Showing posts with label Benching. Show all posts
Showing posts with label Benching. Show all posts

Tuesday, July 28, 2015

R7 260X Stream 2





So in the last stream I only managed to get through 1 benchmark and that was Unigine Heaven. The card managed 1447/1975mhz at 1.46/1.75V and scored 1804 points.
Since then I've prepared some other benchmarks that and I'll push the card to the limit one more time to get them too.
This time around I should be able to run a higher core clock because the weather has gotten colder and I'm going to do a fan swap from the current 1600RPM slim 120x120mm fan to a full size 2400 RPM 120x120mm fan. This should lower VRAM temps and core temps by a couple C. So 1.5Ghz core maybe?

The stream will be up on my twitch at 6PM GMT this Saturday. If you missed the last one you now have a second chance.

Saturday, July 25, 2015

Live stream aftermath

I'm very happy to say the stream was an absolute success. As far as overclocking goes anyway. The R7 260X hit a monstrous 1447/1975mhz at 1.46V core and 1.75V memory and got a score of 1804 points in Unigine extreme. So that's great and all. As far as streaming success goes I think that having only 3 viewers watching the stream speaks for itself. If you wanted to watch this one and missed it well I got good news for you. I will be doing more streams. The next stream will be about the HD 5850 using the water chiller and I plan to make that one pretty interesting because I'm going to be doing the insulation and volt modding of the card on the stream not ahead of time like I did with the 260X. Right now I'm not sure when I'll do the stream but I know it will be on a weekend and it'll either be in August or October because in September I'm moving to the UK.

Also here's a picture of the setup from the stream:

As always thank you to Cooler Master for supporting me.

Wednesday, July 22, 2015

Live stream subzero HD 5850 and the insanity R7 260X.

I'm very sorry for not posting any actual content for the last 23 days. So to make it up to you all I've decided that I will do an overclocking live stream starting Saturday at 4:00GMT. Here are some teasers of the things getting benched:











Other than that I've dropped the you-name-it.com ad because I can't get the Athlon 460 X3 they sent me to work. Which sucks because I wanted to do 3 core Cinebench on LN2 with it. However until this gets resolved I can't recommend them as a good source of CPUs.

Thanks to CoolerMaster for making so much of what I do possible.

Monday, March 16, 2015

Fan review: CoolerMaster BladeMaster 2400RPM


The 4-pin fan header is braided black. This is a nice touch to the overall aesthetic.










Testing Setup:
For testing the fans I'm using my fan test bench. For temperature testing I ran IBT and measured temperatures using AMD Overdrive. I did the noise tests in a room with an ambient noise level of 25dB. The fan was tested at first on a mattress, to absord vibrations. The second test was with the fan mounted to a Corsair H100. To measure the noise level of the fan,I used the Sound Meter app on my Gsmart Mika M2, which was placed about 15cm away.

These are the CPU power settings:
Clock = 4.5Ghz
VCC = 1.525 High LLC
VNB = 1.2 Medium LLC

Cooler Master Blade Master 2400
Fan Specs
Dimensions: 120mm x 120mm x 25mm
RPM: 2400
Power Draw: 0.37A @ 12V

Performance (ambient 25-26C°, 100% FAN SPEED):
Noise level (lower is better): 54B
Noise level radiator (lower is better): 55B
CPU thermal margin IBT load (higher is better): 19.25­C°

Sunday, March 8, 2015

ACCPU CEO Q&A

AC CPU is an American startup aiming to make phase change cooling more accessible to the average person. By both making it cheaper and more simple to use than what is currently available. This is the transcript from an interview with the CEO of the company that we had over skype text chat. I'm in no way being paid/sponsored by AC CPU and I'm only doing this out of my own interest in what they have demonstrated on reddit.

Buildzoid: Why don't you start this of with what you already have?

Matt Russell: Have you seen our new video? We just started a promotional campaign on Facebook to get our Kickstarter out into the aether and the video is the main focus. https://www.facebook.com/ACCPUInc

Matt Russell: Also, this is the preview link for the Kickstarter. It has the video too.

Matt Russell: https://www.kickstarter.com/projects/598954186/938010357?token=f3adf510

Buildzoid: The coil is just to demonstrate that it can cool it is not what the final evaporator will be based on right?

Matt Russell: Absolutely. The final coil will look more simply like a copper block. We are researching casting techniques with 3D prints right now and should have something to show in blueprint form within the week.

Matt Russell: That coil is ugly. Nothing you can say about it. But it’s the fastest and most efficient way to test different configurations.

Matt Russell: Kickstarter requires that you prove your idea works before you offer it up in Kickstarter. That’s one of the ways they have really changed in response to the lack of accountability complaints that they were getting. We had to jump through quite a few hoops to get our project approved, essentially proving that it would work before they would allow us to seek funding through them.

Matt Russell: So though it’s an ugly half-solution, it met the purposes of showing that our idea was based on sound engineering and wasn’t nonsense.

Matt Russell: This is a very rough engineering mockup of what our heat sink will resemble in the future, which is what that coil will really be housed within.

Buildzoid: Ok cool. So the noise of the compressor in the video and the fan are also just part of the prototype

Matt Russell: Absolutely. The fan is currently run by a drill, even though it was originally part of an air compressor. For the purposes of quick removal and control of the confessor temperatures for testing, it was simply easier to mount it on a cordless drill. As well, that compressor is much larger than what we will be using in production.

Matt Russell: Between that heat sink housing the evaporator coil, and the compressor, you have the very essence of why our design is vastly advanced.

Buildzoid: Since we're talking about the fan radiator already we might as well get out of the way the fact that the maximum heat transfer of the compressor is 608W can you fit a radiator of that power into a PC with reasonable noise levels.

Matt Russell: The radiator is in two sections. In HVAC they are referred two as the ‘evaporator coil’ and the ‘condenser coil’. One is hot (condensor) and one is cold (evaporator). We are confident that we will be able to house the evaporator coil in our heat sink so that all cooling is done essentially right on top of the CPU. That’s what our Sketchup file will show you. It’s about 10 feet of copper tubing stuffed into a 8cm cube.

Matt Russell:  As for the condenser coil, that radiator is likely to go in a common liquid cooling loop radiator. That is something that we are going to need to narrow down post-Kickstarter since we simply haven’t got a bunch of expensive liquid radiator lying around. With some slight modification of the fill ports on a liquid loop radiator, we should have plenty of surface area to maintain the condenser temp at 100F. As the condenser temp rises, you simply lose cooling load capacity, the system continues to work, and our compressor is measured up to a condenser temp of 140F.

Matt Russell: As the condenser temp hits 140, you are still at a capacity of 482 watts.

Buildzoid: So let me get this straight. Once you get the refrigarant compressed it will be at 37C and get shoved into a water coooling radiator from where it will then go to the cooling coil/evaporator block or am I missunderstanding something?

Matt Russell: Compressor hits around 300 psi, the compressed gas goes to the condenser where it loses some of the heat gained from compression and then turns into a liquid by the end of the condenser coil. From there, it travels to a metering device which in our case is a capillary tube. Because the capillary tube is very very tiny in diameter, only a small amount can go through despite the tremendous pressure behind it. As the liquid refrigerant exits the cap. tube the drastic drop in pressure causes it to boil (which in the case of r134a at 13 psi. makes a temp of -13C. As the refrigerant boils it cools the surrounding metal to the boiling temperature and turns into a gas by the time it returns the compressor.

Matt Russell: When we go down to 1.2 psi in the evap lines, then we can hit -24C.

Buildzoid: So the condenser will be a watercooling radiator right?

Matt Russell: That’s one of the possibilities that we are looking at. If for whatever reason that is not feasible (which we really believe it will be) then we have other options like an automotive radiator which is definitely able to handle these pressures.

Matt Russell: We only are looking at water cooling radiators because they can come in a very large size if we need it, and they can be installed into a normal PC case.

Buildzoid: Ok cool so I didn't misunderstnad anything. With watercooling radiatorsa I know for a fact that a full copper thick 360mm Radiator with 1850RPM fans does about 300W heat dissipation with a delta T of about 5C between the water and the ambient air. Now I'm not 100% sure what that will give you at 37C compared to 22/24C ambient but it should be plenty to take care of most CPUs. Now then is there any chance of using a more powerful refrigarant like R404 or would those not work?

Matt Russell: R134a is a great refrigerant for most users because you can buy it at an auto parts store to refill the system (which shouldn’t happen but with tinkerers it can.) Also it can be purchased in 1 lb. cans. whereas most other refrigerants are purchased in 25 lb. or larger cylinders which (trust me I just bought a 410a cylinder) cost around $130.

Matt Russell: 404, 410a, ammonia, they would all work relatively the same (with different pressure/temperature requirements) but for us it’s more an issue of ease of use for the end user. Specifically, we are looking at the user who doesn’t want to pay for the whole complete system as he already has the radiator, or other parts and just wants to add our compressor. It would be a huge cost increase for that builder if we used 410a, because even though it would be a better refrigerant in terms of cooling capacity, it would be more dangerous because of its higher pressures, and more expensive because it is sold in larger containers.

Buildzoid: But people with access to better refrigerants could use them if they wanted to right?

Matt Russell: Yes. As long as the vacated the system of the POE oil that 410a and 134a use and make sure that they replace it. A lot of systems use mineral oil instead, or a proprietary oil blend. If they don’t properly replace the oil, the compressor will destroy itself just like any other machine.

Matt Russell: One thing we will have to keep our eye on is the proposed phaseout of 134a in the EU by 2017. The replacement they are looking at is Carbon Dioxide and we will have to look at testing there. However, I doubt the compressor will be able to handle the ridiculous pressure that Carbon Dioxide requires to function as a refrigerant.

Matt Russell: It is worth noting that 404a, 134a, and 410a have all been tested in the compressor by the manufacturer with great results.

Buildzoid: Ok well that's some good news for the really hardcore crowd. This is now the longest post on my blog ever so lets wrap this up. What is the minimum temperature that a 300W load could get to using your cooler.

Matt Russell: 4C.

Matt Russell: But that’s what it will maintain, not the minimum possible.

Matt Russell: If you are in a cooler room than testing conditions for example, you will go lower.

Matt Russell: If we look into a larger compressor which the manufacturer also offers, we would be able to hit -1C with the same conditions.

Buildzoid: That's very impressive for something so small and and basically means Haswell users could run 1.5V through their chips all day every day. Now then what will the pricing and buisness model of AC CPU be.

Matt Russell: At this point I can give you a range of possibility which hinges on how much we can make the heat sink and purchase the compressor for.

Matt Russell: The range is looking like between $499 - $699 very much depending on research in the Kickstarter phase. We simply can’t fund the project on our own anymore, and in order to bring this to the community we need funding. My wife is going to kill me if I put any more of my own money into this. We have probably spent $2,000 of our own money bringing it this far, and we are looking of at least another $2,500 to go on into testing of components to give a better price range.

Matt Russell: The business plan is this. We will sell plans to do everything yourself for $20 including part recommendations, how to take apart a window AC unit, etc.
On our site, we will also sell the ‘lego kit’ assembly method so that users can create their own custom system without buying the many parts (pipe flaring kit, copper tubing, etc) and pieces of equipment to make their own. We are only looking for a small markup in order to cover our costs and labor.
We will release all of our concepts so that if someone wants to create this system on their own without paying $20 for plans, they can do the same thing.
And we will offer support to everyone. We are here for the community, because we are a part of the community. I overclock and build PCs. So does my partner Dave. Between us we have put together 4 PCs this year, and it’s March. I can not wait until someone takes our system up to HWBOT and wins the rookie challenge. It’s going to be awesome to contribute to the community and change it for the better.

Buildzoid: Sounds good to me. It's a little beyond the cost of high end CPU watercooling loops so how long can one expect the compressor to last?

Matt Russell: My manufacturer sells compressors to the US military which have been in continuous use in MRAP vehicles in Afghanistan for over 5 years. They show no signs of failing. We chose a compressor and system with as few moving parts as possible. Seriously, outside of the compressor there are no moving parts. At all. Nothing to break.

Matt Russell: Also, if you spill liquid in your CPU, you are in a bad way. Refrigerant will boil away and leave the system much much faster to do less damage.

Buildzoid: From what google gave me R134a is not conductive so if you do spill it no damage will happen beyond any condesation it might cause. Well I think this is a good place to end so goodbye and I hope this works out for you.

Matt Russell: Thanks a lot.

Sunday, January 4, 2015

AMD RAM Timing Guide

Voltages
The usual DDR3 voltages for 24/7 are anything up to 1.725V actual(checked with DMM). For benching I wouldn't go above 1.85V on air cooling but if but if the sticks scale well with voltage 1.9V is still reasonable depending on what sticks you have.
With AMD chips the north bridge plays an important role in just how high clocks you can run so once going above DDR3 2400 you will need to raise the voltage. For 24/7 on 32nm and 28nm CPUs stay below 1.35V and for benching stay bellow 1.4V.

Primary Timings
The primaries behave like primaries on intel and depend entirely on your kit of RAM. You can run lower ones with more voltage.
The only rule for the primaries is that CL+tRCD+tRP=tRAS ±3

Secondary Timings
The Secondary timings have a massive impact on how high you can clock your RAM. Higher numbers will usually yield higher clocks but going to far will make your system unbootable so do not just max all of them out lower timings will get you more performance for a set frequency.
A free timing is a timing that you can adjust independently from all the other timings without losing stability for any other reason than being too low.


Gigabyte ASUS MSI Asrock Dependancy
tRC WIP WIP WIP free
tRRD WIP WIP WIP free
tWTR WIP WIP WIP free
tWR WIP WIP WIP previous*2±2
tWL WIP WIP WIP previous-3
tRFC0 WIP WIP WIP free
tRFC1 WIP WIP WIP free
tRTP  WIP WIP WIP free
tFAW WIP WIP WIP previous*4
Command Rate WIP WIP WIP free




AMD only settings


Processor on die termination.
Settings: 240ohms 120ohms 80ohms 60ohms.
Lower on die termination allows running higher frequencies.

Drive Strengths:
Higher drive strengths allow higher clocks and lower timings. The DQS and DATA drive strengths top out at 1.25x for me regardless of what I do. I set all the other settings to the absolute maximum of either 2x or 1.5x.

I do not have any pictures because different motherboards have different timing table layouts so you'll just have to read them all anyway.

Friday, December 26, 2014

Friday bench day: 5.16Ghz FX 6350 Cinebench Rampage

This is the secondary computer I have at my grandma's house. Based around an FX 6350 8GB of Corsair Vengeance RAM and a ASUS Sabertooth 990FX motherboard. And since it finally started to snow a little I decided it was the ideal time to do some benching on low ambient temperatures.

So after a few hours of having both windows in my small tower(seriously it's a tower) room open the temperature reached around 8C° at the intake of my Phanteks heatsink.
Cinebench was the first benchmark I ever used and since it's the only benchmark I have installed on this computer that doesn't greatly benefit from a clean install of windows I did Cinebench 15 and Cinebench R11.5. In R15 I got 624points at  5.16Ghz. That's a nice 3rd place on the HWbot.org FX6350 leader board. 
I'll be honest I'm proud of my efficiency on this run because I was running on my Corsair Vengeance sh*t sticks.

I managed to get another 0.2 points in Cinebench R11.5 over my 5Ghz score. Unfortunately the cold got to me and I forgot to do a second run and so the score is ever so slightly lower than it could be.
I'll be doing another session soon on my 2333 9-11-10-32 G.skill Pis. Also it looks like this CPU could run 5.16Ghz all day everyday if I just had a good cooler because all these runs were done on just 1.6V with High LLC settings.
Here's a challenge to all my readers who have an FX 6350. The first to beat my score in either Cinebench R15 or R11.5 by the 27th of January 2015 and I'll give you 10,000 DOGE. !NO LN2 and leave a link to a pic of your system and your dogecoin address in the comments!

BTW I updated the support this blog page to include sources of free dogecoin. If every time someone goes on this blog they went to this site and set the dogecoin to this blog I could get or buy a fan to review every month and it just takes a few clicks.

Sunday, December 21, 2014

ASUS now has it's own external VRM!

This is the new external VRM from ASUS. It's built for up to 500A on 8 phases and a maximum voltage of 2.5V.
You can follow all updates about it in this thread.

Saturday, December 20, 2014

The RAM and FAN test bench overview



 Built on an IKEA pine wood wine rack. For low weight.
 With WiFi for easy screenshot uploading and maximum portability.  (Mounted with zipties)
 Based on the Gigabyte F2A88X-D3H I reviewed ages ago.
(Also mounted with zipties)
 Insulated for use with LN2.

and with a crocodile clip for reading CPU core voltage directly










Running an Athlon II X4 750K cooled by a cooler master Seidon 120V closed loop cooler.
(The rad really is just sitting there)








With video provided by my modified R7 260X
(The cooler is mounted with zipties)
(If I get a bigger PSU and another CPU cooler I will replace it with the water cooled GTX 590)







 The RAM currently in this beast is the G.skill ECO 2x2GB 1333 7-7-7-21 1.35V kit I previewed some time ago.







 The HDD is a 350GB WD blue.
(Mounted with even more zipties)
The PSU is an EVGA 430W 80+ unit.
(Held in place by zipties)
This is built from scrap parts from other projects(GPU MB and CPU Cooler) and my friends old PC(wifi and HDD). The reason why I built this is that my 3960X is never ever going to boot with RAM above 2500mhz so I can't use it for RAM reviews. This 750K already managed to boot 2520mhz on these G.skill sticks so it can do a better job with RAM than my main system. Ideally this would be a Z97 system but good Z97 boards cost about 1.5-3X what this board cost and an i5 4690K is about 3X the price of the 750K. Yes I know there is the Pentium G3258 but IMO pairing a 3000czk MB with a 1700czk CPU is stupid. Plus AMD CPUs are funner to OC.
The idea for using the wine rack from IKEA is that. It can hold as much stuff as a full tower, cost only 199czk(10$) and weighs less than any other case you can buy.
I used the cooler master Seidon for cooling because it allows me to easily swap fans for fan reviews.
If I had to change any things I would get a better PSU(750W Seasonic/EVGA) and a better MB(Crossblade Ranger).

Monday, December 1, 2014

R9 290/R9 290X Refrence PCB Overclocking Guide

The R9 290 and R9 290X are AMD's current flagship single core GPUs. I have 1 of each running in crossfire and have benched both of them extensively.
I expect that you know what is where in afterburner and you know how to configure custom fan profiles.

Software you will need
For daily use and aircooled benchmarking:
MSI afterburner
Sapphire Trixx
For benching on water or LN2
PT1/PT3 BIOS (link in useful links page)
GPU Tweak 2V (link in useful links page)
AtiWinFlash (link in useful links page)

Air 24/7
The Hawaii GPUs are pretty tanky so they can be overvolted for years without any negative effects. For 24/7 overclocking you will only need MSI Afterburner because going above +100mv on the core voltage generally doesn't help your frequency margin much and is about as high as you want to go for 24/7 overclock. There is no reason not to set the power limit to 150% so just set it to 150% and forget about it.

The Core
Most Hawaii GPUs do about 1100mhz without raising the core voltage. Once your card can't go any higher on stock volts start raising the voltage by 12mV. While voltage does increase frequency ranges AMD cards are really sensitive to temperatures and so the increased operating temperatures can counter act the increase in voltage. This is not an issue for people on water cooling but for people on air coolers I recommend tweaking the fan curve too avoid going above 85C° however if you can tolerate the noise of your cooler at 90% just set it to hit 90% at 70C° and you should be all good temp wise.
Afterburner has an AUX voltage setting for the Hawaii cards. Unfortunately no one really knows what the best setting for it is and I haven't fully test it yet so you will just have to test it at settings between -50mV and +100mV going by 50mV increments. I have mine at +100mv to get 1140/1550mhz on my Windforce R9 290X. I do suspect however that the AUX voltage somehow impacts the core voltage either in stability or how high it is.

The VRAM
The VRAM on Hawaii cards is very different from previous cards. ALL Hawaii cards have a golden ratio of
VRAM clock/core clock. This ratio varies from card to card but is generally between 1.25 and 1.4. Once you find this ratio it will generally allow you to run a much higher memory clock than try to incrementally raise VRAM frequency. For example my R9 290X has a ratio of 1.36 and 1140/1500 is not stable on it but 1140/1550 is even though they are at the same voltage.
While the Hawaii cards do not have memory voltage controls available in any software the memory clock does scale with core voltage so if you are trying to push a high memory clock you wil need to raise the core voltage regardless of your core frequency.

For some reason Hawaii cards don't always downclock  the memory when idling so when you are going above 1350mhz VRAM with an OCed monitor or on a multi monitor setup you will need to set Afterburner to constant voltage. This will not negatively impact the life span of you card because the idle core voltage will still be under 1V. I also recommend trying this if you crash when pressing apply because that happens for the same reason. Your VRAM OC kicks in but the voltage stays at idle levels and you get a crash.

Water 24/7
Do what you did for air 24/7 but use sapphire Trixx to get above +100mv core voltage.

Air Benching
Do what you did for the Air 24/7 but use Afterburner to set your AUX voltage then use Sapphire Trixx to get +200mv core. If you have an air cooler just max the fan speed since you're benching.

Water/LN2 Benching
!! COMING SOON !! NOT because my card does not like the available LN2 BIOSs sorry

Friday, November 7, 2014

Air benching my WTF cooling R7 260X


My good old R7 260X from WTF cooling has gotten even more WTF and now does 1450+mhz on the core.



And this is the V mod I did on the GPU. I finally got my hands on a 10Kohm potentiometer so I got the core voltage under manual control. Unfortunately a 10Kohm potentiometer is not enough to keep the voltage in "safe" territory and so the card has a minimum 3D load voltage of 1.47V. The 3rd wire on the back is hooked directly to the + leg of one of the output caps and ends in crocodile clip to allow for easy measurement of core voltage with a DMM.  
                                               Here's a shot of the side of the GPU. You can see both the potentiometer and the crocodile clip behind it
The screws on VRAM in this and the shot above are there to cool the memory because in my first runs the memory overheated and crashed when above 1670mhz. That's a massive problem on a card that has an anemic 128bit bus feeding cores 896 stream processors running at 1450mhz.


And here's the card running in my main system. The Gelid heatsink did a great job and kept the core bellow 70C° throughout my benching session. The Hynix VRAM on the other hand was terrible and kept me bellow the 1600point mark in Unigine Heaven DX11 Extreme.

Sunday, October 12, 2014

G.skill ECO 1333 7-7-7-21 2x2GB Early Expirience


Not long ago I bought a G.skill ECO 2x2GB 1333 7-7-7-21 1.35V ram kit. Now the stock settings on the kit may not sound like much but these things are great. Within 35 minutes of playing with them I had them booting 2400mhz. With better timings than many 2400mhz that you will find available in stores. Now they were not the miracles that I wanted because these sticks use PSC ICs and those are know to do as much as 2600 8-12-8-28 @ 1.85V on air cooling. Now this was early testing so I wasn't pushing very high voltages but it is very obvious that I have a pretty bad PSC kit. The settings I ended up with were 2400mhz 10-12-11-33-1T with sub timings of 5-160-12-6-24-6-7 and 1-1-1-1-3-3-2-1-3-1 using 1.75V. The primary timings are pretty bad as they are barely better than my 24/7 sticks which do 2400 10-12-12-34-2T. However the secondaries and tertiaries are pretty damn good.

I hope that I can get this kit to run CL 9/8/7 above 2400-2000mhz since that should put them way ahead of my 2000mhz 8-9-8-24 Ballistix Elite that got me my Super Pi 32M record.
I will have more updates after I'm done with the kit.

My plans for this kit are to see how high and how low it can go. Now frequencies above 2400mhz are really problematic for the IMC on Sandybridge-e CPUs so I doubt I will hit above 2500 even if the sticks are capable of that. However what I'm more intrested in is just how low I can take the CL because a low CL from my experience is key to getting a HWbot prime score so if these sticks could boot CL6/5 at 1333-1600mhz that be great.

UPDATE:
The kit is optimal at 2333 9-12-11-32-1T with secondaries of 5-156-12-6-24-6-6 using 1.8V. I still haven't done any benching with these because I need to setup my new benching HDD.
I still have to try top the frequency and try the low CL(5,6,7) max clock before doing a full write up this weekend.

Saturday, October 4, 2014

Voltmodding GPUs using the NCP and PCP 81022 voltage controllers

So HWbot recently launched this. I think the new division system is awesome and as someone planning to participate in division 5 I decided that I'll help everyone by compiling the available information on the NCP80122 controller found on the AMD reference design R9 285 and R9 260(X) cards.

So here is the NCP81022
The red pin controls the Vcore you can solder a 10K ohm variable resistor(VR) to this and the ground to get control over the core voltage.
The 2 green pins control the over current protection. By increasing the resistance between the 2 you will get a higher current limit. The only problem is that you have to find the resistor that these pins are attached to because I couldn't find a good enough photo and because R9 285 PCB designs differ quite a lot. If you want to completely disable over current protection just remove the resistor that these pins are attached to.
If soldering directly onto the IC's pins scares you(me too they are freaking tiny) then find the first resistor that the pin connects to and solder your VR onto the resistor. If you don't have a VR you can try using pencil.
All these mods are universally applicable to GPU using the NCP80122 or PCP81022 voltage controller.

Sources:
http://forum.hwbot.org/showthread.php?t=75953
http://www.techpowerup.com/reviews/Sapphire/R9_285_Dual-X_OC/4.html

Sunday, September 21, 2014

I hate digging up crap to write about but oh well here we go SuperPi tweak guide.

The title says it all so lets get going.
SuperPi 32M is a completely single threaded benchmark and is currently one of the most popular benchmarks on HWbot so getting a good score in it can get you a ton of points.


Software
First the OS, you've got to use Windows XP. SuperPi is just weird like that. Windows 7 also works but you get slightly higher times and can't do copy wazza.

Once you have XP installed go into the services and disable everything except for WMI, log event, futuremark sys, windows installer, intel management, drivers, plugnplay and themes.

When benching SuperPi you keep themes enabled because for some reason it runs faster with the olive green theme.

Use task-manager to shutdown as many unnecessary tasks as possible also set SuperPi's affinity to core 1. Set all other tasks to core 0 and put their priority to low. Set SuperPi's priority to high

A really important tweak for getting a good SuperPi time is copy wazza. Here's a video of Splave doing copy wazza.

Hardware
On intel CPUs disable hyperthreading because it adds extra load onto the IMC and only run with 2 cores to increase OC range and lower heat output.

For Haswell CPUs try to get the Uncore/Cache ratio as high as possible.

For AMD CPUs clock the north bridge as high as possible.

For ram you will want to use PSC or Samsung HCH9 as they are capable of high clocks with very low timings. PSCs come in 2GB sticks from a variety of manufacturers but Samsung HCH9s are found on 4GB 2600 10-12-12-34 G.skill sticks.
Another good series of ram to use are 4GB stick Crucial Ballistix Elite that come stock at 1600 8-8-8-24 or 1866 9-9-9-27. These are capable of 2000 8-9-8-25 @ 1.65V the only down side is that they don't scale past that regardless of the voltage I used. But they are still very good sticks that can also serve as daily drivers.


Sorry for not posting for so long. School happened.

Thursday, July 10, 2014

R9 290X VRM design specifications run down.

MSI R9 290X Lighting Physical phases: 12+2+3
Controller: IR 3567B 6+2 for the core and auxiliary. I can't find the memory voltage controller without having the card in my hand. You get control over the 3 key voltages and then you gotta pray that MSI's BIOS and AMD's drivers won't screw you over.

ASUS R9 290X Matrix
12+2+2
Controller: A re-branded  IR 3567B 6+2 0.2Mhz-2Mhz for the core and auxiliary. I can't find the memory voltage controller without having the card in my hand. This is the only card that will let you manually tune the controller's operating frequency and all voltages that exist on an R9 290X.

ASUS R9 290(X) DirectCU II
6+2+2
Controller: A re-branded  IR 3567B 6+2 0.2Mhz-2Mhz for the core and auxiliary. I can't find the memory voltage controller because I never had the chance to take the card apart.

Sapphire R9 290X Vapor-X
12 total 8+2+2
Controller: IR 3567B 6+2 0.2Mhz-2Mhz. The 8 phase design is very likely not a true 8 phase design and is using doubler but luckily for anyone reading this blog I'll be getting a sample so I'll have exact specs then.

Sapphire R9 290 Vapor-X
8 6+1+1
Controller: IR 3567B 6+2 0.2Mhz-2Mhz. 6 Phases for the core and one for auxiliary. Memory has it's own static voltage controller. Unfortunately all the 71A MOSFETs of the AMD reference design have been replaced by 40A IRPowerStages leaving you with only 240A of current for the core and 40A for the auxiliary. The memory still uses the 71A MOSFETs.

ALL other models
5+1+1
Controller: IR 3567B 6+2 0.2Mhz-2Mhz for the core and auxiliary. I can't find the memory voltage controller because there isn't one, no really there isn't the voltage is 1.5XXV idle and drops to 1.4XXV in load which got me really confused when benchmarking Unigine one time.

BTW I will be getting the parts for my AMD rig tomorrow so prepare for even more reviews and guides.

Wednesday, June 11, 2014

GTX 590 el cheapo watercooling

So this how I decided to cool my GTX 590 for when I try benching it with an addon VRM. The coolers I used are Coolermaster Seidon 120Vs. I flipped the intel mounting bracket upside down removed the PCI-e slot bracket and used zipties to attach the coolers to the card.
Here's a Diagram of all the screw holes on the GPU's PCB:
______________________________________
|1       2                     3                4            5|
|       6       7                                 9     10    |
|                                8                               | 
|       11      12                              13     14   |
|__15______16______17_______18______20|
|           ]]]]]]]]]]]]]]]]]]]] [

The first unit has zipties going through holes 1, 3, 15 and 16.
The second unit has zipties going through holes 3, 5, 17 and 20.

On my first mounting attempt I managed to bend the PCB by tightening the zipties too much. Luckily I just had to redo the mounting. The end result of this is that the GPU takes up ALL expansion slots bellow it. I do believe that you could do the mounting slightly different and reduce the number of occupied slots to just 5 but for my purposes of running the card on a test bench this mounting proved most effective. Another thing that isn't necessary is removing the PCI-e slot bracket because you can have that attached by the DVI ports without having it take up screw hole #1. However even with the PCI-e bracket this is definitely not something you would want sitting in your tower case but it is great for open air test benches and it manages to run the card at 51C at stock clocks which is a 34C improvement over the stock cooler. The other good thing about this cooling method is that you still have enough space to attach addon VRMs like the EVGA E-power.