Go Back   Hardware Canucks > PC BUILDERS & TWEAKERS CORNER > Overclocking, Tweaking and Benchmarking

    
Reply
 
LinkBack Thread Tools Display Modes
  #1 (permalink)  
Old October 11, 2013, 01:08 AM
MVP
 
Join Date: Oct 2012
Location: North East Alberta
Posts: 447

My System Specs

Default Understanding Graphic Card Memory Clock

So I've been playing around with my new HD 7970 card and tweaking it and running 3D Mark 11 benchmarks to see how the tweaks work.

I can understand the graphic card core, the "OC" version I have comes clocked at 1000. I was able to take that up to 1060 before the system would reboot when I ran the Mark 11 benchmark. I've backed that down to 1050.

Curiously this does not make it a Ghz edition because the memory clock is still 5500 Mhz not 6000 (among some other things related to boost).

What I don't understand is the clock memory.

My results show my clock memory stock at 1375, which I was able to take up to 1515 and still run a Mark 11 benchmark without the system shutting down. I've backed that down to 1475.

Yet the memory clock is supposed to be 5500 Mhz?

Simple arithmetic tells me 1375x4=5500Mhz but why is only one quarter of the memory clock reported?

So when I compare my overclocking with the Hardware Canucks overclocking of the same card I see that they got 1198 (I got 1060) and they got 6204 while my 1515x4=6060

I'm doing this on an AMD A10 APU - they were using a i7. Does that explain the difference.

OR should I keep tweaking in hopes of narrowing that gap?

My best Mark 11 score was 7391 - quite a bit less than the 8072 HWC got with their system but again my A10 ain't going to compete with an i7 on overall scores.
__________________
________________________________

FKD
Fort Kent, Alberta

2012 - A10 Rig
Reply With Quote
  #2 (permalink)  
Old October 11, 2013, 10:13 AM
"Quote This..."
F@H
 
Join Date: Nov 2007
Location: Hell
Posts: 3,833
Default

The GPU memory is quad data rate. They are multiplying the memclock by 4 to get that result. think it it as the Memory Base clock.
Reply With Quote
  #3 (permalink)  
Old October 11, 2013, 01:05 PM
Allstar
 
Join Date: Oct 2011
Location: Edmonton
Posts: 646

My System Specs

Default

HWC are pros at overclocking, but maybe there's some things you're forgetting.

Stability is inversely related to temps. Running your gpu fans @ 100% with a ton of positive pressure in your case, and direct flow onto the card ensures low temp air for your cards.

Now there is a reason stock clocks are so much lower than max clocks. The top speed a card can reach is a distribution. So one card can be a lot better than another. So all cards are clocked to this minimum standard, and will be stable under standard fan profiles. This is also the reason why some cards have shaders turned off; they are a bit unstable under standard fan profiles.

So when you overclock your card you need to realize that it might overclock a little or a lot, and it's pretty much out of your hands at a certain point. You can liquid cool, but at some point it is still unstable, even under lN2 cooling.

Using an a10 vs an i7 should not effect your overclocking by one mHz. However, the quality of the motherboard can have an effect on overclocking, but not usually for just one card.

For example, my motherboard has an extra power connector on it just for the video cards. It's really odd seeing a 4 pin molex connector on the mobo, but I can't bitch, my cards were stable at very high GPU clocks when mining. When gaming I use stock clocks.

7391 score on an a10 is very good. But for overall 3d mark scores the CPU is going to matter.

Basically if you can overclock better than HWC reviewers you're a pro
Reply With Quote
  #4 (permalink)  
Old October 11, 2013, 01:36 PM
enaberif's Avatar
Hall Of Fame
 
Join Date: Dec 2006
Location: Calgahree, AB
Posts: 10,679
Default

Quote:
Originally Posted by Masteroderus View Post
HWC are pros at overclocking, but maybe there's some things you're forgetting.

Stability is inversely related to temps. Running your gpu fans @ 100% with a ton of positive pressure in your case, and direct flow onto the card ensures low temp air for your cards.

Now there is a reason stock clocks are so much lower than max clocks. The top speed a card can reach is a distribution. So one card can be a lot better than another. So all cards are clocked to this minimum standard, and will be stable under standard fan profiles. This is also the reason why some cards have shaders turned off; they are a bit unstable under standard fan profiles.

So when you overclock your card you need to realize that it might overclock a little or a lot, and it's pretty much out of your hands at a certain point. You can liquid cool, but at some point it is still unstable, even under lN2 cooling.

Using an a10 vs an i7 should not effect your overclocking by one mHz. However, the quality of the motherboard can have an effect on overclocking, but not usually for just one card.

For example, my motherboard has an extra power connector on it just for the video cards. It's really odd seeing a 4 pin molex connector on the mobo, but I can't bitch, my cards were stable at very high GPU clocks when mining. When gaming I use stock clocks.

7391 score on an a10 is very good. But for overall 3d mark scores the CPU is going to matter.

Basically if you can overclock better than HWC reviewers you're a pro
Stability has a lot to do with the chip itself not so much temperatures. You could have a chip run at 20 degrees or 80 degrees and have no issues.

The issue with stability is that a chip needs voltage to remain stable and by increasing voltage you increase heat. At some point no matter the amount of voltage you feed it the chip simply can not remain stable and crashes.

A perfect example of this was with my i5 2500k. I could get it to run anywhere between 4.0 and 4.7ghz with low temperatures due to my cooling setup but the chip would simply not take any more voltage to keep itself steady and thus would crash.

So in return I'd have to turn the cpu speed down and I could decrease the voltages and the chip would remain stable.

This had absolute NOTHING to do with temps but more of voltage limitations of the chip itself.
Reply With Quote
  #5 (permalink)  
Old October 11, 2013, 02:23 PM
trodas's Avatar
MVP
 
Join Date: Feb 2008
Location: Czech republic
Posts: 352

My System Specs

Wink

To put long story short - it is always luck. If you are lucky and your chip can overclock good (and the voltage regulation can handle it), then you can surpass anyone, if they are unlucky and have a poor chip.
And there is no way of telling.

Futhermore, some speculations are, that in some cases, the boards/cards sent to reviewers are cherry picked, witch means that you cannot compete with them at all, since they are in another league - by the HW itself.

I firmly state, that unless we talk about extreme cooling and modifications of the hardware, then the skills of the overclocker hardly matter.
Even beginner can - with some patience and determination - get to the same maximal clocks as the professional, on the same hardware. The problem is, that no two pieces of hardware is ever the exactly same. Simply it is not. Everything differ - caps, voltage regulation chips, soldering joints... and mainly the chip itself. If you get chip from a good waffer, then you can beat anyone. If your chip is from poor waffer, well, there is not much you can do about it.
Not even super-pro-overclocker can help you at all and do "magic" with the HW.

Overclockers are just people, dead fortkentdad :) Some are just lucky suxxkas, playing with given super-toys, not knowing what they ever doing.

You can quickly determine what good are they by simply asking questions about certain bios settings (for example for ram, ram timings) and you quickly realize, that many times they never know, what they are even talking about. Some know, that someone say it have to be this way, but to explain why this setting should be that - they cannot come with the real answer.

So, don't overrate people. In overclockin it is your HW against theirs. Get a better HW, modify your HW (replace current with better components, if the chip show a overclocking potencial - if it show 1000 to 1060Hz "OC" then it would be a waste of time and resources) and you can kick the a$$ of them :)
__________________
"It is dangerous to be right in matters on which the established authorities are wrong." - Voltaire
...just keep folding, just keep folding... :) my config
Reply With Quote
  #6 (permalink)  
Old October 12, 2013, 10:15 AM
"Quote This..."
F@H
 
Join Date: Nov 2007
Location: Hell
Posts: 3,833
Default

You are going to have better luck with a factory overclocked card too as they tend to cherry pick (Bin) the chips for those.
Reply With Quote
  #7 (permalink)  
Old October 12, 2013, 05:16 PM
trodas's Avatar
MVP
 
Join Date: Feb 2008
Location: Czech republic
Posts: 352

My System Specs

Default

Let's hope so.
Sometimes I fear that their pre-selection based on the leakage current is either not very accurate, or...

Back in ancient times, these overclock are (by % gain) much more interesting. On old Riva TNT2 M64 16MB is default clocks 125/100MHz (core/rams). Now I tried on this Riva, that wish just recapped caps to Samxon GC (with Nichicon polymers it would be even better) overclock and get to 167/170Mhz:


That is:
125 -> 167
100 -> 170

So a 34% overclock for core and whooping 70% more for rams. Show me hardware, that can do such overclock todays :)

...and this is where the overclocking just start. Moving sliders up is not overclocking - the reall challenge is now to look at the card, identify voltage regulators and manage to increase the output voltage from the for the core and rams. Unfortunately, for the rams, are used the 3.3V from mainboard, so, no regulator and therefore no way to modify this, w/o completely cutting the power to the rams and providing own regulator from 5V rail to the rams.
The core, on the other hand, use a 2.65-2.66V from onboard regulator, that can be set by changing the components (usually resistors) to give higher voltages, like 2.80V or 2.90V... :) Now that will be overclocking!
__________________
"It is dangerous to be right in matters on which the established authorities are wrong." - Voltaire
...just keep folding, just keep folding... :) my config
Reply With Quote
  #8 (permalink)  
Old October 13, 2013, 06:17 AM
Bond007's Avatar
Hall Of Fame
 
Join Date: Jun 2009
Location: Nova Scotia
Posts: 3,146

My System Specs

Default

I would say the score difference is a combination of CPU and GPU mem/core speeds all being lower on your setup. DDR/ddr2/ddr3/ddr4 you multiply actual frequency by 2 to get effective rate. Gddr5 you multiply by 4 ( as you have figured out).
__________________
Desktop: Antec Three Hundred, Z77-D3H, 3570K undervolt @4.1ghz with Zalman Optima Cooler, Corsair Vengeance 2x4gb@1866mhz, HD 6950(unlocked shaders and OC'd), Bitfenix Fury 750w, WD Black 1TB, Intel 530 120GB
Away from home light Gamer: Acer Aspire 15.6", A8-3500m undervolt and OC to 2Ghz, 2x4gb Corsair Vengeance, Crucial M500 240gb
Reply With Quote
  #9 (permalink)  
Old October 13, 2013, 08:01 PM
MVP
 
Join Date: Oct 2012
Location: North East Alberta
Posts: 447

My System Specs

Default

Interesting discussion.

Overclocking seems to be a hobby onto itself, complete with contests. I have watched with interest the overclocking of the AMD A10 APU - seems that chip is good for over clocking.

RE: Temperatures - while I did not run all fans at 100% all the time I did have my fan controllers set so that as temperatures approach 60 they go to 100%. I also noticed that the temperatures just barely made it to 60 degrees C during the testing. Tried test with temperature monitoring on and off, didn't seem to affect the testing.

RE: voltage ~ I need to read a little more about adjusting voltages. I did push my APU voltage up just a hair but not by much. (not at home right now to check the actual amount - away for Thanksgiving weekend). I did not adjust the video card voltage. I'll have to see what I can learn about tampering with these voltage settings before tinkering ~ any advice or references to check?
__________________
________________________________

FKD
Fort Kent, Alberta

2012 - A10 Rig
Reply With Quote
Reply


Thread Tools
Display Modes

Similar Threads
Thread Thread Starter Forum Replies Last Post
New build, graphic card and memory management Paulo098765 New Builds 10 June 23, 2013 03:52 AM
Understanding RAM Memory Timing Data Malam RAM 9 November 26, 2012 02:50 PM
6670 Memory clock problems hiller741 Video Cards 9 September 26, 2012 05:59 PM
GPU Memory Clock??? BFighter Overclocking, Tweaking and Benchmarking 1 May 12, 2011 07:35 PM
What gets you more FPS, Higher GPU Core Clock or Memory bushwickbill Gaming 7 May 22, 2009 06:33 AM