Cache Frequency vs Processor Frecuency

[TechDookie]

So, I started overclocking my I5 4690K and I got the Processor to run at 3.9 gigahertz at 70 degrees on 1.15 v while on stock cooler. But, I was wondering what is the difference between the Cache Frequency and the Processor Frequency? 

[TechDookie]

Bumpity

[Ryoutarou97]

Bumpity

Don't bump. BAAAAAD. I'm not sure how they affect performance, but cache is like hyper fast RAM that's directly on the CPU since the bus frequency is only 100mhz while the CPU runs at 30-50 times that. That means if the CPU needed data it would have to wait 40ish cycle after requesting it. instead, all of this is loaded into cache which is much faster. The core frequency is the speed of the CPU which I suspect is the part that's going to influence performance. Cache frequency is, well, cache frequency.

[i_build_nanosuits]

So, I started overclocking my I5 4690K and I got the Processor to run at 3.9 gigahertz at 70 degrees on 1.15 v while on stock cooler. But, I was wondering what is the difference between the Cache Frequency and the Processor Frequency?

Your processor has onboard a few megabytes of very fast storage call ''cache''...that's a pool where the applications will store instructions for the CPU to process...it's like memory, but very fast memory onboard the CPU die itself...increasing the speed of the cache can lead to instability and other issues, it will not impact performance any noticeable amount...i would suggest leaving the cache ratio on auto or manualy set it to say..38...

[trag1c]

Your processor has onboard a few megabytes of very fast storage call ''cache''...that's a pool where the applications will store instructions for the CPU to process...it's like memory, but very fast memory onboard the CPU die itself...increasing the speed of the cache can lead to instability and other issues, it will not impact performance any noticeable amount...i would suggest leaving the cache ratio on auto or manualy set it to say..38...

It also stores frequently used data as well as well as data that is queued to be to processed.

[TechDookie]

It also stores frequently used data as well as well as data that is queued to be to processed.

So, its basically faster RAM? Well, in that case I guess I won't be overclocking it.

[trag1c]

So, its basically faster RAM? Well, in that case I guess I won't be overclocking it.

Yeah Cache is crazy f****** fast. it makes your ram look like a hard disk, especially L1 cache.

[Prysin]

L1 cache is like autofill function in browsers (where the browser automatically fills out the adress you are looking for), except L1 cache is running on both crack and steroids at the same time.

[harrynowl]

Cache or uncore (uncore refers to things on the CPU die that aren't the CPU- as a lot of things that used to be separate ICs are now included on the CPU package) is basically the speed of the integrated northbridge, it will generally impact the speed of your memory operations when you increase it.

 

The frequency to prioritise is CPU frequency, cache frequency can require quite a bit of fine tuning to get right. I personally have just left my cache at 40x (up from 32x) as it runs at that frequency without changing anything else just fine.

[Lays]

So, its basically faster RAM? Well, in that case I guess I won't be overclocking it.

 

 

It does help with video editing and things like that, it won't help in games though I don't believe.

 

If you're benchmarking and going for high scores, you definitely should OC the cache for that extra 2-5% performance bump.

[i_build_nanosuits]

you definitely should OC the cache for that extra 2-5% performance bump.

2-5%? my cinebench runs output exactly the same results +/-15points (this varies all the time it's normal you never get the same score exactly) wheter my CPU ring is at 32 or 44...so not even 1%...not even 0.5%.

[Lays]

2-5%? my cinebench runs output exactly the same results +/-15points (this varies all the time it's normal you never get the same score exactly) wheter my CPU ring is at 32 or 44...so not even 1%...not even 0.5%.

Weird, are you closing every program before running it?

I gain lots of points in xtu benchmark, and a quite a few in cinebench r15 as well.

I'm competing with people that are at 5 ghz when I'm at 4.9 because I ran 4.8 ghz cache as well.

[i_build_nanosuits]

Weird, are you closing every program before running it?

yes, all tests i run in the same conditions...fresh reboot...wait 1 minute...run test 3 times in a row...do the average...average is like +10 when pump the ring bus to 44...

[Lays]

yes, all tests i run in the same conditions...fresh reboot...wait 1 minute...run test 3 times in a row...do the average...average is like +10 when pump the ring bus to 44...

Any added points are good in my book, +10 or +20 are good for me, sometimes makes the difference on getting you higher on that leader boards than other people with higher core clocks.

It doesn't seem to help much in real world scenarios other than long video editing projects though.

[i_build_nanosuits]

Any added points are good in my book, +10 or +20 are good for me

yes...

[Lays]

yes...but as i was pointing out, at 4.4ghz core clock my score is 870...i can go up to 885 or so if i clock the crap out of the cache ratio...that's a difference of 0.1724%...nowhere near the 2 to 5% you are claiming...that's all

 

To get a 5% improvement you need like around a 250mhz extra on the core clock...

 

That's only in one benchmark though.  In XTU for instance, you will definitely get closer to that 5% margin.

 

Also 4.4 ghz on the cache is still quite a medium OC, you should be able to get closer to 4.5 or 4.6, 4.8 if you have a good chip.

[i_build_nanosuits]

That's only in one benchmark though.  In XTU for instance, you will definitely get closer to that 5% margin.

 

Also 4.4 ghz on the cache is still quite a medium OC, you should be able to get closer to 4.5 or 4.6, 4.8 if you have a good chip.

i messed up the math...it's 1.724%....so your 2% guesstimate wasn't too far off...but 5% is out of question.

it might be different in other tests though...i don't know.

 

...and i don't have a good chip lol

 

My daily speed on this CPU (sweespot) is 4.2ghz @1.185v and cache ratio (ring bus) is 4.0ghz at 1.14v

i can push it to 4.3 with a little less than 1.3v but to me it's not worth the extra heat and power consumption...so i settled at 4.2ghz...it's plenty for my needs anyways.

[TechDookie]

Cache or uncore (uncore refers to things on the CPU die that aren't the CPU- as a lot of things that used to be separate ICs are now included on the CPU package) is basically the speed of the integrated northbridge, it will generally impact the speed of your memory operations when you increase it.

 

The frequency to prioritise is CPU frequency, cache frequency can require quite a bit of fine tuning to get right. I personally have just left my cache at 40x (up from 32x) as it runs at that frequency without changing anything else just fine.

So whats your CPU frequency?

 

From what I've read, RAM doesn't do much to increase performance especially when your running on a graphics card. So, I'm guessing this would be similar... 

 

I think what I'll do is overclock the cache frequency a bit so it isn't squashed by the CPU frequency. I don't want the CPU to not be getting the cached memory fast enough (like a mini bottleneck).

[Lays]

i messed up the math...it's 1.724%....so your 2% guesstimate wasn't too far off...but 5% is out of question.

it might be different in other tests too...i don't know.

 

...and i don't have a good chip lol

 

Have you tested cache beyond 4.4? If you change ring voltage to 1.325, VCCIN to 2 or 2.05, you should be able to get in the neighborhood of 4.5 or 4.6 maybe 4.7.

 

XTU's built in benchmark really likes cache overclocks, it should be closer to the 5% increase mark.

 

I think it takes me roughly 2.075 vccin, 1.5-ish v core, and 1.415ring voltage to bench stable at 4.9 ghz core, 4.8 ghz cache with ram at 2600 CL9 timings 9-11-11-22 1T

Got 1005 CB score from that, people with 5 ghz cpu and lower cache overclocks are getting around 1000-1015, so I'm already right next to people with 100 mhz core over me just because of super high cache, also the ram helps a few points as well.  At 5 ghz I would get around 1040-ish I believe, but I need to delid and get a bigger radiator + ice bucket, or dry ice.

 

I really really need to delid though, because my chip gets ridiculously hot after like 1.3v, and my 4790k could do 1.4-1.45v at same temps.  To bench that 4.9 cinebench I did the other day I needed an ice bucket to put my radiator in lol.

 

[harrynowl]

So whats your CPU frequency?

 

From what I've read, RAM doesn't do much to increase performance especially when your running on a graphics card. So, I'm guessing this would be similar... 

 

I think what I'll do is overclock the cache frequency a bit so it isn't squashed by the CPU frequency. I don't want the CPU to not be getting the cached memory fast enough (like a mini bottleneck).

I am running 4.5Ghz CPU.

 

I'd overclock it but don't overclock it as aggressively, as the performance gains are pretty minimal.

[i_build_nanosuits]

Have you tested cache beyond 4.4? If you change ring voltage to 1.325, VCCIN to 2 or 2.05, you should be able to get in the neighborhood of 4.5 or 4.6 maybe 4.7.

 

I think it takes me roughly 2.075 vccin, 1.5-ish v core, and 1.415ring voltage to bench stable at 4.9 ghz core, 4.8 ghz cache with ram at 2600 CL9 timings 9-11-11-22 1T

Well..the thing is i personaly care more about quiet operations, low temps and energy efficiency...

So for example, my CPU is now running 100% stable at 4.2ghz with only 1.185v on the Vcore, 1.65v on the VCCIN and 1.135v on the ring bus...so overall very low voltages...i plan on using that chip a couple more years so i like it running cool...right now with that 240mm AIO i got on it, the CPU never reach even 60c (mostly around 55c) while gaming...and around 65c under prolonged full load...so overall very happy with those settings...i might crank it up when i will have a more powerful graphics solution and the CPU will start showing it's age when it comes to running the latest AAA titles...i know i can get it stable at 4.4ghz with around 1.37v...at those settings it gets up to around 80c+ under full load...

[Lays]

Well..the thing is i personaly care more about quiet operations, low temps and energy efficiency...

So for example, my CPU is now running 100% stable at 4.2ghz with only 1.185v on the Vcore, 1.65v on the VCCIN and 1.135v on the ring bus...so overall very low voltages...i plan on using that chip a couple more years so i like it running cool...right now with that 240mm AIO i got on it, the CPU never reach even 60c (mostly around 55c) while gaming...and around 65c under prolonged full load...so overall very happy with those settings...i might crank it up when i will have a more powerful graphics solution and the CPU will start showing it's age when it comes to running the latest AAA titles...i know i can get it stable at 4.4ghz with around 1.37v...at those settings it gets up to around 80c+ under full load...

 

 

I'm not talking about 24/7 clocks at all btw.

 

I'm talking about benchmarking for short periods of time with higher voltages.  I run lower voltages 24/7 as well because I like quiet operation as well.  I usually run closer to 1.2v for 24/7 use with 1.8vccin 1.15 ring bus.

 

This nets me ~4.4 ghz core + 4.2 ghz cache.  I haven't done anything for stress testing yet, I will need to go and do that sometime this week when I'm not busy benchmarking trying to get highscores on HWBOT