Sign in with
Sign up | Sign in
Your question
Closed

Core i7 bus aka quickpath

Last response: in CPUs
Share
August 28, 2009 10:00:53 PM

i know most of you are noobs, but im sure there is at least one that knows the answer for this qustions.

ok busses, for intel core 2 extreme

Processor Family FSB
QX9775 4 3.20 GHz 1600 MHz 12 MB 45nm
QX9770 4 3.20 GHz 1600 MHz 12 MB 45nm
QX9650 4 3.00 GHz 1333 MHz 12 MB 45nm
QX6850 4 3.00 GHz 1333 MHz 8 MB 65nm
QX6800 4 2.93 GHz 1066 MHz 8 MB 65nm
QX6700 4 2.66 GHz 1066 MHz 8 MB 65nm
X6800 2 2.93 GHz 1066 MHz 4 MB 65nm

i do understand what 1600 MHz mean(FSB), now what i wanna know is this

Processor Family FSB or intel quick path
i7-950 3.06 GHz 4.8 GT/sec 8 MB 45 nm
i7-940 2.93 GHz 4.8 GT/sec 8 MB 45 nm
i7-920 2.66 GHz 4.8 GT/sec 8 MB 45 nm

whats this FSB 4.8GT/sec for core i7, how much is that in MHz, or better yet what does that mean.

More about : core bus aka quickpath

August 28, 2009 10:09:36 PM

4.8 GT/s is data bits transferred at a rate of roughly 4.8 GHz. There's some technical discussion around the phrase "Giga-transfers per second" vs "Gigahertz per second" which I'm not going to get into here.

What's not being listed is the number of lanes of 4.8GHz traffic.

But the FSB and QPI data rates are not directly comparable.
August 28, 2009 10:15:14 PM

which im not going to get into there??? why not get in there and explain please.
Related resources
August 28, 2009 10:18:37 PM

Because it's largely a semantic argument between engineers who think "gigahertz" should be reserved for discussing clock speeds and other engineers who think "gigahertz" is a generic term for "rate".
August 28, 2009 10:27:54 PM

It's not 4.8 ghz it's 2.4 ghz. And the 6.4 gt/sec is 3.2 ghz. Take the gt/sec and divide by two and that's your frequency. It's how fast the quick path interconnect moves data, (i think from the processor to the ram, but I don't know for sure). Right now mine is 7.2 gt/sec, but there isn't a performance difference between that and 4.2 because ram can't handle quick data very well yet (see memory scaling articles).
August 28, 2009 11:02:02 PM

yea the front side bus/ or quick path is how Cpu gets to the Ram, but why would they compare its speed with GT's not MHz like they always used to, but as i can understand that 4.8GT is much faster than 1600 MHz right? and thats more than enough to Full accsess the fastest DDR3 Ram. good to hear.

but if someone could get me the speed with MHz, i'll be thankful.
a b à CPUs
August 28, 2009 11:14:18 PM

Actually, it's irrelevant for RAM speed anyways. The memory controller is on a completely separate pathway on i7, and will always allow for full speed access.
August 29, 2009 12:56:22 AM

what u mean its on a completely separate pathway on i7? isnt the quick path aka the front side bus for the memory, if not please explain so i can understand how this works.
a b à CPUs
August 29, 2009 3:03:05 AM

surda said:
what u mean its on a completely separate pathway on i7?
Think of it like a freeway. The number of cars you can move per second depends not only on the speed limit (ie, the GHz or GT/sec), but also the on number of lanes. Double the number of lanes, and you've doubled the capacity of the freeway.

At the end of it all no matter what technology or protocols are being used, the key thing you're looking for is the throughput of the bus, which is about 25GB/sec for the high-end Core i7 chips. That's 12.5+ GB/sec in two directions.

This Wikipedia article explains it in detail, noting that:

"The rate is computed as follows:

* 3.2GHz
* x 2 bits/Hz (double data rate)
* x 20 (QPI link width) <--- this is the number of "lanes"
* x (64/80) (data bits/flit bits)
* x 2 (bidirectional)
* / 8 (bits/byte)
* = 25.6 GB/s"
a b à CPUs
August 29, 2009 3:31:15 AM

surda said:
what u mean its on a completely separate pathway on i7? isnt the quick path aka the front side bus for the memory, if not please explain so i can understand how this works.

Nope. The QPI is the connection between the processor and the northbridge, which controls the hard drives, the graphics cards, and most other things on the computer. However, the memory is connected to the CPU via the integrated memory controller, which is completely separate from the QPI. Therefore, no matter how much you use the QPI, it won't slow down the memory at all and vice versa.
a b à CPUs
August 29, 2009 3:56:59 AM

archibael said:
4.8 GT/s is data bits transferred at a rate of roughly 4.8 GHz. There's some technical discussion around the phrase "Giga-transfers per second" vs "Gigahertz per second" which I'm not going to get into here.

Considering that Hertz refers to cycles per second, "Gigahertz per second" would be cycles per second per second. QPI accelerates! :bounce: 
August 29, 2009 5:28:53 AM

randomizer said:
Considering that Hertz refers to cycles per second, "Gigahertz per second" would be cycles per second per second. QPI accelerates! :bounce: 

True that.
August 29, 2009 5:44:01 AM

that makes a complete sense.

ok everything is connected to the Motherboard, north and south bridge as shown in the pic.




on the north bridge that's where the CPU, Ram, And graphics card send buses to so they can connect to each other right.

so no matter how fast or slow the CPU Quickpath is, its not gonna effect the speed of the ram bus, right, because its going to send its data to the north bridge to the memory controller and from there, the cpu will send its bus to read the data correct?

so lets say for example 1333 mhz fsb and 1600mhz fsb, how would a 1600 operate faster? by reading the memory faster than the 1333mhz in the north bridge not the ram it self right? the memory is there in the north bridge it just waits for the cpu to read it, so thats how 1600 is faster.

memory controller some have dual channel and some have triple channel, its determined by the motherboard right not the cpu.

am i still on the right side, or am i going off? please let me know

and really thanks alot for your help guys.




a b à CPUs
August 29, 2009 6:13:47 AM

Your image is for LGA775 and older. This is what LGA1366 looks like:

http://images.tweaktown.com/imagebank/ici7l_block.gif

Notice that there is no FSB and the memory has no connection to the North Bridge. The NB is also no longer called the Memory Controller Hub (MCH) since the MC is on the CPU die instead.
August 29, 2009 6:56:35 AM

this is really confusing, i was looking at the other motherboard (pretty much im f'd up)

k whats this loool, i see the memory connect with the cpu, no north bridge. only cpu and pci express that are connected to the north bridge, so what was that north bridge talking about?

so there is no bus between cpu and ram, it uses the quick path insted its like having ram and cpu connected to each other no need for bus speed to effect its transfer data speed.
August 29, 2009 10:35:24 AM

Basically with an "FSB" CPU (pre-i7 or pre-socket_754), everything goes through the FSB, which can reach up to 12.8GB/s on some CPUs, so when the CPU accesses the RAM or northbridge everything squeezes through the FSB

With a QPI(Intel) or HT(AMD) and integrated memory controller, the RAM has it's own dedicated connection of up to 21.3GB/s (AMD) or 38.4GB/s (Intel), PLUS the QPI/HT connection to the northbridge of up to 25.6GB/s (QPI) or 44.8GB/s (HT).

Basically the QPI/HT method can currently be up to about 5 times as fast as the "old" FSB way.
a b à CPUs
August 29, 2009 10:55:44 AM

randomizer said:
Your image is for LGA775 and older. This is what LGA1366 looks like:

http://images.tweaktown.com/imagebank/ici7l_block.gif

Notice that there is no FSB and the memory has no connection to the North Bridge. The NB is also no longer called the Memory Controller Hub (MCH) since the MC is on the CPU die instead.


Well QPI and FSB aren't really cross-comparable.

One is quad pumped the other is deca-linked.
August 29, 2009 6:28:24 PM

randomizer said:
Considering that Hertz refers to cycles per second, "Gigahertz per second" would be cycles per second per second. QPI accelerates! :bounce: 



Doh! Sorry about that!
August 29, 2009 7:38:25 PM

ok now i get everything.

in the old motherboards (not core i7), the memory controller was in the the north bridge (memory controller is where the ram data goes to using dual or triple channel), now for cpu to get there, it had to use its FSB to get to the north bridge and then get the memory controller to accesses the ram data.

but in the new motherboards with the new core i7, it have on-die memory controller that is inside the cpu (core i7)

so there is no longer memory controller in the north bridge, its on the separate way on the core i7 it self, so there is no need to use fsb or intelquickpath to get to the memory controller to accesses the ram, because the ram is sending its data right inside the core i7, and insted of using the quickpath to get ram now its used to get to the north bridge to get the other data not including ram.

please correct me if im wrong.
August 29, 2009 9:58:26 PM

You've got it.
August 29, 2009 11:24:05 PM

thanks to all who helped me with this.

and i need cjl to confirm on this please.

and if there is something els you guys wanna add, please do so i can know more things about this.
a b à CPUs
August 30, 2009 1:52:01 AM

Just as a bit of extra info, AMD uses a similar approach. They don't have an FSB either, and haven't since about 2004 (?).
a b à CPUs
August 30, 2009 4:35:39 AM

surda said:
ok now i get everything.

in the old motherboards (not core i7), the memory controller was in the the north bridge (memory controller is where the ram data goes to using dual or triple channel), now for cpu to get there, it had to use its FSB to get to the north bridge and then get the memory controller to accesses the ram data.

but in the new motherboards with the new core i7, it have on-die memory controller that is inside the cpu (core i7)

so there is no longer memory controller in the north bridge, its on the separate way on the core i7 it self, so there is no need to use fsb or intelquickpath to get to the memory controller to accesses the ram, because the ram is sending its data right inside the core i7, and insted of using the quickpath to get ram now its used to get to the north bridge to get the other data not including ram.

please correct me if im wrong.

Exactly. AMD uses this same approach as well, as randomizer pointed out.
August 31, 2009 7:33:17 AM

thanks, good to know that too

ok i got another question, they replaced the MHz with GT, so i read some pages and it appears that they gonna be using GT for speed insted of MHz from now on.

but i have no idea on how much of MHz=GTs some said its not comparable but can i just an idea on how much it is, or some pics just to get an idea please.

thanks in advance.
a b à CPUs
August 31, 2009 7:54:08 AM

GT/s is slightly more descriptive, but all it really means is gigatransfers (billions of transfers) per second. So, a 6.4GT/s bus makes 6.4 billion transfers per second. To know how fast the bus truly is, you also need to know how much data is in each transfer. The old front side bus ran as fast as 1600 MHz (1600 million transfers per second, or 1.6 billion), but it was 1/2 the speed of a 6.4GT/s QPI rather than 1/4 the speed if I remember right, since it moved more data per transfer.
August 31, 2009 5:48:09 PM

ok i get it.

1600 MHz = 1.6 billion transfers per second

4.8GT/s = 4.8 billion transfers per second.

and also in the old FSB they used QDR(Quad Data pumping), on the rising and falling edges, and at two intermediate points between them, so really it wasnt 1600MHz it was actually 400MHz which = to 400 million Transfers per second.

but Quickpath it transfers on the rising and falling edges, But using 4 bit in each edge

so 4.8GT is 2.4 on each edge

2.4/4 (because it transfers 4bit in each edge) = 600

lets do it backwords

600 X 4 = 2.4 per edge, 2.4 X 2= 4.8GT

thats what i read, not 100% sure though
a b à CPUs
August 31, 2009 6:47:10 PM

surda said:
ok i get it.

1600 MHz = 1.6 billion transfers per second

4.8GT/s = 4.8 billion transfers per second.

and also in the old FSB they used QDR(Quad Data pumping), on the rising and falling edges, and at two intermediate points between them, so really it wasnt 1600MHz it was actually 400MHz which = to 400 million Transfers per second, just like the DDR ram but Quad insted of double.

thanks for your help cjl.

All right except for the QDR part - it's true that it was quad pumped, but that means that it was transferring 4 times per clock cycle at 400MHz. The physical clock signal was 400MHz, but it was still transferring 1600 million times per second. That's why they called it 1600MHz effective.
a b à CPUs
September 6, 2009 1:32:26 PM

randomizer said:
Just as a bit of extra info, AMD uses a similar approach. They don't have an FSB either, and haven't since about 2004 (?).


They still have a clock multi tho.
a b à CPUs
September 6, 2009 9:07:38 PM

amdfangirl said:
They still have a clock multi tho.

That's true with the i7 too.
a b à CPUs
September 8, 2009 10:35:29 AM

I wonder what the next frontier of CPU dev is...

Probably cache related.
October 20, 2009 9:00:09 PM

after the 6 and 8 core i7's - 25nm die's then ≈15nm die's.
February 14, 2010 12:37:43 PM

Hehe actually they're kinda hitting the hardware wall when it comes to silicon... i would say they soon will replace silicon for something else (carbon nano fibers?) which should enable them to go well beyond 15nm. the thing is that silicon becomes unusable at these small sizes and electrons would jump from one lane to the other if they're too close, causing errors.. But anyway, i think the next serious frontier in CPU dev would be quantum computers(imagine that? it can theoreticly read 8 bits at a time v.s 1 for current cpus!!) but still they have a long way to go for these to pop up in your fav hardware store, hehe.
Anonymous
a b à CPUs
June 18, 2010 4:43:10 AM

INTEL Core i7-920 Processor-------------------------- AMD Phenom II X4 965 Black Edition Processor--------

Frequency (MHz) ---- 2.66 GHz ----------------------- 3.4 GHz

Bus Processor ------- 4.8 GT/s QPI ------------------- 2.0 GHz HyperTransport

Size L2-Cache ------ 4 x 256 KB ---------------------- 4 x 512 KB

Architecture -------- Core i7 Bloomfield --------------- Phenom II X4 (Deneb)

MMX ------------------Yes --------------------------- Yes

3D Now! -------------- No ---------------------------- Yes

SSE ------------------ Yes --------------------------- Yes

SSE 2 ---------------- Yes --------------------------- Yes


NOW GUYS HELP ME OUT >>> WHICH ONE IS BETTER .... PLEASE
a b à CPUs
June 18, 2010 2:50:11 PM

^ You probably want to delete one of your redundant posts, if not start a new thread with your question instead of tagging onto this old thread. However if you go to http://www.anandtech.com/bench/Product/102?vs=47 and look up what apps you most commonly use, then you can compare the two at stock speeds. Both win some and lose some...
a b à CPUs
June 18, 2010 5:08:56 PM

This topic has been closed by Mousemonkey
!