and since you can’t breakout the x16 lanes to 16x x1 devices
Isn't there something similar to the Bitcoin risers which could do this? Instead of connecting 4 GPUs to share one lane, why not give them one each?
Ian&Steve C. wrote:
if you have unstable power connections you will have issues
I've had a lot of that. If the power plugs are tarnished, they get thrown out, if the sockets on the GPU are tarnished, I bypass them by soldering the power wires onto the board. I think the problem is the silver coloured plugs on cheap PSUs. the gold coloured ones on Corsair supplies never tarnish. Unfortunately I can't find where to buy gold coloured plugs to fit to my 1kW 12V only supplies.
Ian&Steve C. wrote:
if you have poorly shielded or bad connections on your PCIe lines you will have issues
I just use the stuff I got from an ex-bitcoin miner person, it seems to be fine. Just some standard looking USB cables and some well made looking cards at each end.
Ian&Steve C. wrote:
if you have an unstable CPU or memory overclock you will have issues
I never overclock, tried that in the past, you get a tiny speed increase at the expense of instability. I trust the manufacturer has already set it at the correct speed. I don't mess around with the revs on my car engine either!
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I've worked in a university where they researched new techniques for creating microchips. I was told everything had protection diodes nowadays.
Protection structures are not new. All the Intel parts had provisions when I joined in 1974. People have worked nonstop on making them more effective. I knew someone at my church in Silicon Valley who was a very long-term employee of National Semiconductor who spent much of his career on the things. I even spent a few days of mine trying to improve the provision which Mel Bazes had designed into his re-do of the 8080 (yes, this was a very long time ago).
If you think you are immune, you are wrong. You've just been lucky. This stuff is a second cousin to lightning, and everyone knows how hard it is to guess just when and where that will take practical effect.
I'm talking about in the 90s, when they said they were going to get much better. I heard of a lot of 90s consumer parts having problems, and I was careful back then. But in this century I think they're good enough. I'm going to go with humidity. If you have low humidity, you can get a static shock that you can feel when you touch a grounded object. If that's enough to make you jump, I'm sure it's enough to break a fragile little semiconductor. But in Scotland it's 50-90% RH, probably static just doesn't happen in these conditions.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
and since you can’t breakout the x16 lanes to 16x x1 devices
Isn't there something similar to the Bitcoin risers which could do this? Instead of connecting 4 GPUs to share one lane, why not give them one each?
natively the boards just aren’t setup to do this.
There are some Mining boards that are setup with a ton of x1 slots or USB ports (which are electrically PCIe x1 with a USB connector). These boards run these ports through the chipset and are ultimately limited by the chipset<->CPU link bandwidth. The chipset basically acts like a PCIe switch. On these boards you’ll still find the main x16 slot tied to the CPU as normal. It’s not using those lanes for all it’s ports.
there are AIC devices which either bifurcate the slot, or have a PCIe switch.
in the case of bifurcation, it physically chops up the lanes. So a x16 slot can be x8x8 or x8x4x4 or x4x4x4x4. I’ve never seen any other configuration. And it also requires motherboard BIOS support to be able to do it. You can’t just throw in a bifurcation card and expect it to work without proper setup. These kinds of cards are relatively cheap.
in the case of a PCIe switch, these treat the slot as a single “pipe” so to speak and can be configured with many devices and even over provision the theoretical max bandwidth. You can take 4x x8 devices and throw them down a single x16 pipe for example. These kinds of cards will work without BIOS support as long as you don’t run into other limits like memory allocation or device number limits. The downside is that these devices require their own controller chip, can be a bit power hungry, and are generally quite expensive. Those cheap 4-in-1 USB splitters fall into this category, but they only switch on a single lane since to keep costs down they use a very cheap PCIe switch that’s only setup for a single lane. They can (and do) make ones that switch on a x16 slot, but are prohibitively expensive for most people/uses.
in the case of a PCIe switch, these treat the slot as a single “pipe” so to speak and can be configured with many devices and even over provision the theoretical max bandwidth. You can take 4x x8 devices and throw them down a single x16 pipe for example. These kinds of cards will work without BIOS support as long as you don’t run into other limits like memory allocation or device number limits. The downside is that these devices require their own controller chip, can be a bit power hungry, and are generally quite expensive. Those cheap 4-in-1 USB splitters fall into this category, but they only switch on a single lane since to keep costs down they use a very cheap PCIe switch that’s only setup for a single lane. They can (and do) make ones that switch on a x16 slot, but are prohibitively expensive for most people/uses.
Can you link to such a switch? I've only seen the ones that quadfurcate a single lane to four single lane devices.
I had a search and only found one similar to what you speak of, 16x to 8 of presumably 1x:
I’ve bought a few PCBs from this guy, believe he’s located in Germany. I never bought any of his PLX boards since they’re expensive and I can get by with bifurcation since my hardware supports it.
but other server vendors offer similar products more aimed at the server space. Pricing is similar. A lot of times you’d need to be aware of the product you’re looking for and search by the part number, rather than stumbling across it on eBay.
I’ve bought a few PCBs from this guy, believe he’s located in Germany. I never bought any of his PLX boards since they’re expensive and I can get by with bifurcation since my hardware supports it.
What percentage of boards allow bifurcation? And how would I find out if my board supports it? I found my best board supports it, by chance someone else had asked, but my second best board google doesn't help.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
percentage is impossible to answer. in terms of older systems that aren't server boards, probably relatively few. but it's become more common on even consumer boards in recent years. it's going to vary on a board-by-board basis. Asrock seems to have a good record of providing this feature in their BIOSes though.
sometimes you have to ask the manufacturer. It's not something most people are looking to use and it's not usually present in marketing. other times you can just look in the BIOS yourself (or the motherboard manual) and look for bifurcation settings. if you don't find any, then it likely doesn't support it.
percentage is impossible to answer. in terms of older systems that aren't server boards, probably relatively few. but it's become more common on even consumer boards in recent years. it's going to vary on a board-by-board basis. Asrock seems to have a good record of providing this feature in their BIOSes though.
sometimes you have to ask the manufacturer. It's not something most people are looking to use and it's not usually present in marketing. other times you can just look in the BIOS yourself (or the motherboard manual) and look for bifurcation settings. if you don't find any, then it likely doesn't support it.
My best board (an MSI X470 Gaming Plus Max) does support it. But I can find no such setting in the BIOS. Seems Asus are good at it, they make the cards to do it too, so I'll keep them and ASRock in mind next time I get a board.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
My best board (an MSI X470 Gaming Plus Max) does support it. But I can find no such setting in the BIOS. Seems Asus are good at it, they make the cards to do it too, so I'll keep them and ASRock in mind next time I get a board.
Your board only supports bifurcation in the general sense that most consumer boards do. not really in the way I've described in earlier posts. it only bifurcates in one way and it's not user configurable. the bifurcation happens automatically based on how you populate the slots. you can't plug multiple devices into a single slot unless you use the PCIe switching cards.
most consumer boards (yours included) recently have two physical x16 slots. but the CPU only has 16 lanes. the "top" x16 slot (closest to the CPU) will be electrically wired for all 16 lanes. the secondary x16 slot will usually only be wired for x8 lanes even though it's x16 physically. this is how your board is, and it's easy to see that the second slot only has pins in half (x8) of the x16 slot.
so if you have only one GPU in the top x16 slot, it will run with all 16 lanes of bandwidth.
however if you add a GPU (or other PCIe device) to the second x16 slot, the board will auto bifurcate that first slot to x8 and use the other x8 lanes for the second slot.
the third x16 PCIe slot on your board is run through the X470 chipset and is only PCIe 2.0 x4 electrically.
this is a very common layout for both Intel and AMD consumer boards.
Static electricity discharge is something to beware of as well while manipulating the pins.
Cheers, Mike.
Wasn't that only a problem in the 90s? I've never been careful with any computer parts and I've never electrocuted one.
In my varied career in the computer business I spent about six years actually employed in Intel's Reliability activity.
Yes, electrostatic discharge is a real problem. I've seen the high-resolution SEM imagery of the aftermath of plenty of parts.
No, people don't believe it. Their logic resembles the logic of a person who has successfully crossed a road blindfolded several times, and concludes that since they did not die, there is no reason to confine themselves to a crosswalk, or even to look for traffic.
Not just people like us building their own machines. I had a conversation with the manufacturing manager of a small company which built workstations for computer-aided design in the mid 1980s who actually believe that there were no real ESD problems and that 3M was just building on a myth in order to sell him stuff to help manage the problem in his facility.
I taught basic electricity for nearly 20 years at a community college (among other things automotive) before I retired, so this subject of Electrostatic Discharge is relatively important to me. I’d like to make users aware of some things so that they can be informed about what is actually happening within their computer systems.
To quote a NASA article on the use of a non-effective grounding wrist strap:
NASA Interagency Working Group on Electrostatic Discharge (IAWG-ESD)
“It typically takes an ESD discharge of more than 2,500 to 3,500 volts for a person to feel a shock, however, a small discharge of 25 volts can damage electronic components without the engineer feeling a thing.”
GamersNexus’ Steve Burke made actual reference to this very article in his lengthy video (just shy of 36 minutes).
GamersNexus video on “Static Electricity Misconceptions”
One of the biggest common misconceptions is touching your hand to the computer case before doing anything within the case to prevent an electrical shock. The point not clearly made is that they must touch a clear metal portion of the case (not a painted surface or a plastic piece) while the power supply's power cord is still plugged in to a grounded outlet. Since most people will at least follow some directions, like turning off the PSU and then unplugging it, this procedure is irrelevant.
Ian&Steve C. wrote:EPYC
)
128?! Oooh! You just made me.... https://youtu.be/o-x9w2HzZf8
Isn't there something similar to the Bitcoin risers which could do this? Instead of connecting 4 GPUs to share one lane, why not give them one each?
I've had a lot of that. If the power plugs are tarnished, they get thrown out, if the sockets on the GPU are tarnished, I bypass them by soldering the power wires onto the board. I think the problem is the silver coloured plugs on cheap PSUs. the gold coloured ones on Corsair supplies never tarnish. Unfortunately I can't find where to buy gold coloured plugs to fit to my 1kW 12V only supplies.
I just use the stuff I got from an ex-bitcoin miner person, it seems to be fine. Just some standard looking USB cables and some well made looking cards at each end.
I never overclock, tried that in the past, you get a tiny speed increase at the expense of instability. I trust the manufacturer has already set it at the correct speed. I don't mess around with the revs on my car engine either!
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
archae86 wrote: Peter Hucker
)
I'm talking about in the 90s, when they said they were going to get much better. I heard of a lot of 90s consumer parts having problems, and I was careful back then. But in this century I think they're good enough. I'm going to go with humidity. If you have low humidity, you can get a static shock that you can feel when you touch a grounded object. If that's enough to make you jump, I'm sure it's enough to break a fragile little semiconductor. But in Scotland it's 50-90% RH, probably static just doesn't happen in these conditions.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Peter Hucker
)
natively the boards just aren’t setup to do this.
There are some Mining boards that are setup with a ton of x1 slots or USB ports (which are electrically PCIe x1 with a USB connector). These boards run these ports through the chipset and are ultimately limited by the chipset<->CPU link bandwidth. The chipset basically acts like a PCIe switch. On these boards you’ll still find the main x16 slot tied to the CPU as normal. It’s not using those lanes for all it’s ports.
there are AIC devices which either bifurcate the slot, or have a PCIe switch.
in the case of bifurcation, it physically chops up the lanes. So a x16 slot can be x8x8 or x8x4x4 or x4x4x4x4. I’ve never seen any other configuration. And it also requires motherboard BIOS support to be able to do it. You can’t just throw in a bifurcation card and expect it to work without proper setup. These kinds of cards are relatively cheap.
in the case of a PCIe switch, these treat the slot as a single “pipe” so to speak and can be configured with many devices and even over provision the theoretical max bandwidth. You can take 4x x8 devices and throw them down a single x16 pipe for example. These kinds of cards will work without BIOS support as long as you don’t run into other limits like memory allocation or device number limits. The downside is that these devices require their own controller chip, can be a bit power hungry, and are generally quite expensive. Those cheap 4-in-1 USB splitters fall into this category, but they only switch on a single lane since to keep costs down they use a very cheap PCIe switch that’s only setup for a single lane. They can (and do) make ones that switch on a x16 slot, but are prohibitively expensive for most people/uses.
_________________________________________________________________________
Ian&Steve C. wrote:in the
)
Can you link to such a switch? I've only seen the ones that quadfurcate a single lane to four single lane devices.
I had a search and only found one similar to what you speak of, 16x to 8 of presumably 1x:
https://www.ebay.co.uk/itm/125099145100
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Here’s an example.
)
Here’s an example. https://c-payne.com/products/pex-8747-plx-pcie-switch-card-x8x8x8x8-3w
I’ve bought a few PCBs from this guy, believe he’s located in Germany. I never bought any of his PLX boards since they’re expensive and I can get by with bifurcation since my hardware supports it.
but other server vendors offer similar products more aimed at the server space. Pricing is similar. A lot of times you’d need to be aware of the product you’re looking for and search by the part number, rather than stumbling across it on eBay.
_________________________________________________________________________
Ian&Steve C. wrote:I’ve
)
What percentage of boards allow bifurcation? And how would I find out if my board supports it? I found my best board supports it, by chance someone else had asked, but my second best board google doesn't help.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
percentage is impossible to
)
percentage is impossible to answer. in terms of older systems that aren't server boards, probably relatively few. but it's become more common on even consumer boards in recent years. it's going to vary on a board-by-board basis. Asrock seems to have a good record of providing this feature in their BIOSes though.
sometimes you have to ask the manufacturer. It's not something most people are looking to use and it's not usually present in marketing. other times you can just look in the BIOS yourself (or the motherboard manual) and look for bifurcation settings. if you don't find any, then it likely doesn't support it.
_________________________________________________________________________
Ian&Steve C.
)
My best board (an MSI X470 Gaming Plus Max) does support it. But I can find no such setting in the BIOS. Seems Asus are good at it, they make the cards to do it too, so I'll keep them and ASRock in mind next time I get a board.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Peter Hucker wrote: My best
)
Your board only supports bifurcation in the general sense that most consumer boards do. not really in the way I've described in earlier posts. it only bifurcates in one way and it's not user configurable. the bifurcation happens automatically based on how you populate the slots. you can't plug multiple devices into a single slot unless you use the PCIe switching cards.
most consumer boards (yours included) recently have two physical x16 slots. but the CPU only has 16 lanes. the "top" x16 slot (closest to the CPU) will be electrically wired for all 16 lanes. the secondary x16 slot will usually only be wired for x8 lanes even though it's x16 physically. this is how your board is, and it's easy to see that the second slot only has pins in half (x8) of the x16 slot.
so if you have only one GPU in the top x16 slot, it will run with all 16 lanes of bandwidth.
however if you add a GPU (or other PCIe device) to the second x16 slot, the board will auto bifurcate that first slot to x8 and use the other x8 lanes for the second slot.
the third x16 PCIe slot on your board is run through the X470 chipset and is only PCIe 2.0 x4 electrically.
this is a very common layout for both Intel and AMD consumer boards.
_________________________________________________________________________
archae86 wrote: Peter Hucker
)
I taught basic electricity for nearly 20 years at a community college (among other things automotive) before I retired, so this subject of Electrostatic Discharge is relatively important to me. I’d like to make users aware of some things so that they can be informed about what is actually happening within their computer systems.
To quote a NASA article on the use of a non-effective grounding wrist strap:
NASA Interagency Working Group on Electrostatic Discharge (IAWG-ESD)
“It typically takes an ESD discharge of more than 2,500 to 3,500 volts for a person to feel a shock, however, a small discharge of 25 volts can damage electronic components without the engineer feeling a thing.”
GamersNexus’ Steve Burke made actual reference to this very article in his lengthy video (just shy of 36 minutes).
GamersNexus video on “Static Electricity Misconceptions”
https://www.youtube.com/watch?v=m54ifTCgtN4&t=9s
And Jay from JayzTwoCents did a slightly disparaging video showing how robust the computers of today are, while still making a point.
JayzTwoCents video on “We built a computer wrong on PURPOSE!”
https://www.youtube.com/watch?v=CDUZRNp4pxg
One of the biggest common misconceptions is touching your hand to the computer case before doing anything within the case to prevent an electrical shock. The point not clearly made is that they must touch a clear metal portion of the case (not a painted surface or a plastic piece) while the power supply's power cord is still plugged in to a grounded outlet. Since most people will at least follow some directions, like turning off the PSU and then unplugging it, this procedure is irrelevant.
Proud member of the Old Farts Association