Timeline for Why do CPU's typically connect to only one bus?
Current License: CC BY-SA 3.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 26, 2016 at 10:04 | comment | added | pjc50 | @DrZ214 a "lane" is a single bit wide, a clock/data signal sent as two pairs. The unique thing about PCIe compared to, say, the DDR bus is that PCIe can freely aggregate and deaggregate lanes, whereas most buses you have to take all or nothing. | |
| Feb 26, 2016 at 8:59 | comment | added | Chris H | What's interesting is that this represents a shift back towards the architecture of even older systems, where the memory and storage buses (etc.) went directly to the CPU. | |
| Feb 26, 2016 at 5:17 | comment | added | Tom Carpenter | @DrZ214 They are not PCI, they're PCIe which is a serial bus rather than a parallel one. PCIe is either 2.5Gbps, 5Gpbs or 8Gbps per lane in both directions (full duplex) - with 16 lane being the widest usually seen giving a theoretical maximum 128Gbps in both directions. And yes, the comment above about the speeds of the processor was intentionally overly simplistic, but not unrealistic - unrelated to CPUs, I am working on an FPGA design at the moment which processes data at 315Gbps, and that is nowhere near the full performance possible of the FPGA, what's limiting that is lack of data! | |
| Feb 26, 2016 at 4:52 | comment | added | DrZ214 | P.S., quick question: In your diagram, the PCI connections are labeled with lanes. How many bits are in a lane, and why arent the other things (like memory) labeled with lanes? | |
| Feb 26, 2016 at 4:47 | comment | added | DrZ214 | It is definitely amazing how fast things are getting, but I hafta say your math is a little too ideal. Most programs don't use multi-threading (the trend is definitely towards it, but there are many cs problems that just can't be done with parallel processing), so it seems too ideal to assume all 4 cores can work on a saturated bus of that bandwidth. So 256 Gbps, if that core is only running 1 process, also a bit ideal. But still fast and I get your point and your answer was really good. I would accept it but unit128_t's answer is also great. | |
| Feb 26, 2016 at 3:43 | comment | added | Tom Carpenter | @DrZ214 Take a modern 4GHz, 64bit, Quad-Core processor. That's 4*64*4=1024Gbps. QPI which is the latest Intel link between CPU and Chipset is 307Gbps. So while shocking, it's not actually that far fetched. Just amazing how fast things are getting! | |
| Feb 26, 2016 at 3:19 | comment | added | DrZ214 |
Thank you, especially the second paragraph hit home. However, on the die you can achieve speeds on the order of Tbps! Yikes, isn't that beginning to outrun the CPU's ability to process it fast enough?
|
|
| Feb 26, 2016 at 3:03 | comment | added | uint128_t | +1, you beat me to it :) Nice answer, especially for the historical reasons for architecture design. | |
| Feb 26, 2016 at 2:51 | history | answered | Tom Carpenter | CC BY-SA 3.0 |