Analyzing 4GB RAM


EvilDragon

Administrator
Staff member
Joined
Mar 4, 2003
Messages
29,986
Age
46
Location
Ingolstadt
Did you miss me?
I have to admit - I had some annoying issues happen in my "normal" life (bike stolen, smartphone broken, etc.), so I was pretty much absent from the boards and also twitter.

That should now change again!
I'll try to resurrect my Twitter with even more Micro-Pyra-News than before, and of course, have more information for you more often here on the boards as well.

Okay, time to start with the news...


1. Remaining work for the Mainboard

After Nikolaus came back from holidays, he resumed working on the last fixes for the mainboard.
He will modify an existing one with all the proposed fixes and will test them right away.
Most of them have already been implemented and quick tests assume these work fine.
He'll continue to work on those during this week and I'll let you know about the results (quick news via Twitter this time).

The mainboard dummy for measuring and testing the Wifi antenna is already in the hands of the RF expert.
He didn't give us a date yet when he'll find the time to do that, but he tries to work on it as soon as possible.

So the mainboard is almost ready now.


2. The 4GB RAM issue

Well, I had hoped to tell you that's all fixed now.
That's on top of my personal wishlist, as this is the issue that really pulls me down...
It's really demotivating, but don't worry: The last thing I'll do is give up!

We managed to resurrect the Pandora, so we'll manage to fix this as well!

And we have some good ideas what the issue is.
We've tried quite a few things in the last week, and here is what we know:

Tony's Pyra works with 4GB RAM, if he powers it with at least 1.39V or if he disables DLL (then it also works with 1.35V)

Usually, it should work worse with DLL disabled - so this is a bit weird.
He provided patches - but those didn't work for others (it's still unstable for them).

However, he uses his own compiled U-Boot and bootloader - so it could very well be he's got some other different settings we didn't think of.
To confirm whether that's the case or not, I'll try his compiled binaries on multiple Pyra 4GB RAM boards later tonight. I'll make sure to update you here on the boards and on Twitter about the result!

But why does the 2GB RAM board work whereas the 4GB might not?
There's an easy answer:
Juliano simulated the noise and voltage drops of the VDDR3L ground plane (which is the power supply of the RAM chips).
The noise is too high compared to the allowed value in the datasheet and the voltage drops too much on some balls.
This is not necessarily a problem, as all chips have tolerances and the values provided in datasheets ENSURE that everything will work fine, so they have even higher tolerances here.

The 2GB RAM chips are so-called single-rank chips, which means they consist of four 512MB chips.
The 4GB RAM chips are dual-rank chips: each chip has one 512MB chip stacked on top of a 512MB chip - so these are basically eight 512MB chips.

These, of course, need more power than the 2GB RAM version.
All 2GB RAM boards work reliable.
As mentioned, Tony's Pyra works with 4GB RAM, which means: we're not THAT far off!

This means that the CPU boards probably just need a BIT of enhancement (by using capacitors, some more vias for the VDDR3L ground plane and other changes).

This is something Nikolaus needs to check: What is possible in our already pretty much cramped setup?
Or maybe we CAN get the 4GB RAM to work without a hardware fix - that's something we'll know tonight when I tested the boards!

There's also nother possible solution:
Nikolaus has reworked two CPU boards. One with different 2GB RAM chips (for comparison), one with different 4GB RAM chips.
If these 4GB RAM chips have a higher tolerance, they simply work. More information about that when this has been properly tested.

Well, let's assume for now the 4GB RAM needs a hardware change on the CPU board:

  • a) Ask all preorderers whether they would be okay switching to 2GB versions. If at least 400 want to do that, we can produce those with the boards we already have lying at GC while we can work on improving the design and produce 4GB later on. As a little compensation, I'd offer the new 4GB boards to everyone who agreed to get the 2GB version for production cost price. We still plan to produce a stationary motherboard - so you could use the old CPU board to power this one.

  • b) Sell the existing bare CPU boards as keychains (17 EUR each) and improve the design. This would cover the costs for the already produced CPU boards and we could deliver 4GB RAM boards to everyone.

Of course, Option a) is preferred (even though it means a lot more work for me) and production will start earlier, but if not enough people are willing to switch to 2GB, Option b) would be the better one.

The only real bummer would be if we can't find the space to fit more capacitors onto the CPU board... in this case we'd have to try to sell the 4GB Memory chips we already have on stock (worth a whopping 23,000 EUR), but we hope that won't happen.

I guess you can imagine how much that issue is weighing on my mind. But there are multiple options for us, it's not a dead-end for the Pyra.
We can also get help from a company that the memory distributor suggested to us. They have the possibility to do live-measurements with the memory while the system is running. That company isn't located too far from us, so in worst case, we can find out more details what's going on with their help.
That would cost quite a bit though, so let's first see if Nikolaus can improve the design enough. We can easily test that thanks to Julianos simulations!

Speaking of costs...


3. What's the financial situation?

If you followed the newsposts, you probably can imagine that these issues have added quite a lot to the development costs. And yes, these have cost more than I ever had planned.
However, we don't have any financial problems - because my shop is growing more and more popular and the profit has increased quite a lot - which is great!
I'm still keeping my wage as low as possible (I'm earning 690 EUR per month, in case you're interested) to make sure as much money as possible is available for the Pyra.

The production of the first 550 Pyras (first production run) has already been fully paid. This includes the full PCB production including all needed parts.
We already own 3000 batteries and 1500 LCDs as well as 10,000 nubs.

We still have more 100k EUR on our bank account (which is more than enough to start the production of the next 550 Pyras) and the shop is making a profit of about 8000 - 10,000 EUR per month.

So yes, unless there's a sudden mass cancellation (which I certainly don't hope!), the development and production is safe.

We won't give up as long as you don't give up on us! Promise!


4. What's the status on the cases?

I'm getting more information on those tomorrow.
My contact will be visiting the companies in Greece and checks on the status.

So yeah, expect another newspost within the next two days :)



Well, I hope todays newspost didn't disappoint you too much.
I know you had hoped the 4GB RAM are working now - I had hoped the same.
But the most important thing is that we've not hit a wall but have different options to proceed.

We can do it! And yes, if ANYONE wants the Pyra to be finished as soon as possible, it's me (because it's all really stressful).

Thanks as usual for your support.
Now off to test the 4GB RAM boards I've got here. I'll report back later!
 
Thanks for the update, and best wishes for the testing.

The news post isn’t displaying correctly on Tapatalk on my end for some reason, so I had to open it in web view to see the whole thing. For me, it starts with the real bummer paragraph.
 
Did you miss me?

I'm still keeping my wage as low as possible (I'm earning 690 EUR per month, in case you're interested) to make sure as much money as possible is available for the Pyra.

I really wish Pyra production will start soon. Not to get my pre-pre-order one but to start cash flowing into your pockets. You deserve it! Thank you for the news.
 
I just got the 4GB version because I could.
If 2GB is sufficient, I just as easily switch my order to that if that's convenient. But yeah, let's see what some more or heavier caps can do.
It would be a shame to get things measured for a lot of money and then get told exactly that. "Sorry sir, She just needs to be able to draw a little higher current."
 
Last edited:
OK -- if you don't stop the development on the Pyra, switch me to the 2Gig Model. I hope for the 4Gig some time into the future ... but will go with whatever get's you into "normal" mode sooner then later

so for me it would be a) ...

peb
HP200lx admin
 
I never switched from 2GB to 4GB, you're telling me that this cozy 2GB pre-order boat is going to get more people?
Was kind of starting to like it but... Fine by me, I guess. It does sound like getting the RAM fixed might be possible, still. Hopefully, maybe? :/
 
Quick update (as promised):
Tonys binaries didn't work on my PCBs as well, U-Boot complained about faulty DRAM. I have one more idea (combining both of Tonys working solutions: 1.5V and DLL off, and if that works, try lower voltage values).
 
Thanks for the news! I'm subscribed to your twitter feed (shop and personal) and am trying to keep up to date with the Kernel mailing list threads, so I'm never going to complain about getting more news!

I also pre-ordered a 2GB model over a 4GB one, because I'm planning to get a new CPU board when they come out and I'll have a better idea if there any use for that extra 2GB of RAM by then; I suspect at present there will not be.
 
If this was a livestream i would be centimetres away from the screen.

Still up and running at the desktop. Need to run memtester at some time though.
And 1.5V is not really an option, but a test...

However: If that works, it probably means that our voltage drop is too high but the noise is not that much of an issue.
That should be an easier fix.
 
As a little compensation, I'd offer the new 4GB boards to everyone who agreed to get the 2GB version for production cost price.

Is there a rough estimate for that production cost? Just to get an idea, really.

Also, thank you for persevering through adversity, and for being so open with the accounting and finances involved. Your blend of remarkable project management and transparent practices truly is one of a kind.
 
Back
Top