Jump to content
HWBOT Community Forums

Nuts and Volts

Members
  • Posts

    15
  • Joined

  • Last visited

About Nuts and Volts

  • Birthday 02/21/2006

Converted

  • Location
    New Zealand

Converted

  • Occupation
    https://www.youtube.com/@Nuts_and_Volts

Converted

  • realname
    Valen

Recent Profile Visitors

454 profile views

Nuts and Volts's Achievements

Apprentice

Apprentice (3/14)

  • Reacting Well Rare
  • Dedicated Rare
  • Collaborator Rare
  • One Year In Rare
  • One Month Later Rare

Recent Badges

0

Reputation

  1. Unleashing New Zealand | HD 7990 | Part 1: Power limits and Temps

    Today's the day, after months of waiting for the last pieces to ship, the last piece of the puzzle finally turned up. Pretesting and overclocking the HD 7990.
    Funny that the HD 7990s codename is New Zealand, despite it not technically being a "Tropical" Island for AMDs nomenclature for these GPUs. (NZ is considered a "Sub-tropical" island.)

    I found it fitting being a kiwi that I should take first with this card; I have a long way to go before I can get gold though. Perfectly segueing into the next part...

    Overclocking; Giving the card a spin in FireStrike with the power limit maxed out and voltage slider cranked to a WhOoPiNg 1.3v, the core maxed out at 1150MHz on both cores and memory at 1650 MHz. Final FS score below \/ , TES disabled along with other tweaks etc. 8th place aint bad and with a better CPU like a 5900X I would be sitting around 5th place. 

    FireStrike Score

    My lil Ryzen 5 3600 is trying its best but should retire soon. The bad single threading performance isn't helping in tests such as Heaven and Vantage...

    Temps were interesting; even with the 180 watt power limits in place for each of the cores, the temps were surprisingly hot at 45-50c during FS (Room ambient at 17c), VRM was reaching 60-65c. I made sure I was running max fans and pump speed for OPTIMAL thermals. I have a couple theories' on why the temps are so high:
    1. Bad contact on the die either from thermal pads or the metal shim
    2. MORE COOLING!!!
    3. Waterblock design

    Regrettably, I don't have the EK water block for comparison, which I would have loved to include. The rest of the cooling loop consists of a PE 360mm CoolStream radiator, SPC 160 Pump/Reservoir combo, CoolerMaster SickleFlow fans, EK Classic STC 10/16mm compression fittings, ZMT tubing, demineralized water, and of course, the XSPC HD 7990 water block (using GD-900 thermal paste and ADWITS thermal pads). Given the elevated temperatures, I'm likely to disassemble the setup once I've either removed the power limit or completed all desired benchmarks. Ideally, I won't need to "delid" the cores if the waterblock is designed to make proper contact with only the cores. The VRM temperature makes sense since the block's internal design lacks a flow channel over the Vcore VRM area. Memory, VDDCI, Display Drive, and the PLX chip are situated over flow channels. While lacking temperature sensors is a frustration, I can reasonably assume they are fine due to their relatively low heat output compared to the Vcore VRM.

    Speaking of the Vcore VRM, voltage control!

    Using a modified version of Trixx to increase the voltage, I noticed that even at the stock 1.3v, the card was throttling. It seemed like an easy fix, so I turned to VBE7 to address this. I made sure to keep copies of both the master and slave BIOS. After flashing both, GPU-Z displayed no changes. I tested FireStrike again to verify, and the card was still throttling at 1150MHz. I downloaded the supposedly flashed BIOS to check if it had even flashed. While VBE7 indicated an increased power limit, the changes weren't reflected in GPU-Z and monitoring. I attempted to increase the limits using HxD and guidance from BZ's blog post. Nada, AMDVBFlash complained that for some cursed reason that it couldn't find the file?!? 

    I'm guessing it's probably a mixture of the last driver version AMD put out along with how the drivers deal with the card's power limit. Attaching the EVC ain't gonna help yet until I'm hitting a current limit from the controller itself. Until then I'm going to keep looking for a way to remove the power limit and really let the HD 7990 stretch its legs. Even at 1150 MHz I still got pretty good eff despite dropping clocks left and right, however I can imagine once it's removed my eff is going to go through the roof. 

    If any of ya'll have a solution please let me know below or on discord (HWbot Discord server) ❤️sUMbxh3K4I6HeAu7zqLVsSsOD966iRGwpWfYI0yzjLfbIeElAXSz5T-Y83piWqsk8HSriXG0UlO69_3FGiRWWpusfKa1_UJeU9LQ4580sj5PUfhtdgLBW0wpdJLvyMiZ6OjSMirB-PKNHX76QY3d69c

    Once I throw on some custom timings and voltage control for the memory this card will finally be running at the speed it should have been a decade ago.

    Next steps for this are going to be removing the power limit, memory timings, EVC, and better cooling. That first place is going to be mine!

    1343436086_3mbImage.png

  2. Bad ASIC Quality ≠ Potato Core| GTX 770 Lightning | Part 1: Voltage Control

    Yep that's right, more Kepler baby! In all seriousness though Kepler is a joy to overclock at ambient for me. Easy voltage scaling, good selection of cards, tons of tweaks and tricks, and SLI if you want to go down that route.

    In this case there was a GTX 770 Lightning for sale on FB marketplace with the OG box and accessories. Not owning a Lightning and wanting to have a second 770 to compare against and torture. Haggled the price down from 100 NZD to 85 plus shipping. Arriving packaged well, the box was in near mint condition.

    After cleaning the card and giving a healthy amount of thermal paste I went straight to GPU-Z to make backups of both the LN2 and Normal BIOS's. Checking the ASIC showed that it was only better than "%30 of GPUs on Techpowerup database". Testing the card at stock voltage would seem to agree with that statement. However I have a "secret" tool, more voltage. Giving a quick test of EVGAs Classified tool not seeming to have any effect on voltage I swapped to drastic measures.

    Not finding a download link for afterburner extreme within the first page of google, I decided to try GALAXs HOF software for the 780 Ti since they not only share the same voltage controller for Vcore, but also Vmem (up1612Q). After getting a copy from StrunkenCap and firing up the non BIOS checked version, raising Vcore to 1.25v and checking with a DMM it works! Memory also works, however the 20mV steps are a little inaccurate to what the real voltage is; make sure to check. Load line also works however I haven't attached the scope to see which is the best setting.

    Speaking of the scope, the GPU reactor on the back of the core. That's a thing which exists, begging to be modded. Part 2 or 3 will probably include upgrading the GPU reactor with a metric butt-ton of caps.

    I was expecting pretty bad temps given the heatsink quality however I was pleasantly surprised that it didn't surpass 60c running Firestrike at 1.4v (Ambient of 15c)

    Not finding a download link for afterburner extreme within the first page of google, I decide to try GALAXs HOF software for the 780 Ti since they not only share the same voltage controller for Vcore, but also Vmem (up1612Q). After getting a copy from StrunkenCap and firing up the non BIOS checked version, raising Vcore to 1.25v and checking with a DMM it works! Memory also works, however the 20mV steps are a little inaccurate to what the real voltage is; make sure to check. Load line also works however I haven't attached the scope to see which is the best setting.

    Speaking of the scope, the GPU reactor on the back of the core. That's a thing which exists, begging to be modded. Part 2 or 3 will probably include upgrading the GPU reactor with a metric butt-ton of caps.

    I was expecting pretty bad temps given the heatsink quality however I was pleasantly surprised that it didn't surpass 60c running Firestrike at 1.4v (Ambient of 15c)
    The fans and shroud I can't say as much though. Who ever designed the way the fans attach to the heatsink I despise. You are gargantuan to end up damaging the finstack and scratching the heatpipes in a valiant attempt to remove them. Some small other details the card has are blue LEDs behind each phase for Vcore and Vmem along with one on each of the fans. 

    Final lazy testing results ended up with the core running at 1424 MHz 1.4v (Voltage read points from the card) and 2000 MHz on the mem at stock voltage. Memory chips are Samsung and increasing voltage from 1.6v to 1.7v did reduce artifacting at higher clock speeds however didn't fully go away. The core still isn't as good as my GTX 770 Classified and certainly not as good as the memory on that card. I would personally say the core is pretty average, maybe a little above. Memory however is mid at best.

    Firestrike results below.

    Next part is probably running with a waterblock attached, extra controls such as switching frequency, scope measurements, and maybe the GPU reactor.

    438888074_GTX770Lightning.png

    Screenshot_7.PNG

  3. Making a "Poor man's Classified" Project | GTX 680 FTW | Part 1: Broken Vapor Chamber

    Recently I purchased a GTX 680 FTW for some more fun with Kepler. Seeing that the FTW shares very a similar VRM to the classified I thought it would be fun to start a project on turning it into a "Poor man's classified".

    I'm talking voltage control for core, memory and PEX, unlocked BIOS etc. Right now that can all wait as I run some baseline tests on the stock cooler. One small problem...

    After thoroughly cleaning the card and reassembling it: repasted the core, however left the pads since I'm out at the moment, and they were in good enough condition.
    I noticed that the core temps were particularly bad even for a blower. Running even a light workload such as GPU-Zs render test would make the card shoot up to 80c within 20 seconds (Room ambient was 17c). Thinking it was a bad mount I removed the heatsink to check if it had bad contact. Nothing as such, fairly even pressure across the die. Reapplied just in case and tried again. Same issue. 

    Suspecting the cooler was at fault and not the thermal paste nor spread, checking the vapor chamber revealed that part of the solder near where the chamber is filled was cracked and had a hole. Giving how much dirt and dust had built up around the vapor chamber near the blower fan, I suspect that it got corroded enough to where the solder cracked and failed. Bit of a shame but let's be honest, the stock cooler wasn't going to be staying on for long after I get voltage control.

    In the meantime I'll be buying a freeze mod VGA block for the card so that I run it with the baseplate still attached for extra cooling on the memory and VRM.
    The Heatkiller block doesn't clear the baseplate and metal clips for the shroud hence a new block.

    Still got some quick results out of the card in FireStrike before it ended up overheating. Memory chips are Hynix (MFR) and seem to do around 1800-1825MHz.
    The core however is a potato. ASIC of %75 and can only do 1190MHz at stock voltage. Core can def run higher clocks however it crashes late into GT2 after temps get up to around 80c when set to high.

    Scores if anyone cares - 
    FireStrike (GPU) ScoreFireStrike Score.

    Stay tuned for future updates as I get this card up to acceptable speed and maybe even deserve the classified title.

    6MB.png

  4. Wait lmao is that the one I modded to 1.5v? It should be the one with +200mv offset from 1.3 stock. I remember posting the file in HWbot. Trixx 4.9.1 is what I'm using.
  5. So dam happy with the set of benchmarks. All of my personal favs. Apart from vantage, vantage takes to dam long...
  6. So far AMD cards seem to be winning, I would love to see crossfire for the comp since more motherboard support it than SLI. PCIe lanes will barely matter for bandwidth.
×
×
  • Create New...