FSB vs Quickpath

[PerplexShyt]

I know quickpath is faster. But if two computers had the same specs, but one with FSB and the other with Quickpath, how much faster would the latter be?

And excuse this noob question, but is there a way to modify your setup from FSB to Quickpath?

 

[DoFoT9]

Interesting question, really!

FSB is no longer around basically, so the scenario you put forward would never happen - just FYI

But anyway.

In a FSB scenario, the CPU talks to other parts through the northbridge chip, this was the controlling mechanism in it all. The northbridge had memory, hard drive, graphics, etc connecting into it. The rate of frequency of all connected devices were in ratio. For example, CPUs might be in ratio of 4:1, so a 500MHz FSB would = 2000MHz CPU (very simple scenario). RAM might be in a 5:4 ratio, etc. This northbridge acted like a bottleneck, apparently maxing out roughly around 12,800MBs throughput. When you consider that high end GPUs these days push easily past 10,000MBs you understand why!

And BTW, a GT/s = 1000MT/s (megatransfer/s), which equates to FSB * transfers per clock cycle.

QPI is a decentralised method of transferring data. The bus lanes for transferring are smaller, but they are uni-directional, and can pass directly to the component without the latency/queuing of other requests getting stuck at the northbridge.

Why ask about QPI, what DMI is much more prominent these days? (DMI is also decentralised also).

edit: Didn't really answer your question - It's kind of hard to compare, given there are a few key differences. I would say that QPI would undoubtedly be faster though, especially under high load.

edit2: Oh, FYI - DMI (Direct Media Interconnect) is a more consumerised Intel product released for Single Socket Motherboards. Where as QPI is for Dual Socket Motherboards.

 

[Inconsequential]

Interesting question, really!

FSB is no longer around basically, so the scenario you put forward would never happen - just FYI

But anyway.

In a FSB scenario, the CPU talks to other parts through the northbridge chip, this was the controlling mechanism in it all. The northbridge had memory, hard drive, graphics, etc connecting into it. The rate of frequency of all connected devices were in ratio. For example, CPUs might be in ratio of 4:1, so a 500MHz FSB would = 2000MHz CPU (very simple scenario). RAM might be in a 5:4 ratio, etc. This northbridge acted like a bottleneck, apparently maxing out roughly around 12,800MBs throughput. When you consider that high end GPUs these days push easily past 10,000MBs you understand why!

And BTW, a GT/s = 1000MT/s (megatransfer/s), which equates to FSB * transfers per clock cycle.

QPI is a decentralised method of transferring data. The bus lanes for transferring are smaller, but they are uni-directional, and can pass directly to the component without the latency/queuing of other requests getting stuck at the northbridge.

Why ask about QPI, what DMI is much more prominent these days? (DMI is also decentralised).

QPI is used in Mac Pros since the 2009 MP, so perhaps that is why?

All much of a muchness tbh.

And no, there isn't any chance of going FSB to QPI/DMI.

 

[DoFoT9]

QPI is used in Mac Pros since the 2009 MP, so perhaps that is why?

All much of a muchness tbh.

And no, there isn't any chance of going FSB to QPI/DMI.

How many people do you know these days that chase MacPros? But yes, I guess there is the rare case.

From what I understand, FSB stopped around 10GBs, where as QPI at introduction in 2008 was at 25.6GT/s (12.8GBs). I would assume this has advanced since then

 

[goMac]

The main advantage of QPI is bypassing the front side bus. It makes memory operations much faster because the CPU is able to directly talk to memory.

On non memory heavily applications, I would expect little difference. On very memory heavily applications, I would expect a major major difference, possibly in a 2x range.

But you've got to have enough memory and the right sort of app to get that.

And no, you can't modify. QPI works by using special features on the newer CPUs, and newer motherboards without a FSB. In order to make that happen you'll notice Apple actually moved the CPUs right next to the RAM in the new machines so they could directly talk to one another.

 

[DoFoT9]

The main advantage of QPI is bypassing the front side bus. It makes memory operations much faster because the CPU is able to directly talk to memory.

Though in saying that, FSB does deal with higher data blocks per "transfer", given that QPI is limited to 20 in + 20 out, and FSB is the full 64. (QPI does get 84 flints transferred in that time though, so I am a little confused by that).