Overclocking getting a bit overwhelming these days.

Man of Honour
Joined
25 Oct 2002
Posts
31,869
Location
Hampshire
A bit of background, I started getting into overclocking in the late 90s, around the time it moved away from being done via physical jumpers and some motherboards started to offer some features like being able to set bus speed in BIOS. In those days, you had very few options to choose from, overclocking basically meant:

1) Research what CPU and mobo to buy
2) Get some overspecced RAM that could run at a higher bus speed than the default bus speed for your chip (FSB/RAM were linked)
3) [Optional] Invest in a decent cpu cooler
4) Increase the bus speed (in set increments like 66 / 75 / 83 / 100 / 105 etc), perhaps adding a bit of VCoreif your motherboard allowed (not all did), and maybe increasing CAS latency (just the one setting) to allow the RAM to go higher.

So after an ill-fated experiment on my old Cyrix CPU (which used to run baking hot even at stock) I cut my teeth on legendary combos like Celeron 300A, Abit BH6, 64MB PC100 RAM, crank it up to 450mhz+ and you had something that matched the top of the line CPU of the time, the P2-450.

Broadly speaking, whilst things evolved a bit with more options coming in, maybe some extra voltage or whatever, this patter stayed largely the same through the 00s. There were of course some stuff like golden fingers devices, pencil tricks etc for certain CPUs but by and large the BIOS options were just an evolution not revolution.

Now, what I'm finding these days on the Ryzen platform, is there is hundreds of things to keep track of. Multiple voltages, numerous RAM timings, IF Clock, memory clock, PBO, Auto OC, elaborate fan profiles, offsets, LLC etc etc. And CPUs seem to automatically overclock themselves with turbo modes, voltage ramping up or whatever, so it's really difficult to know how much I'm gaining / what the best route to go is. In the old days I'd basically upgrade and have a very clear 40%+ improvement over stock based on frequency vs stock.

On the one hand, back in the 90s I probably would've craved a bit more flexibility to tune overclocks, bigger range of bus speeds, more voltage headroom etc. So I shouldn't be too critical of giving that flexibility. But I must confess, part of me longs for the simpler time, when overclocking was really quite simple, you bought specific components like CPUs on lower speeds along with the relevant mobo/ram/cooler to be able to crank them up, that was about it. Nowadays, it seems a bottomless pit of possibilities, an endless cycle of tweaking until you muck something up and have to reset the CMOS.

Of course, just because the functionality doesn't mean you have to use it, but like many overclockers the OCD starts to kick in and you just can't help yourself, or if you just go for a simple fire and forget OC, you are then sitting there thinking you could get more out of it.
 
AMD CPUs come pretty much maxed out the box so any extra gains from overclocking are quite small, overclocking these days is something to do for fun rather than the big performance gains of old.
 
True enough HangTime. The last CPU I had that I could get a really big improvement from and without too much complexity was an Intel Core 2 Quad Q6600 - the clock speed was 2.4 GHz, mine would go up to 3.6 with only a little extra voltage, a nice 50% boost.

These days a manual overclock will still likely give you an improvement on the all-core clock speed and faster results in heavily multithreaded tasks, but will likely reduce the single core performance.

I think it stems from the manufacturers trying to eke out more performance by pushing them closer to their limits as standard, to compensate for the lack of gains they used to be able to get from regular manufacturing process improvements. As we get closer and closer to the limits of how small transistors can be made, the improvements are becoming smaller and harder to achieve.

Right now I haven't even got a manual overclock on my current CPU (a Core i9 9900K) as I could only improve on the standard boost clock by 200 Mhz - which is almost nothing on a CPU with a 4.9 GHz boost clock to start with. Throwing a lot of cooling power at them to make sure they can boost without hitting thermal limits is about all it's worth doing.
 
AMD CPUs come pretty much maxed out the box so any extra gains from overclocking are quite small, overclocking these days is something to do for fun rather than the big performance gains of old.

It comes down to automation software.

In past years, there was no way to bin CPU's without putting them in a PC and overclocking them - because of this, Intel and AMD has to be super conservative with clocks to avoid RMA issues, so CPU's were shipped with a clock speed that guarantees minimum RMA and it was up to the user to find out how far that CPU could actually go.

Fast forward to now and all CPU come with boost software that figures out what clock the CPU can do before instability kicks in and runs it at that clock speed. Intel and AMD"s boost tech is just an automatic overclocking software - if this software didn't exist, Intel would still be selling 4ghz CPU
 
Yeah the best balance IMO was the Core 2 era.

A lot of settings i.e. sub-timings, sub-system voltages, etc. there is a science to and in most cases used to be automated behind the scenes but the information to calculate the required setting from other settings and that relationship often isn't readily available and usually there is a fairly small range of actual viable settings - now a lot of those settings are exposed to the end user but not really that useful unless you are trying to set a, probably unstable, WR - for a lot of people arbitrary best guesses/trial and error, settings you've seen on Reddit, etc. etc. are probably not properly stable unless you got lucky.
 
Yeah the best balance IMO was the Core 2 era.

A lot of settings i.e. sub-timings, sub-system voltages, etc. there is a science to and in most cases used to be automated behind the scenes but the information to calculate the required setting from other settings and that relationship often isn't readily available and usually there is a fairly small range of actual viable settings - now a lot of those settings are exposed to the end user but not really that useful unless you are trying to set a, probably unstable, WR - for a lot of people arbitrary best guesses/trial and error, settings you've seen on Reddit, etc. etc. are probably not properly stable unless you got lucky.


I went from a stock athlon all the way to an i7 860 back in the day and oh boy I loved that thing - I pushed it 1.3ghz higher than the stock speed, today that's only possible on LN2
 
He's saying you don't need LN2 to get a 1.3ghz overclock :)
My dad's old dell that's been replaced actually has a i7-860 in it, I might harvest it sometime to swap with an i5-750 as I have a 1156 board that can overclock.

So I think you guys have explained it quite well, basically the issue I have is that in order to keep making gains the manufacturers have had to start putting in auto-overclocking features and hence a reduced number of high performing cpus just put in a lower speed bin for marketing purposes. In the old days there were loads of slower cpus with headroom which made end-user overclocking pretty easy but not so much these days. Don't know why I've never really thought about it this way before.

I'm seeing similar things on the GPU side, it's now all about 'boost clocks' and somewhat counter-intuitive approaches of undervolting to try and maintain higher boosts due to reduced temps etc, rather than a case of ramp up the voltage and clocks as high as you dare. Again I get a bit confused by it all, it's getting to the stage now where I might not bother with overclocking any more, maybe in future I should consider buying prebuilt systems (have always built my own due to the desire to overclock).

In a way it is ironic; for years we moaned about attempts to clamp down on overclocking (fixed multipliers, voltage limits etc) and now they have embraced it, it's actually got more difficult to eek out extra performance.
 
In a way it is ironic; for years we moaned about attempts to clamp down on overclocking (fixed multipliers, voltage limits etc) and now they have embraced it, it's actually got more difficult to eek out extra performance.

The clamping down still exists, to cause the lose what we were used to, take that Celeron 300A a CPU at 1/8 the price of the PII-450 and with a few hardware settings or changes in the BIOS you have yourself a PII-450 but with integrated L2 cache.
Intel stopped letting you change the multi's so you can't just pop out and buy a 10400f and get the same performance as the 10600K even if you wanted to, at they stop you doing it, even though most of the CPU's could probably hit 5GHz+.

It was about saving money and getting (almost) the best performance, now they just make SKU's that are made for OCing taking the entire point way in the first place.
 
He's saying you don't need LN2 to get a 1.3ghz overclock :)
My dad's old dell that's been replaced actually has a i7-860 in it, I might harvest it sometime to swap with an i5-750 as I have a 1156 board that can overclock.

So I think you guys have explained it quite well, basically the issue I have is that in order to keep making gains the manufacturers have had to start putting in auto-overclocking features and hence a reduced number of high performing cpus just put in a lower speed bin for marketing purposes. In the old days there were loads of slower cpus with headroom which made end-user overclocking pretty easy but not so much these days. Don't know why I've never really thought about it this way before.

I'm seeing similar things on the GPU side, it's now all about 'boost clocks' and somewhat counter-intuitive approaches of undervolting to try and maintain higher boosts due to reduced temps etc, rather than a case of ramp up the voltage and clocks as high as you dare. Again I get a bit confused by it all, it's getting to the stage now where I might not bother with overclocking any more, maybe in future I should consider buying prebuilt systems (have always built my own due to the desire to overclock).

In a way it is ironic; for years we moaned about attempts to clamp down on overclocking (fixed multipliers, voltage limits etc) and now they have embraced it, it's actually got more difficult to eek out extra performance.
CPU/GPU manufacturers don't want to give people that extra performance headroom atleast not for free which is why everything is so tightly locked down these days. If you want an extra 15% boost to the GPU then you have to buy the next tier up, if you want your CPU to hit 4600mhz instead of 4400mhz then you have to buy the 3800XT instead of the 3700X etc.
 
overwhelming? easiest it has ever been.

You used to have the northbridge and FSB to northrbridge ratios right?

some RAM sticks you actually had to think about compatibility

Some mobos had v-core droop issues (cough Asus cough)

Now even the auto OC functions do a fairly decent job!
 
There's vastly more options than there used to be. I mean you go in RAM timings and it's pages of the stuff, used to literally just be CAS latency that's it.
It's easier now to do a 'simple' overclock but you're then sat there wondering how to tweak a bit more - but instead of half a dozen variables to play with, now there's dozens of them.
 
There's vastly more options than there used to be. I mean you go in RAM timings and it's pages of the stuff, used to literally just be CAS latency that's it.
It's easier now to do a 'simple' overclock but you're then sat there wondering how to tweak a bit more - but instead of half a dozen variables to play with, now there's dozens of them.

Sometimes the mobo has predone settings to overclock. Usually the very top end gaming mobos.
 
Back
Top Bottom