How much quicker is QPi over FSB

[iamrogue]

Is there a short answer to my question? How much quicker is QPi/HT over FSB? I mean, lets say we have a 2.1 GHz CPU Dual Core, one of them is QPi and the other FSB, how much quicker would the QPi be? Also, would be awesome if you could show me the "formula" or whatever.

 

[Lmeow]

To be honest, it's pretty hard to compare, I don't think there's a real formula for defining how fast QPI is over FSB. There are no QPI based dual cores.

 

[johnnyq8]

its complicated question some cpu's have something called QPI the most of this cpu's is the quad-core the qpi is the main link between the cpu and the ram and the gpu when you overclock your cpu the qpi will be higher that means better performance and i dont know much about FSB but this is regular things you dont need to compare

 

[unknown_13]

QPI is being use on one architecture (Nehalem aka i7-9xx), FSB is being used on a different architecture (LGA775 aka Pentium Duals, Core 2's etc.)

So you can't really compare those two things. QPI is faster, but how much faster (exactly how much), i don't know. There aren't exact numbers provided by Intel, at least i haven't seen those "numbers".

 

[johnnyq8]

QPI is being use on one architecture (Nehalem aka i7-9xx), FSB is being used on a different architecture (LGA775 aka Pentium Duals, Core 2's etc.)

So you can't really compare those two things. QPI is faster, but how much faster (exactly how much), i don't know. There aren't exact numbers provided by Intel, at least i haven't seen those "numbers".

check intel website for i7 9's processors and you will find it have the QPI speed is 4.8 GT/S but the i7 8's is 2.5 GT/S so the 9 series is better and will use the HT technology efficeintly but the extreme editions of the i7 they use the full speed of the qpi link wich its 6.4 GT/S thats the highest qpi link

 

[unknown_13]

check intel website for i7 9's processors and you will find it have the QPI speed is 4.8 GT/S but the i7 8's is 2.5 GT/S so the 9 series is better and will use the HT technology efficeintly but the extreme editions of the i7 they use the full speed of the qpi link wich its 6.4 GT/S thats the highest qpi link

Aren't the i7-8xx series using the DMI? They're on the P55 platform. And DMI is also slower than QPI but is faster than FSB... so FSB is several times slower than QPI then???

 

[johnnyq8]

Aren't the i7-8xx series using the DMI? They're on the P55 platform. And DMI is also slower than QPI but is faster than FSB... so FSB is several times slower than QPI then???

yeah exactly the 9's is better because the qpi this is the bloomifield archticture the lynifield archticture is DMI so the QPI ( i7/9's) win's

 

[unknown_13]

yeah exactly the 9's is better because the qpi this is the bloomifield archticture the lynifield archticture is DMI so the QPI ( i7/9's) win's

Yes, that's the right answer

So, the answer to the thread is that the FSB is alot slower than the QPI, and slower than DMI.

 

[blackhawk1928]

QPI is much quicker then FSB, from I've and read and understand, in FSB, the cpu and ram communicate over the northbridge while in QPI it skips it and the CPU and ram communicate directly through a link so it elimates a huge amount of latency.

 

[unknown_13]

QPI is much quicker then FSB, from I've and read and understand, in FSB, the cpu and ram communicate over the northbridge while in QPI it skips it and the CPU and ram communicate directly through a link so it elimates a huge amount of latency.

Yeah i forgot that

 

[rohn_avni]

yes in QPI cpu directly communicate with the RAM.... so it's faster... maybe twice than FSB...

 

[Scotteq]

It depends on how you look at it. In absolute terms, the standard QPI buss is equal in total/theoretical bandwidth with the fastest (1600) FSB. The difference is that FSB uses the same lanes for traffic in both directions (these can NOT both be active at the same instant), whereas QPI has dedicated lanes for each. So given the artificial situation where we have perfect utilization in both directions, with all lanes fully saturated, at all times, the slowest implementation of QPI is theoretically twice as fast as the fastest (*1600) FSB. That never happens in real life, though.

QPI also has an advantage of higher clock speed and lower latency (so in some few artificially created cases it could theoretically 'feel' more responsive to the end user). There is also the indirect architectural advantage of having the L3 cache on the processor on a QPI equipped system. With a FSB, communication between Processor cores has to be routed through the FSB. In a (Nehalem) that is handled by the *much* faster L3, and any bandwidth used in this manner never hits the QPI.

http://www.hardwaresecrets.com/article/610


AMD's Hypertransport has the same architetural advantages, but less overall bandwidth.

"In Practice", none of this really matters on a desktop computer. On the Server side, it may or may not matter - depending on the type of workload.

 

[Scotteq]

Also, would be awesome if you could show me the "formula" or whatever.

Example at a 3.2GHz QPI speed (THIS IS DIFFERENT FROM THE PROCESSOR SPEED, SO DON'T CONFUSE THE TWO)

3.2 GHz
× 2 bits/Hz (double data rate)
× 20 (QPI link width)
× (64/80) (data bits/flit bits)
× 2 (two links to achieve bidirectionality)
÷ 8 (bits/byte)
= 25.6 GB/s