Topic: how does cpu speed affect gpu mining? (Read 8900 times)
CPU speed is close to irrelevant with GPU bitcoin mining. However there is a small but demonstrable benefit to changing CPU frequency governor to high speed with scrypt GPU mining so therefore CPU speed matters slightly. I wouldn't get too excited though, it's only marginally more than a trivial speedup.
The speed of the processor is not as important as the number of PCIe lanes the CPU supports.
PCIe uses pathways between the CPU, RAM and disk called "lanes."
Each lane has an inbox and an outbox. The in and out boxes can communicate with the rest of the components on multiple in and out routes at the same time.
PCIe slots come in many speeds from 1x to 32x. The higher the number, the more in and out routes (lanes) that slot has.
For example, a PCIe 1x connector has one lane composed of four connections. a 32x PCIe has 32 lanes with a total of 128 connections.
To use an analogy, imagine you have a FedEx depot with 1x truck docking bays. Well, all the trucks are going to have to line up and load/unload one at a time. This would take days.
Now imagine you have a FedEx depot with 128 bays. You could load up every single truck at the same time and still have plenty of bays available.
The more connections a PCIe slot has the faster it can communicate. However, assigning more lanes to one PCIe slot takes them away from other slots, and there's a limit to how many lanes any CPU can have based on its architecture.
8GB/sec of bandwidth, either in the form of PCIe 2 x16 or PCIe 3 x8, is necessary to operate a Radeon 7970 at full load, so for 2 of them you need 16GB/sec bandwidth.
A Sandy Bridge CPU has 2x8 lanes (8GB/sec) whereas an Ivy Bridge CPU has 2x16 lanes (16GB/sec).
Or in other words, if you want to run two Radeon 7970's to maximum potential at load you need 2 x16 PCIe slots (PCIe version 2) or 2 x8 slots (PCIe version 3) and an Ivy Bridge CPU.
Core i5-3570 CPU's are Ivy Bridge.
Core i5-2500 CPU's are Sandy bridge.
Core i7-3770K CPU's are Ivy Bridge.
Core i7-2600K CPU's are Sandy Bridge.
Ivy Bridge processors are vastly more expensive than Sandy Bridge CPU's because they are newer and because managing all those lanes of traffic requires a lot more L2 (on dye) and L3 (off dye) cache. This type of RAM that is built directly onto the chip is extremely difficult and expensive to produce, resulting in lots of bad parts that have to be thrown out and re-fabricated at high cost.
It should also be noted that if you are running your GPU's in Crossfire/SLI configuration, 8 lanes are devoted to interdevice communication.
However mining software (Diablominer) addresses each GPU separately, dispatching work to them in chunks called jobs.
So you should switch Crossfire/SLI off before you begin mining.
PCIe uses pathways between the CPU, RAM and disk called "lanes."
Each lane has an inbox and an outbox. The in and out boxes can communicate with the rest of the components on multiple in and out routes at the same time.
PCIe slots come in many speeds from 1x to 32x. The higher the number, the more in and out routes (lanes) that slot has.
For example, a PCIe 1x connector has one lane composed of four connections. a 32x PCIe has 32 lanes with a total of 128 connections.
To use an analogy, imagine you have a FedEx depot with 1x truck docking bays. Well, all the trucks are going to have to line up and load/unload one at a time. This would take days.
Now imagine you have a FedEx depot with 128 bays. You could load up every single truck at the same time and still have plenty of bays available.
The more connections a PCIe slot has the faster it can communicate. However, assigning more lanes to one PCIe slot takes them away from other slots, and there's a limit to how many lanes any CPU can have based on its architecture.
8GB/sec of bandwidth, either in the form of PCIe 2 x16 or PCIe 3 x8, is necessary to operate a Radeon 7970 at full load, so for 2 of them you need 16GB/sec bandwidth.
A Sandy Bridge CPU has 2x8 lanes (8GB/sec) whereas an Ivy Bridge CPU has 2x16 lanes (16GB/sec).
Or in other words, if you want to run two Radeon 7970's to maximum potential at load you need 2 x16 PCIe slots (PCIe version 2) or 2 x8 slots (PCIe version 3) and an Ivy Bridge CPU.
Core i5-3570 CPU's are Ivy Bridge.
Core i5-2500 CPU's are Sandy bridge.
Core i7-3770K CPU's are Ivy Bridge.
Core i7-2600K CPU's are Sandy Bridge.
Ivy Bridge processors are vastly more expensive than Sandy Bridge CPU's because they are newer and because managing all those lanes of traffic requires a lot more L2 (on dye) and L3 (off dye) cache. This type of RAM that is built directly onto the chip is extremely difficult and expensive to produce, resulting in lots of bad parts that have to be thrown out and re-fabricated at high cost.
It should also be noted that if you are running your GPU's in Crossfire/SLI configuration, 8 lanes are devoted to interdevice communication.
However mining software (Diablominer) addresses each GPU separately, dispatching work to them in chunks called jobs.
So you should switch Crossfire/SLI off before you begin mining.
You typed a wall of text that is full of FAIL. How many GPUs do you have mining? I have over 40 7970s and I'm getting almost 700MH/s on each running off of 1x extenders and controlled by a lowly Celeron G530. I think you're confusing the bandwidth needed for applications like GAMING. We happen to be on a forum about BITCOIN MINING which uses trivial amounts of bandwidth on the bus. Death and Taxes has posted all the math before in his threads and had one of the nicest racks ever seen (haha I'm so childish).
GPU mining is nearing the end but no need to spread incorrect information (WRT mining at least).
Citation needed.
The speed of the processor is not as important as the number of PCIe lanes the CPU supports.
Nope. Bitcoin mining is incredibly low bandwidth. If they made a serial card GPU it would work fine. Plenty of miners have used GPU in PCIex1 slots with no loss in hashpower.
ok thanks for all the tips.
Im going to give cgminer a try. Out of curiosity what exactly makes it better than phoenix miner?
Whichever software you use usually doesn't matter because mostly they all use the same computing kernels. They are just different interfaces to run on top of the kernel.
Whichever software you use, you should use the Diablo kernel because it is provably faster on Radeon 7970 than any other.
I am using ccc 11.10 and sdk 2.1. General arguments are:
-k phatk2 AGGRESSION=13 WORKSIZE=256 BFI_INT VECTORS fastloops=false
Im going to give cgminer a try. Out of curiosity what exactly makes it better than phoenix miner?
-k phatk2 AGGRESSION=13 WORKSIZE=256 BFI_INT VECTORS fastloops=false
Im going to give cgminer a try. Out of curiosity what exactly makes it better than phoenix miner?
The speed of the processor is not as important as the number of PCIe lanes the CPU supports.
PCIe uses pathways between the CPU, RAM and disk called "lanes."
Each lane has an inbox and an outbox. The in and out boxes can communicate with the rest of the components on multiple in and out routes at the same time.
PCIe slots come in many speeds from 1x to 32x. The higher the number, the more in and out routes (lanes) that slot has.
For example, a PCIe 1x connector has one lane composed of four connections. a 32x PCIe has 32 lanes with a total of 128 connections.
To use an analogy, imagine you have a FedEx depot with 1x truck docking bays. Well, all the trucks are going to have to line up and load/unload one at a time. This would take days.
Now imagine you have a FedEx depot with 128 bays. You could load up every single truck at the same time and still have plenty of bays available.
The more connections a PCIe slot has the faster it can communicate. However, assigning more lanes to one PCIe slot takes them away from other slots, and there's a limit to how many lanes any CPU can have based on its architecture.
8GB/sec of bandwidth, either in the form of PCIe 2 x16 or PCIe 3 x8, is necessary to operate a Radeon 7970 at full load, so for 2 of them you need 16GB/sec bandwidth.
A Sandy Bridge CPU has 2x8 lanes (8GB/sec) whereas an Ivy Bridge CPU has 2x16 lanes (16GB/sec).
Or in other words, if you want to run two Radeon 7970's to maximum potential at load you need 2 x16 PCIe slots (PCIe version 2) or 2 x8 slots (PCIe version 3) and an Ivy Bridge CPU.
Core i5-3570 CPU's are Ivy Bridge.
Core i5-2500 CPU's are Sandy bridge.
Core i7-3770K CPU's are Ivy Bridge.
Core i7-2600K CPU's are Sandy Bridge.
Ivy Bridge processors are vastly more expensive than Sandy Bridge CPU's because they are newer and because managing all those lanes of traffic requires a lot more L2 (on dye) and L3 (off dye) cache. This type of RAM that is built directly onto the chip is extremely difficult and expensive to produce, resulting in lots of bad parts that have to be thrown out and re-fabricated at high cost.
It should also be noted that if you are running your GPU's in Crossfire/SLI configuration, 8 lanes are devoted to interdevice communication.
However mining software (Diablominer) addresses each GPU separately, dispatching work to them in chunks called jobs.
So you should switch Crossfire/SLI off before you begin mining.
PCIe uses pathways between the CPU, RAM and disk called "lanes."
Each lane has an inbox and an outbox. The in and out boxes can communicate with the rest of the components on multiple in and out routes at the same time.
PCIe slots come in many speeds from 1x to 32x. The higher the number, the more in and out routes (lanes) that slot has.
For example, a PCIe 1x connector has one lane composed of four connections. a 32x PCIe has 32 lanes with a total of 128 connections.
To use an analogy, imagine you have a FedEx depot with 1x truck docking bays. Well, all the trucks are going to have to line up and load/unload one at a time. This would take days.
Now imagine you have a FedEx depot with 128 bays. You could load up every single truck at the same time and still have plenty of bays available.
The more connections a PCIe slot has the faster it can communicate. However, assigning more lanes to one PCIe slot takes them away from other slots, and there's a limit to how many lanes any CPU can have based on its architecture.
8GB/sec of bandwidth, either in the form of PCIe 2 x16 or PCIe 3 x8, is necessary to operate a Radeon 7970 at full load, so for 2 of them you need 16GB/sec bandwidth.
A Sandy Bridge CPU has 2x8 lanes (8GB/sec) whereas an Ivy Bridge CPU has 2x16 lanes (16GB/sec).
Or in other words, if you want to run two Radeon 7970's to maximum potential at load you need 2 x16 PCIe slots (PCIe version 2) or 2 x8 slots (PCIe version 3) and an Ivy Bridge CPU.
Core i5-3570 CPU's are Ivy Bridge.
Core i5-2500 CPU's are Sandy bridge.
Core i7-3770K CPU's are Ivy Bridge.
Core i7-2600K CPU's are Sandy Bridge.
Ivy Bridge processors are vastly more expensive than Sandy Bridge CPU's because they are newer and because managing all those lanes of traffic requires a lot more L2 (on dye) and L3 (off dye) cache. This type of RAM that is built directly onto the chip is extremely difficult and expensive to produce, resulting in lots of bad parts that have to be thrown out and re-fabricated at high cost.
It should also be noted that if you are running your GPU's in Crossfire/SLI configuration, 8 lanes are devoted to interdevice communication.
However mining software (Diablominer) addresses each GPU separately, dispatching work to them in chunks called jobs.
So you should switch Crossfire/SLI off before you begin mining.
You're using a 5970, but what driver/SDK are you using? Can you post your exact Phoenix arguments? It sounds like you're running into the dreaded 100% CPU bug. You prolly didn't notice it before.
With 4 cores (and HT enabled), 100% on one core is 12-13%, or barely noticeable. With only 1 core (2 virtual cores), 100% on one core is 50% overall.
I also agree with Jack, switch to CGMiner. You won't be disappointed.
With 4 cores (and HT enabled), 100% on one core is 12-13%, or barely noticeable. With only 1 core (2 virtual cores), 100% on one core is 50% overall.
I also agree with Jack, switch to CGMiner. You won't be disappointed.
Try CGMiner, Phoenix uses Alot more CPU, And at higher intensities even uses upto 99% (a bug)
I went from GUIminer, To Phoenix, To Diablo, To CGMiner, and by far, CGMiner is the best
I went from GUIminer, To Phoenix, To Diablo, To CGMiner, and by far, CGMiner is the best
the cpu usage spike is due a bug in opencl runtime.
I have a gpu mining setup of a 5970. /i am using phoenix miner (latest version) and get around 790mhs. I use a core i7 3770, quad core at 3.4ghz. However I recently deactivated 3 of the cores so Im running off 1 core and I downclocked the cpu to 1.6ghz to save power. I know that I am using my gpu for mining, but for some reason my cpu is always at 50% usage when mining, regardless of cores active or clockspeed. Now mining with the 1 active core at 1.6ghz I am averaging around 720mhs according to my pool, even though phoenix miner still reports the exact same hash rate as with all cores active. These results have been consistent over several days now, and there are no other variables, so I know that the cpu speed/cores is affecting my mining speed. I was wondering does anyone know why this is, or have the same problem?
Jump to: