Topic: Acorn M.2 FPGA based GPU Accelerator - page 28. (Read 73507 times)

Yah I don't think it has anything to do with usb ports on the mobo.  As mentioned higher, I'm sure the reason is due to the increased cost of shoving 4 m.2 ports and 4 usb ports on the same card!

Why would you do that when you can just get one for 30 bucks?    https://www.amazon.com/Inateck-Superspeed-Ports-PCI-Expansion/dp/B00FPIMICA/ref=zg_bs_229185_2?_encoding=UTF8&psc=1&refRID=T195ZQ898YCA5X65QFM1

Hello there,

What is the Discord channel for this forum?

Thanks.

This makes no sense to me. How is the number of USB ports on the mobo relevant? I thought this was all about pcie lanes and x8/x16 connections.

We're still waiting to hear a response to our inquiry to them. Over a week ago now.

They don’t serve the same function.

The 2 card works well with boards that don’t have lots of USB ports.


The four card is designed for boards that have lots of ports.


The 2 card is really good for threadripper boards which are short ports.

The four card is good for 1 full slot board with 13 ports.

I have a threadripper 1920 board it can do only five cards.

Add 1 two port jumps me to eight cards.
Add 2 two port jumps me to eleven cards.

And the four port does not help with this board.

I guess if you have a threadripper and the four card added lots of ports you may want it over the two port.
So I can see you complaint.

Thanks this is helpful. I would probably pay an extra $100 for the 4x USB ports on the X4, but oh well.

Anyone received a shipment notification/tracking info?

My info comes from lurking on discord. The reason the X2 has USB ports and the X4 does not is the expense of chip required to support 4 M.2 + 4 USB ports is too high. I read on discord it was like $100 for the chip alone that would be required. Thus the USB ports did not make it into the X4.

My understanding is that the X2 and X4 nests serve the same basic function, i.e. hosting multiple Acorns on a single x8/x16 PCIe. So I don't understand why the X2 has USB connectors but the X4 doesn't. Wouldn't it be just as helpful to connect your X4 nest via 4x USBs as it would for the X2?

Posted ~10 minutes ago on the FPGA/#sqrl-products discord channel

  
 

 funny 2018

Do you happen to have a link.. i am guessing my search techniques are getting old.. the cheapest i can find currently is ~150 after a mail in rebate..

in which  year?

Yeah you will need more than x1 for the Acorns I'm pretty sure of that.  I find it hard to buy an Acorn, when a 470 8G MSI model is on sale on newegg at 95$ (seems to be new...but who knows). 

Latest info on acorns from discord



Lokar -=BoA=-Today at 5:38 PM
Latest update on acorns is currently going through final testing and things are looking good for monday ship.


that is monday the 5th of Nov

Thanks, this is very helpful. I wonder how efficiently the acorn can divide its compute between two gpus. For example, boost two gpus at 15 MHS each, thus avoiding the x1 bandwidth bottleneck.

It also seems like I could avoid the bandwidth issue by switching to e.g. raven coin.

Note that if you're going to be mining eth, you may be limiting the performance due to the limited bandwidth of the x1 PCIe risers where you have the GPUs. For the ethash algorithm, the acorns are used to perform the two keccak hashes which occur before and after the DAG sampling loop, ie. the process to perform one ethash result would be:

1) acorn performs the initial keccak hash on the block header plus nonce (ie. 40 bytes initial input)
2) 64 byte result of this initial hash is sent to the GPU over PCIe (ie. acorn writes the result into the GPU memory)
3) GPU performs DAG sampling loop to mix with hash result
4) GPU sends 96 bytes of DAG loop mixing reduced/compressed result to acorn over PCIe (ie. acorn reads the result from the GPU memory)
5) acorn performs the final keccak hash

If you look at just step #2, the 64 bytes must be sent over the PCIe lane connected to the GPU. With a x1 riser, you effectively have a bandwidth of a single x1 PCIe 3.0 lane, which would be just under 1 GB/s bandwidth.

So if you're sending 64 bytes per hash being performed over a 1 GB/s PCIe lane, then that bandwidth can effectively support only about 16 MH/s hashrate.

You really need a x2 (or higher) PCIe 3.0 connection to the GPU so the bandwidth can support around 32 MH/s (or higher) and won't throttle the performance for ethash. Considering your particular cards you might even use x4 lanes.

(The info above came from the various livestream interviews GPUHoarder held with a couple of Youtube channels a few months ago. Best to check with SQRL directly in case the config has changed for ethash use.)

I mined ethereum with a single GPU (MSI R7-370 4GB) rig that earned me 16 coins before it became nothing more than a noisy heater. The rig has morphed and some of the coin it mined has been put into the current iteration. I used it to learn about gnu mining and mess with new gear. This rig is the one that will get the Nest X4 with CLE-215+s added to it. Will it ROI in 3 months? Unlikely. in 12 months? Dunno. That will depend on what coins I mine and what the market thinks of them.

I also bought ethereum, but not at $1000.

so, numbers yet?

wait for real testing.

i do think the 4 slot nest is correct.

but you may be better  getting 2x 215+   then another 2x 215+ a month later

without real world tests simply don't buy yet.