doesn't keep his room as cold as mine and I wanted to be sure everything would be ok when the days warm up.

 

[Retrorockit]

The Vega 64 at it's 3 lower power settings 206W, 220W, and 273W falls between the GTX 1070, and 1080 in performance.

AMD Radeon RX Vega 64 8 GB Review

Our AMD Radeon RX Vega 64 review confirms that the company achieved major performance improvements over their last-generation Polaris and Fiji cards: Vega is faster than the GTX 1080. We tested six different performance configurations of the Vega 64, with surprising results.

www.techpowerup.com

It seems like exactly the right card for that system.
You might not want to give it away, but it would provide some interesting benchmarks.
If I were running an 8GB GPU I would bump up the system RAM to 18GB or so.

Dell Precision WorkStation T3500 Performance Results - UserBenchmark

 

[frankr2994]

Hey all. looking for help. system is a dell precision T5810 with a intel 1650v3 xeon 32gb ddr4 ecc ram. Got it overclocked to 4.5Ghz without much issue. Trying to go to 4.6 triggers some throttles. Sadly I just don't know what they mean to know what I can adjust. Here is my screen shots for trying to go to 4.6. I'm also pretty sure that I can't use 4.6 right now. the stock heatsink does ok at 4.5 but i'll need better for 4.6. but i'll need the knowlege none the less

 

[totalfreq]

The Vega 64 at it's 3 lower power settings 206W, 220W, and 273W falls between the GTX 1070, and 1080 in performance.

AMD Radeon RX Vega 64 8 GB Review

Our AMD Radeon RX Vega 64 review confirms that the company achieved major performance improvements over their last-generation Polaris and Fiji cards: Vega is faster than the GTX 1080. We tested six different performance configurations of the Vega 64, with surprising results.

www.techpowerup.com

It seems like exactly the right card for that system.
You might not want to give it away, but it would provide some interesting benchmarks.
If I were running an 8GB GPU I would bump up the system RAM to 18GB or so.

Dell Precision WorkStation T3500 Performance Results - UserBenchmark

I would imagine a w3680 will bottleneck a vega64.

I was running a 1070ti on a T7500 with dual X5697 and 48gb ram and getting a 30% penalty. A gtx 1060 runs full out but the 1070ti was severely handicapped.

 

[frankr2994]

Ok just an update. I've tried increasing turbo limits, disabling turbo limits, using powercut and increasing pp0 power limit. no luck. however lowering my voltage 100mv made it take longer before throwing those same throttle flags. however at that lower voltage that frequency is not stable and I got BSOD. soooo ya. I did increase the voltage 100mv as well and it instantly and more severely throttles the cpu.

 

[SamirD]

Ok just an update. I've tried increasing turbo limits, disabling turbo limits, using powercut and increasing pp0 power limit. no luck. however lowering my voltage 100mv made it take longer before throwing those same throttle flags. however at that lower voltage that frequency is not stable and I got BSOD. soooo ya. I did increase the voltage 100mv as well and it instantly and more severely throttles the cpu.

I'd think that keeping it cooler would help with your goal, so if the cpu and case fans aren't already at 100%, I'd manually set them to 100% and see if that helps. Aside from that, you may need to do a cooler upgrade.

 

[frankr2994]

Ok I'll turn the jet engine up to 100 lol. And no I haven't set it that high. I'm assuming the fan is probably 75% when trying for 4.6 before it throttles. I would still like to know what power budge, edp current, and max turbo limits are. I really can't assume much because I think I tried modifying enough settings that I should have seen one of those flags go away.

 

[totalfreq]

Ok I'll turn the jet engine up to 100 lol. And no I haven't set it that high. I'm assuming the fan is probably 75% when trying for 4.6 before it throttles. I would still like to know what power budge, edp current, and max turbo limits are. I really can't assume much because I think I tried modifying enough settings that I should have seen one of those flags go away.

TCase is 66.7C on that chip, looks like you are hitting a max core temp of 77C. Depending on your cooling solutions, thermal paste etc. You can be anywhere from 10-20c different in your core vs TCase max temps but you're already showing at 10C difference which means you are likely already close to that limit. I'd agree heat is likely causing the instability and either improving the transfer material (repaste or change the cooling system) or turn up the cooling as high as you can and try and get that core temp down if you want to clock higher.

 

[frankr2994]

The last one I overclocked which was the same system specs t5810 with a 1650v3 I put that chip in there and used artic silver. Temps ran hotter on that chip and I was only able to get 4.3 out of it. The heat pipes on the bottom of the Dell cooler have space beside them. It's not totally flush and it requires a ton of paste to make a good connection. The paste that Dell uses seems to be much thicker in viscosity and I thought it seemed to work better on their poorly designed heatsink. So if it really is just temps causing it I should be able to max the fans out to test. If it works I'll be buying another better cooler.

 

[totalfreq]

The last one I overclocked which was the same system specs t5810 with a 1650v3 I put that chip in there and used artic silver. Temps ran hotter on that chip and I was only able to get 4.3 out of it. The heat pipes on the bottom of the Dell cooler have space beside them. It's not totally flush and it requires a ton of paste to make a good connection. The paste that Dell uses seems to be much thicker in viscosity and I thought it seemed to work better on their poorly designed heatsink. So if it really is just temps causing it I should be able to max the fans out to test. If it works I'll be buying another better cooler.

I believe your chip maxes out at 1.3v so I wouldnt go doing a liquid nitrogen or fish tank cooler setup but you may have room to go with a little more airflow and contact. Also depends on the paste between the chip and ihs unless you've already delidded and lapped.