Is BClock just a reference clock?

[OC-Chronic]

I recently ired a bunch of know it all posters by arguing the BClock is mroe than a reference & they finally conviced me untiol I started playing around with my i7 960 & DDR3 2000 & noticed a trend. A higher Bclock results in lower RAM latency, all other things being equal, including uncore speeds. If it's just a reference clock, this wouldn't be a logical result. There has to be a reason that BClock has an effect on this. Why?

 

[iconiK]

I recently ired a bunch of know it all posters by arguing the BClock is mroe than a reference & they finally conviced me untiol I started playing around with my i7 960 & DDR3 2000 & noticed a trend. A higher Bclock results in lower RAM latency, all other things being equal, including uncore speeds. If it's just a reference clock, this wouldn't be a logical result. There has to be a reason that BClock has an effect on this. Why?

Because all clocks are a multiple of the base clock. That's why it's CALLED a "base clock". Increase it, you increase all other clocks. This is exactly why you can't increase it on Sandy Bridge, because the ratio of ALL the clocks, including SATA, PCIe, USB, etc. is fixed. Increase 100 to 120 you've just lost SATA and maybe PCIe. Definitely USB.

It results in lower RAM latency because the memory controller operates the same as everything else, so it gets clocked proportionally higher. Does it help with performance? No, not really. In fact, barely at all.

 

[OC-Chronic]

Because all clocks are a multiple of the base clock. That's why it's CALLED a "base clock". Increase it, you increase all other clocks. This is exactly why you can't increase it on Sandy Bridge, because the ratio of ALL the clocks, including SATA, PCIe, USB, etc. is fixed. Increase 100 to 120 you've just lost SATA and maybe PCIe. Definitely USB.

It results in lower RAM latency because the memory controller operates the same as everything else, so it gets clocked proportionally higher. Does it help with performance? No, not really. In fact, barely at all.

I'm not talking about raisng all clocks, just the BClock...If I keep all other things equal, like Core, Uncore, QPI & RAM, I get better latency with a high BClock & lower multis vs a lower Bclock & higher multis. If I increase Uncore frequency, QPI or Core, I don't get as much of a latency savings. Even RAM speeds don't have as much effect as BClock on latency. That's why you don't break the 40ns mark unless you go 200+ Bclock. If the Bclock was jsut a reference clock, this wouldn't make any difference& uncore frequency would have the most effect of all. It's got to be some sort of bus clock. Check it out for yourselvse if you don't believe me but BClock has the lartgest effect, by far, on memory latency. Again, why? Because it is a bus clock. It has to be....

 

[sirmonkey1985]

raising the bclk raises everything with it including the ram.. easy way of explaining how bclk works, look at AMD's overclocking with locked processors. both are identical. so as you raise the bclk you raise the NB, ram clock and QPI along with it. thus your latency time goes down because the NB is now at a higher clock speed.

intel's Bclk is one in the same as AMD's HT ref. clock. they both do the exact same thing.. where Intel changed it is with sandy bridge due to the onboard IGP which is also tied into the Bclk and why you can't overclock using the Bclk option.

 

[OC-Chronic]

If you check out Clunk's review of the UT x58, he has a Bclock in the 200 range but his core, uncore, mem * QPI are not that high but the memory latency is low 30's...for example this link shows his benchmarks & OCCT time....

http://www.clunk.org.uk/images/reviews/dfi-ut-x58-t3eh8/results/4ghz-1600mhz-oct-evst-wpr-spi.png

If you'll notice it's a modest oc with:

200-BClock
Core @ 4Ghz
Uncore @ 3.2Ghz
QPI @ 3.6Ghz
MEM @ 800(DDR 1600) 8,8,8,24,74,1T

Yet, his memory latency is 30.6 ns...

 

[OC-Chronic]

raising the bclk raises everything with it including the ram.. easy way of explaining how bclk works, look at AMD's overclocking with locked processors. both are identical. so as you raise the bclk you raise the NB, ram clock and QPI along with it. thus your latency time goes down because the NB is now at a higher clock speed.

intel's Bclk is one in the same as AMD's HT ref. clock. they both do the exact same thing.. where Intel changed it is with sandy bridge due to the onboard IGP which is also tied into the Bclk and why you can't overclock using the Bclk option.

Obviously, you didn't read the whole thread before responding...

I stated that I used lower multi's to keept the Core, MEM, Uncore & QPI at the same speeds, while increasing ONLY BClock....yet I get lower memory latency...your explanation doesn't fit...

This has to be mroe than just a reference clock...It has to be a dataclock of some sort or it wouldn't have this effect on latency...

 

[iconiK]

This has to be mroe than just a reference clock...It has to be a dataclock of some sort or it wouldn't have this effect on latency...

The 1600 MHz CL8 memory is what gives you the latency. Higher CPU clocks might also help get a more accurate memory latency measurement. For the fuckin' billionth time, base clock is only a configurable clock generation, which sends a stupid signal every so many times a second (default 133 million times a second for 1366).

If increasing base clock gives you lower latency, everything else being set manually to the same, then the motherboard must hide something that gets proprotionally increased with the base clock and affects memory latency. I can't think of anything right now, I'm no Nehalem architecture expert. You might want to ask IntelEnthusiast, send him a PM.

 

[OC-Chronic]

The 1600 MHz CL8 memory is what gives you the latency. Higher CPU clocks might also help get a more accurate memory latency measurement. For the fuckin' billionth time, base clock is only a configurable clock generation, which sends a stupid signal every so many times a second (default 133 million times a second for 1366).

If increasing base clock gives you lower latency, everything else being set manually to the same, then the motherboard must hide something that gets proprotionally increased with the base clock and affects memory latency. I can't think of anything right now, I'm no Nehalem architecture expert. You might want to ask IntelEnthusiast, send him a PM.

I admit, I am missing something here...I can't figure out how this guy is getting such low latency...

 

[cyclone3d]

I've done quite a bit of testing, and i am pretty sure it has to do with the QPI speed when you raise the base clock.

Also, you can tighten up the sub-timings on your RAM if your BIOS supports it. The sub timings can bring the latency down quite a bit.

 

[OC-Chronic]

I've done quite a bit of testing, and i am pretty sure it has to do with the QPI speed when you raise the base clock.

Also, you can tighten up the sub-timings on your RAM if your BIOS supports it. The sub timings can bring the latency down quite a bit.

I think you are on to something but It's not the QPI. I did some more experimienting to see what affects latency & the Uncore frequency seems to make the most difference, which makes sense.

Another possibility that occurred to me:

My i7 960, I believe, has more memory multi's than the 920's that I see reviewed everywhere. If I am correct, then wouldn't it makes sense that the internal latency of the memory controller would be increased to accomodate the higher dividers? My chip allows a 12 multi on the memory(1600) & the 920's have a 10x multi limit(1333). No?

If this is true, & I'm pretty sure it is, then the 920 would have tighter internal timings between the mem controller & the RAM which greatly affects the memory latency.

 

[Tsumi]

I think you are on to something but It's not the QPI. I did some more experimienting to see what affects latency & the Uncore frequency seems to make the most difference, which makes sense.

Another possibility that occurred to me:

My i7 960, I believe, has more memory multi's than the 920's that I see reviewed everywhere. If I am correct, then wouldn't it makes sense that the internal latency of the memory controller would be increased to accomodate the higher dividers? My chip allows a 12 multi on the memory(1600) & the 920's have a 10x multi limit(1333). No?

If this is true, & I'm pretty sure it is, then the 920 would have tighter internal timings between the mem controller & the RAM which greatly affects the memory latency.

I don't know what memory dividers you're talking about, I think it depends on the motherboard. For my motherboard (MSI Big Bang Xpower), at 200 BCLK I get 1600 mhz with a multiplier of 4. At 133 BCLK, I get 1600 mhz with a multiplier of 6 I believe. And my memory multipliers go up to 10 or 12. So that's definitely not the issue here. Unless it's a motherboard-specific issue.

 

[iconiK]

I admit, I am missing something here...I can't figure out how this guy is getting such low latency...

There's some settings in the BIOS he tweaked for sure, but I don't know which. Probably not directly related to the base clock.

 

[OC-Chronic]

I don't know what memory dividers you're talking about, I think it depends on the motherboard. For my motherboard (MSI Big Bang Xpower), at 200 BCLK I get 1600 mhz with a multiplier of 4. At 133 BCLK, I get 1600 mhz with a multiplier of 6 I believe. And my memory multipliers go up to 10 or 12. So that's definitely not the issue here. Unless it's a motherboard-specific issue.

My 960 has a max Ucore multi of 24 which would allow a 3200 Uncore @ 133 BClock & 1600 RAM(2:1 for stability) What's the highest uncore multi on your chip & what i7 are you using for the above 1600 RAM?