PCIe 1.1 and mining

[PovertyByte]

I was taking a serious look at this random old HP I may convert to a 3 GPU miner with some of those PCIe x1 risers. I began to question if PCIe 3.0 was needed to mine effectively and better yet if this old HP is a 2.0 or 1.1 board. It's a 1.1

The mining speed I get in the PCIe x16 slot is perfectly normal but I do not know if the x1 will cut it

My attempt at looking up if modern GPU's can run on older PCIe boards kept turning up results about how the primitive legacy CPU's and RAM would bottleneck the GPU's which is not a miners concern.

Is 1.1 too slow? Can 2.0 cut it? Or does it have to be a 3.0?

[PovertyByte]

If I get this right an x1 3.0 slot can only do 2GB/s of speed meanwhile a GTX 1070 starts at 8GB/s of memory bandwidth?

Something isn't right, you guys should be bottlenecked here?

[LoneRangir]

ETH requires little actual PCIe bandwidth. The numbers you quote are the theoretical max of the channel. Some algorithms may require more bandwidth than ETH. I remember ZEC took quite a bit more PCIe in the earlier days. I'm sure with all the optimizations over the last few months, the requirements of the channel have gone down

[PovertyByte]

ETH requires little actual PCIe bandwidth. The numbers you quote are the theoretical max of the channel. Some algorithms may require more bandwidth than ETH. I remember ZEC took quite a bit more PCIe in the earlier days. I'm sure with all the optimizations over the last few months, the requirements of the channel have gone down

Not sure what to look at in GPU-z or any other hardware sensor program to figure out how much speed the PCIe slot is using.

By the look of things, all these GPU's exceed PCIe x16 3.0 speeds now

Maybe ETH uses little but that may be one exception. Just because many people are using x1 risers on 3.0 boards and probably a few 2.0 boards doesn't mean that the 1.1 will work. Those are old motherboards and most people buy 6 GPU boards and build mining rigs gound up from newer product listings

[JaredKaragen]

The data processed by the video card is all processed (for the most part anyways) on the video card.

The PCIE bandwidth is just that, how much data can be sent at once, vs how fast can it be sent at once.  You are just adding more independent connections by increasing the PCIE spec up from 1.... so the actual speed of the transmission isn't really any faster in basic terms; there are just more lines you can send larger amounts of data at the same time.

Mining apps aren't sending butt-loads of data to the video card like a game is;  it is sending an instruction set and a few strings/integers for it to calculate instead of a giant map and a bunch of texture data as well.

The bandwidth on the bus that's actually used by the miner app is a fraction of a percent overall; compared to the slot's bandwidth capability; in many if not all respects.  Remember, the video card is basically just returning jobid and nonce information for the most part.... that's a tiny set of bits compared to the amount of data that can be sent over a 1.1 bus.

Can anyone here that knows more than I, reference a coin that is bus-heavy when it comes to it's workload?  I can't think of one that I know of...

[Viperho]

I remember ZEC took quite a bit more PCIe in the earlier days. I'm sure with all the optimizations over the last few months, the requirements of the channel have gone down

Actually ZEC was more CPU intensive early on, not really PCIe intensive. My i7s were substantially faster with the same GPU then some of my older core 2 duos.

[PovertyByte]

The data processed by the video card is all processed (for the most part anyways) on the video card.

The PCIE bandwidth is just that, how much data can be sent at once, vs how fast can it be sent at once.  You are just adding more independent connections by increasing the PCIE spec up from 1.... so the actual speed of the transmission isn't really any faster in basic terms; there are just more lines you can send larger amounts of data at the same time.

Mining apps aren't sending butt-loads of data to the video card like a game is;  it is sending an instruction set and a few strings/integers for it to calculate instead of a giant map and a bunch of texture data as well.

The bandwidth on the bus that's actually used by the miner app is a fraction of a percent overall; compared to the slot's bandwidth capability; in many if not all respects.  Remember, the video card is basically just returning jobid and nonce information for the most part.... that's a tiny set of bits compared to the amount of data that can be sent over a 1.1 bus.

Can anyone here that knows more than I, reference a coin that is bus-heavy when it comes to it's workload?  I can't think of one that I know of...

That is very good to know. It would be very useful to utilize that old HP. Already have an aftermarket PSU. I can probably lay the original tower on the side and weld some bars to hold the GPU's over the rig

[JaredKaragen]

The data processed by the video card is all processed (for the most part anyways) on the video card.

The PCIE bandwidth is just that, how much data can be sent at once, vs how fast can it be sent at once.  You are just adding more independent connections by increasing the PCIE spec up from 1.... so the actual speed of the transmission isn't really any faster in basic terms; there are just more lines you can send larger amounts of data at the same time.

Mining apps aren't sending butt-loads of data to the video card like a game is;  it is sending an instruction set and a few strings/integers for it to calculate instead of a giant map and a bunch of texture data as well.

The bandwidth on the bus that's actually used by the miner app is a fraction of a percent overall; compared to the slot's bandwidth capability; in many if not all respects.  Remember, the video card is basically just returning jobid and nonce information for the most part.... that's a tiny set of bits compared to the amount of data that can be sent over a 1.1 bus.

Can anyone here that knows more than I, reference a coin that is bus-heavy when it comes to it's workload?  I can't think of one that I know of...

That is very good to know. It would be very useful to utilize that old HP. Already have an aftermarket PSU. I can probably lay the original tower on the side and weld some bars to hold the GPU's over the rig

The only drawback to older machines and faster GPU's, is the miner apps typically verify the nonce before submitting; and that's done on the cpu.   

It isn't a huge impact, but I do know that can be a facet with multiple GPU rigs on older gear. (lets say pre lga 1155 or so?) and algos that are pretty intensive such as X11evo and neoscrypt.

[PovertyByte]

The data processed by the video card is all processed (for the most part anyways) on the video card.

The PCIE bandwidth is just that, how much data can be sent at once, vs how fast can it be sent at once.  You are just adding more independent connections by increasing the PCIE spec up from 1.... so the actual speed of the transmission isn't really any faster in basic terms; there are just more lines you can send larger amounts of data at the same time.

Mining apps aren't sending butt-loads of data to the video card like a game is;  it is sending an instruction set and a few strings/integers for it to calculate instead of a giant map and a bunch of texture data as well.

The bandwidth on the bus that's actually used by the miner app is a fraction of a percent overall; compared to the slot's bandwidth capability; in many if not all respects.  Remember, the video card is basically just returning jobid and nonce information for the most part.... that's a tiny set of bits compared to the amount of data that can be sent over a 1.1 bus.

Can anyone here that knows more than I, reference a coin that is bus-heavy when it comes to it's workload?  I can't think of one that I know of...

That is very good to know. It would be very useful to utilize that old HP. Already have an aftermarket PSU. I can probably lay the original tower on the side and weld some bars to hold the GPU's over the rig

The only drawback to older machines and faster GPU's, is the miner apps typically verify the nonce before submitting; and that's done on the cpu.   

It isn't a huge impact, but I do know that can be a facet with multiple GPU rigs on older gear. (lets say pre lga 1155 or so?) and algos that are pretty intensive such as X11evo and neoscrypt.

Some sort of LGA 775 based motherboard, Q6600 CPU

No issues mining one GTX1070 with it right now. I can just get 1 powered riser and just connect the 1070 from my main PC into it for a test run to make sure it can handle 2 GPU's

[bathrobehero]

GPU bandwidth doesn't matter in terms of PCI-e speed; it's just how fast the GPU can use its own memory but not everything in the memory gets transferred via PCI-e, only very very little.

In fact even PCI-e 1.0 1x (USB risers) are enoguh for mining virtually all coins which only have 250 MB/s bandwidth.