Acorn M.2 FPGA based GPU Accelerator

[dbalz]

Where is the chart, I can't find a link anywhere!

[1715073935]

1715073935

[1715073935]

1715073935

[1715073935]

1715073935

[GPUHoarder]

alll risers support 3.0 (x1). well if motherboard support only 2.0, than the riser will do 2.0 as well.

Its simple the usb3 cables dont have the quality to negotiate pcie 3.0, think about it usb3 cables are made for 8b/10b same as what pcie 2.0 run, but pcie 3.0 use 128b/130b.

Now there is faster standards like USB 3.1 Gen2 or USB 3.2 that do use 128b/130b but i would imagine that they also have much shorter cables, and its not cheap

EDIT: Already at pcie 2.0 these usb risers suck, if you have problems with gpu fallen of the bus errors in your syslog, and your gpu's are not too much clocked, you can usualy fix it by forcing pcie 1.0, so the risers are iffy even for 2.0, you say they can 3.0 I say NO !

8/10 vs 128/130 has absolutely nothing to do with this. That is simply the encoding for DC Balancing/ensure there is a periodic bitflip on the transcievers so they can stay locked. It was changed in PCIe 3.0 due to improved transceivers and to lower overhead.

Also for everyone checking lspci - most GPUs now adays are designed for low-power idle. They run the link at x1 unless they are actually being used for transfer! Run a bandwidth heavy test or activity and run lspci, you’ll see LnkStatus kick up to 3.0 (8 Gbps)

We just spent weeks testing dozens of risers and slots on 3 dozen motherboards. I haven’t found a single riser that can’t sustain PCIe 3.0 speeds. While monitoring PCIe for any errors.  I have, however, found crap “mining” motherboards that have PCIe errors on their slots even with a x1 device plugged straight in. I’ll take that actual testing over your anecdote.


Software (miner) and APIs: This is a Pre-Order for the hardware. The software will launch with the hardware shipping late August. We’ve already asked devs to reach out for API and SDK information that will be available before launch. Developer boards haven’t even been sent out yet to other devs.

[Marvell2]

alll risers support 3.0 (x1). well if motherboard support only 2.0, than the riser will do 2.0 as well.

Its simple the usb3 cables dont have the quality to negotiate pcie 3.0, think about it usb3 cables are made for 8b/10b same as what pcie 2.0 run, but pcie 3.0 use 128b/130b.

Now there is faster standards like USB 3.1 Gen2 or USB 3.2 that do use 128b/130b but i would imagine that they also have much shorter cables, and its not cheap

EDIT: Already at pcie 2.0 these usb risers suck, if you have problems with gpu fallen of the bus errors in your syslog, and your gpu's are not too much clocked, you can usualy fix it by forcing pcie 1.0, so the risers are iffy even for 2.0, you say they can 3.0 I say NO !

8/10 vs 128/130 has absolutely nothing to do with this. That is simply the encoding for DC Balancing/ensure there is a periodic bitflip on the transcievers so they can stay locked. It was changed in PCIe 3.0 due to improved transceivers and to lower overhead.

Also for everyone checking lspci - most GPUs now adays are designed for low-power idle. They run the link at x1 unless they are actually being used for transfer! Run a bandwidth heavy test or activity and run lspci, you’ll see LnkStatus kick up to 3.0 (8 Gbps)

We just spent weeks testing dozens of risers and slots on 3 dozen motherboards. I haven’t found a single riser that can’t sustain PCIe 3.0 speeds. While monitoring PCIe for any errors.  I have, however, found crap “mining” motherboards that have PCIe errors on their slots even with a x1 device plugged straight in. I’ll take that actual testing over your anecdote.


Software (miner) and APIs: This is a Pre-Order for the hardware. The software will launch with the hardware shipping late August. We’ve already asked devs to reach out for API and SDK information that will be available before launch. Developer boards haven’t even been sent out yet to other devs.

darn so to use the miners you have to have acorn running ? so if you have say one nest acord adaptor board with 1 or two acorn installed would the miner be implemented for all cards on the rig or only the acorn boosted gpus?

[dsmn]

8/10 vs 128/130 has absolutely nothing to do with this. That is simply the encoding for DC Balancing/ensure there is a periodic bitflip on the transcievers so they can stay locked. It was changed in PCIe 3.0 due to improved transceivers and to lower overhead.

Also for everyone checking lspci - most GPUs now adays are designed for low-power idle. They run the link at x1 unless they are actually being used for transfer! Run a bandwidth heavy test or activity and run lspci, you’ll see LnkStatus kick up to 3.0 (8 Gbps)

We just spent weeks testing dozens of risers and slots on 3 dozen motherboards. I haven’t found a single riser that can’t sustain PCIe 3.0 speeds. While monitoring PCIe for any errors.  I have, however, found crap “mining” motherboards that have PCIe errors on their slots even with a x1 device plugged straight in. I’ll take that actual testing over your anecdote.

Software (miner) and APIs: This is a Pre-Order for the hardware. The software will launch with the hardware shipping late August. We’ve already asked devs to reach out for API and SDK information that will be available before launch. Developer boards haven’t even been sent out yet to other devs.

I must have some real shitty risers then, because they refuse to do 3.0 in any of my boards including normal non mining boards, not only on the chipset's pcie ports, but also the cpu's x16 port

And for the lspci and idle thing, if you get 2,5 GT/s that could be it yeah, But im getting the 5 GT/s. And its while mining eth so the bus is used.

And yes some chipsets have shitty pcie that will make errors with just a standard network card, and its not only on mining boards, ive seen it on normal b250 boards also.

[vuli1]

I thiink on Z77 I had max Pci e 1.1 with Celeron and hd 6870. But on Z170 all RX are runing on 3.0 and 006 ver . eitherway, 1.0 is more than enough for mining cards. For Acron, not sure, didnt pay attention.

[Longsnowsm]

I think this looks very interesting, especially the 215+ stand alone mentioned on the web site.  I am going to hold off and hope you guys share your experiences and recommendations on the right combo for mobo etc for these rigs.  I would like to avoid some of the fumble farting around that I typically do on these things and just get something working without spending the next weeks and months trying to figure out how to make something work.  

Thanks to all you guys in this thread that are mentioning what you are building with and what you are thinking.  Looking forward to the reports on how this works.  I would love to see how this works in stand alone mode.

[ronnieb]

whats the cheapest board I can get away with with 4 rx 580s and a acorn.

[ronnieb]

Why not set up 32 Acorn CLE 215+s on a octominer and let them mine by themselves using little power.

The Octominer is completely starved of PCIE Lanes, which is precisely what you need to run lots of Acorns. 

The Acorn 215+ needs PCIE 2.0 x4 (or equivalent) speed to operate and ideally the GPU(s) it is paired with would have that speed as well.  Every mining specific motherboard that I am aware of limits you to 1x speed (either gen. 2.0 or 3.0), which is too slow to be as effective.

Take a few minutes to research PCI Express lanes, what the different generations are relative to each other speed wise.  Then research your motherboard(s) and see how many PCIE lanes it has (don't forget ones from the CPU too) and how are they routed on your motherboard.  You will find that the cheaper motherboards that we like to get as miners are poorly suited for Acorn acceleration.

All I have is cheap boards, looks like I may need an upgrade

[dragonmike]

Can anybody tell me if I could make good use of a 101 in the M.2 slot of my 7x-RX570 rigs powered by weaksauce Celerons 1620's?

[crazydane]

I changed a 2nd Asus PRIME Z270-A from Gen2 to Gen3.  This one is running Win10 and 7x 1080Tis.  Unfortunately one of the GPUs dropped out after the change:



That one GPU is not being assigned an Interrupt Line.  All 7 GPUs in this rig are identical and use identical risers.  Switched back to Gen2, and all 7 GPUs are working again.

I'll try this on another Win10 rig (identical this this one) to see if I run into the same issue.

[philipma1957]

I thiink on Z77 I had max Pci e 1.1 with Celeron and hd 6870. But on Z170 all RX are runing on 3.0 and 006 ver . eitherway, 1.0 is more than enough for mining cards. For Acron, not sure, didnt pay attention.

I have 4 or 5 z170 boards from biostar  https://www.amazon.com/Biostar-RACING-Z170GT7-Intel-Motherboard/dp/B06XDKHN1S


I think  that these boards  with the nest x2g  and  acorn 215+  will be true beasts

[TheYankeesWin!]

phil

if you buy 4x x2g nests = 400
if you buy  8x acorn 215+ = 2800

3200 to fill the board = $$

[philipma1957]

@ yankees  I have 12 1080ti's and 2 1050ti's

I don't need to fill the boards

I could do 2 nests per board and 2 1080ti's

[Marvell2]

the jury is out anyways on weather it will be better to buy next gen cards vs these fpga deals too many moving oarts imo, need new motherboards specific cpus etc , how bout i grab a gtx 1180 for $600 that hashes at stright up twice the hash rate of an 1080 and worry about fpga later when new 8 card boards with better features come out.

It makes no sense to replace hardware (mother boards and cpus) with new gear unless you are building a new system or app specific boards are released.

fuck risers

[bswilmington]

the jury is out anyways on weather it will be better to buy next gen cards vs these fpga deals too many moving oarts imo, need new motherboards specific cpus etc , how bout i grab a gtx 1180 for $600 that hashes at stright up twice the hash rate of an 1080 and worry about fpga later when new 8 card boards with better features come out.

It makes no sense to replace hardware (mother boards and cpus) with new gear unless you are building a new system or app specific boards are released.

fuck risers

where have you found the hashrates of the new GPUs?  I can't even find a release date

[Pennywis3]

the jury is out anyways on weather it will be better to buy next gen cards vs these fpga deals too many moving oarts imo, need new motherboards specific cpus etc , how bout i grab a gtx 1180 for $600 that hashes at stright up twice the hash rate of an 1080 and worry about fpga later when new 8 card boards with better features come out.

It makes no sense to replace hardware (mother boards and cpus) with new gear unless you are building a new system or app specific boards are released.

fuck risers

I seriously doubt you will get 1180 for 600$ 

[philipma1957]

MARVELL2 = correct on fuck risers

As to what  is best to do  buy  a shit ton of these acorns or wait.  Waiting is better.

Buying 2 and a nest  at 330+330+99 = 769 + shipping  is also okay

Buying  02 of these  with or without nests  is hard to do  it is a lot of coin up front  and   a clear picture  is not here.



Ie a 2 card nest does x hash at x watts  =  I don't know


If I knew a 2 card nest did  40mh of x16r at 60 watts I would like that.

as a 1080ti  does 20mh at 200 watts

[ZedZedNova]

MARVELL2 = correct on fuck risers

As to what  is best to do  buy  a shit ton of these acorns or wait.  Waiting is better.

Buying 2 and a nest  at 330+330+99 = 769 + shipping  is also okay

Buying  02 of these  with or without nests  is hard to do  it is a lot of coin up front  and   a clear picture  is not here.

Ie a 2 card nest does x hash at x watts  =  I don't know

If I knew a 2 card nest did  40mh of x16r at 60 watts I would like that.

as a 1080ti  does 20mh at 200 watts

From what I understand by reading the info on the squirrels web pages and reading the discord (before the message traffic went crazy, mostly with the same questions over and over), the CLE-215 and CLE-215+ consume up to ~15 watts each.

So your proposed scenario of 2 card nest would consume ~30 watts. The question then becomes one of what the hash rate of the CLE-215+, the only one so far that GPUHoarder and team say can mine stand-alone, would be on X16R. The stand-alone mining numbers have not been released yet, but should be coming Soon^TM.

When a CLE-215+ is used to boost a 1080Ti in a 1:1 configuration, you get a 76% improvement (1.76x) in hash rate on X16R, so the 20mh of the 1080Ti alone becomes ~35mh when boosted, and the combined total power is 216 Watts.

Additionally, it appears that using a CLE-215+ can allow dual mining. If I understand correctly, the CLE-215+ handles/shares the core heavy algorithm while the GPU handles the memory heavy algorithm.

[OurManInHavana]

With current low prices for the coins an Acorn can help with, all the scenarios I've looked at show that they'll pay for themselves in 12-18 months. Is that what other people are coming up with too?

Can't wait to see some real $/day improvements when they ship: so many GPU/MB combo's to try!

[dsmn]

With current low prices for the coins an Acorn can help with, all the scenarios I've looked at show that they'll pay for themselves in 12-18 months. Is that what other people are coming up with too?

Can't wait to see some real $/day improvements when they ship: so many GPU/MB combo's to try!

Hmm they do x2 x3 the speed on some algorithms if your gpu's are set up right. If thats not real $ per day improvements I dont know what is hehe, the big problem right now is that we are in a major reversal trend.