Discussion in 'Pyra News' started by EvilDragon, Jun 5, 2016.
640 k ought to be enough for anybody.
Right, because anybody really needs that much!
Thank you Mr Gates. Actually didn't he push for it to be even larger and it was going to be 512 at first?
Although this saying has been attributed to Bill Gates, Gates has denied ever saying it.
But they -would- be willing to state that a given CPU/SoC will have an optimal minimum/maximum range related to it's expected tasks and usage. In the case of the OMAP 5 Pyra, that has been proven, through actual usage and testing on the development, to be something less than 2GB.
--- Double Post Merged, Jun 7, 2016, Original Post Date: Jun 7, 2016 ---
The direct question was asked for me to spec a SERVER that cost LESS than a Pyra that would suit the stated tasks that were claimed to justify excessive RAM on a Pyra. I did so.
I am not stopping them from using a reasonable amount of RAM on a Pyra - or the insanely fast /SWAP capabilities afforded by the eMMC to do these additional tasks.
Why should their desire to do a set of tasks, that are not important enough for them to dedicate hardware to, that are clearly ill-suited to the mobile & pocket nature of the device, be allowed to upend production via scope creep at such a late stage? Why are people ignoring the actual, real world, rational testing performed by people, not me, who had/have access to actual developer hardware that say 2GB is plenty for this SoC? Why should the rest of us who are being reasonable and rational about the device and it's capability, continually having to defend reasonable, tested and existing specs, have to suffer through the resulting scope creep and additional costs driven by the irrational need for, "MOOOAAARR SPECS - MUST HAVE MOOOAAAARR!!"
When asked why they, "MUST HAVE MOAAARRR!", the response never fails to be an irrational, "need" based on tasks that are not particularly suited to a pocket sized mobile device. We, as a community, have been through this argument many, many times over the last few years.
More recently ED thought he found an apparent solution in double-sized same-footprint RAM chips to what was used in initial development. Now it turns out that the $30 price increase that he thought would cover the higher capacity chips would actually be more like $100 price increase.
So, now ED and his team are spending more time, researching more options, and needing to do more board population test runs at more expense and MUCH more time. Why? Because a few very loud and sadly ill informed individuals could not understand that the device simply does not need it due to Linux being far more RAM efficient, and cannot really take proper advantage of it because the OMAP 5 is simply not that kind of a CPU.
Now that units have been pre-sold with 4GB, though, ED now -has- to figure out a 4GB solution or go through a refund/reorder nightmare. This simply translates into more delays.
I think we can now place the 'blame' for a good 2-4 Months or more of delays squarely on the shoulders of those who irrationally demanded more RAM for a device that simply didn't need it and could perform it's targeted tasks (games and general purpose computing and mobile connectivity center) quite well with what was originally planned, tested and confirmed for the device.
Scope creep is VERY expensive.
--- Double Post Merged, Jun 7, 2016, Original Post Date: Jun 7, 2016 ---
Where do you get your numbers from? ED already stated that there's the possibility to get 2 2GB chips instead of 4 1GB ones, and the "good 2-4 months or more of delay" is simply ridiculous and you know it. With all of the parallel tasks going on in development and mandatory waiting times for things like case colours and keymat versions, I'd be surprised if this delayed anything in the long run.
And while I agree with you that most tasks should work fantastically with just 2GB, I personally ordered the 4GB version, just in case there are more RAM hungry tasks like virtual machines for Android that I want to try and because the price difference isn't that big, at least the expected one at that time wasn't. And I'd be surprised if that changed a lot.
If you make up arbitrary storage requirements, yeah. But if the 32GB internal storage is enough, then really just $600.
Would also cost more than $225/yr in electricity, if running it 24/7 so that you can remotely log into it from anywhere.
Might power that server for 12 minutes if you're lucky. Even less as the batteries age and need replacement after 2-3 years. Replacement batteries from the manufacturer cost 75% as much as the UPS itself.
The UPS itself wastes more power, as it must keep the batteries trickle-charged; adding more or bigger batteries to get more runtime makes that even worse.
Server $225 + shipping $53 + power $225/yr (+ cooling, if necessary) + UPS $109 = $612 in the first year alone.
Pyra is $600, power costs will be very low, it runs silently, and battery runtime should be a few hours without a UPS.
The obvious differences are that the server has better performance for heavier workloads, and ECC memory. And the Pyra has its own features that a server does not.
My own use case involves using the Pyra as a handheld by day, sometimes as a devboard, and leave it doing software builds overnight. For the latter, 4GB RAM should allow me to run more parallel jobs than 2GB (because I think the CPU on this is very good and would not be the main bottleneck). I expect it will perform better than the BeagleBoard X15 which has a similar CPU to the Pyra but only 2GB RAM. It may even perform as well as the ODROID-XU4 does in practice (the spec of which looks good on paper but for some reason they run kinda slow).
Actually, I think you're greatly misunderstanding basic project development math. Finding and understanding a failure counts in the calculation of delays when considering project scope creep.
For documentation of delay, read the first post in this thread, the ordering thread start, then impute.
To find out that the four 1GB chips were not available in the quantities needed at a suitable cost for the project would have required ED and his team TIME that they would otherwise have had for building drivers and other tasks on the existing prototypes. Time since announcing that they were going to do 4x1GB to now when they were found out to be unfeasible? 40 days.
The 2x2GB Samsung option is being -investigated- and will require acquiring samples AND test populating them on one or more boards. From his post I understand that investigation as just now beginning - and we're lucky that HNS put in the wiring for this alternate chip form factor. However, all of the testing time and any configuration that was needed for the previous RAM configuration gets chucked in this model - one or more weeks easily. 7 days.
So, now they're ordering samples of another chip component. Assuming that the supplier has them in-stock, between figuring out the component, ordering, shipping, receiving, call that a week. 7 days.
Once the component is in, it will still need to be installed and tested on one or more pre-prototype boards. This is insanely fine-detailed work done under dissection scopes with tweezers and pointers, carefully transported to an oven and heated for the solder to flow. A day to hand-populate heat and cool. Another day to check electrically if the placement was 'good enough'. If it isn't, start over with a new set of components and repeat. Then repeat, one at a time, for as many boards as are needed to do the initial evaluations. Assuming it works right the first, second and third times so they get 3 working examples to then test on - minimum of 5 days.
IF the hand populated boards prove out to work flawlessly, then it's time to move to the pick & place to do a test run to ensure that the new component with it's new size and pin-out gets set properly. Those machines only make money when they're working. Time for a test run would need to be scheduled into down-time between jobs at the pick & place company. Minimum of 2-3 weeks lead time on scheduling that visit. 14 - 21 days.
Run pre-production test production (prototype run) on the pick & place to generate 20 or so examples as they would be in production units. 1 day. Test these boards by hand. One person, 2 boards per day = 10 days.
Case & ship them to developers for testing to make sure that the code written for 4x512GB Pyras works on the 2x2GB Pyras. 2days + 1 week in the mail = 9 days.
Wait for reports from developers confirming function - or not. Two weeks. = 10 days.
Assuming tasks overlap of 35% and getting everything right the first time from here on out, the minimum delay from the demanded 4GB change is approximately 60 days. Maximum is unknown, but unlikely to be more than 120.
When RAM is exhausted on the Nokia N900, the latency of swapping to internal flash is *debilitating*. Basic web browsing can leave the phone interface so laggy it is difficult to answer a call. Its 256MB RAM is mostly what makes the hardware obsolete now.
If the Pyra plans to be useful for a long time to come, it should have plenty of RAM now for what *future* use cases might demand from it.
Well, I do need 4GB, and I wouldn't have pre-ordered at all if there wasn't a 4GB RAM option. My, and others' 4GB pre-orders (which you'll notice there are rather many of), help *you* get your device produced sooner, or perhaps even at all.
"Scope creep" here is also "versatility", making something that's useful to enough people that it's viable to actually produce.
If your only requirement is 32GB of flash, sitting around doing headless server chores can also be done by one of the HDMI stick computers for $60-$80. Doing those tasks with a Pyra is a waste. If you don't need the power of a server, but want to leave it sitting on a shelf turning bits, get a gum stick or Raspberry Pi. If you need a server, get a server.
I own two servers of similar spec to the above.
1U Supermicro Dual 6 core 2419EE Athlon with 32GB of RAM and four 2TB drives.
They cost me a combined $30 of electricity per month when on. I recently took them offline and compared bills. The hardware is getting re-purposed and re-combined into a single server with yet more disk. The 5 year old Cyberpower 800VA UPS held the PAIR of servers for 2 hours solid (disks spun down, light CPU load).
Actual tests done by actual devs on actual hardware that VERY closely resembles the Pyra have shown that the OMAP 5 becomes the bottleneck before 2GB of RAM usage.
Real world tests. Not based on belief. Right tools performing right job using components sized properly for each other. It does not have to be difficult or based on guessing, wishing, faith or a specs chase.
--- Double Post Merged, Jun 7, 2016, Original Post Date: Jun 7, 2016 ---
N900, using an OMAP 3430 SoC with 256MB of RAM and it's slow 2009 era eMMC being used as /SWAP.
Pyra, using an OMAP 5430 SoC with 2GB of RAM and it's roughly 4X+ faster 2015 eMMC being unneeded as /SWAP.
Yes, the N900s 256MB of RAM was limiting. They 2GB Pyra has 8 times as much, much faster RAM with an eMMC that should outperform the one in the N900 by a factor of 5-10.
What then, exactly, is your requirement for 4GB of RAM that cannot be done on a 2GB Pyra but is a suitable task for a mobile device?
If you, 'wouldn't have pre-ordered at all' without a 4GB option, then you clearly don't actually understand the Pyra or the hardware it contains. The OMAP 5 pairs very well to 2GB. Although 4GB or even 8GB are theoretical possibilities, they're not really all that beneficial as the RAM capacity outpaces the SoC's processing capability. Linux, as you are aware, is far more resource optimized than Windows - which removes the perceived necessity for large amounts of RAM on this device.
Had the loud and irrational 4GB scope creeps not pounded the podium and raised their voices, the vast majority of those 4GB orders would simply have purchased the 2GB version. So why did they buy the more expensive one? Because the more expensive units are assumed to better support the project. Because the more expensive units have a perceived aura of 'better'. Because of, "herd mentality".
And we're re-hashing, again, a topic that has been hashed and re-hashed repeatedly. But then you're relatively new to this and are unlikely to have read those past threads. Look particularly for posts by @TrashyMG detailing his actual application tests on an OMAP 5430 dev board.
Your insisting on delays required to get to 4GB units and purchasing a 4GB unit is not helping me get my device produced sooner or 'even at all'. It is, in fact, likely delaying it greatly in scope creep driven changes. If anything, it is the other way around. You see, I'm one of the 8 people who dropped $1700 (with shipping) in October 2015 to purchase a prototype and help fund the initial pre-run that showed you a working prototype. However, thank you for your order adding in with everyone else's to hopefully move production onward. Yes, every order does count. Your rather arrogant assumption that your 4GB order was in some way subsidizing mine is, frankly, rather amusing though.
It's not just a difference between base and not-base models, there is also the price difference which is quite small compared to the price of Pyra (expecially the 4G versions), so even if the improvement is small, it looks like it's going to be worth the cost.
Wow you must be right, I've totally ordered the wrong thing, I should cancel my pre-order and buy something else /s
I've no idea what resources Windows uses these days. But my Linux desktop is idling right now with Firefox using 1.27GB _resident_ RAM, leaving not much RAM to do anything else. Or am I also not allowed to use the Pyra to browse the web with many tabs open or multitask?
I think the overlap is much higher and, as I said before, there still are a lot of other things like fine-tuning the cases and waiting for new ones to arrive in Ingolstadt for testing going on that demand a lot of waiting. While ED somehow gets his schedule always filled up, I'd be surprised if he wasn't able to do most of the work in between waiting for other comopanies. But I have to admit, I forgot about the long waiting times at the pick-and-place machines, so I stand corrected on the claim it wouldn't delay anything in the long run. But I still don't think it will be as significant as you say.
I even go so far as to say that ED wouldn't have offered this if he knew he couldn't do it in a moderate timeframe. I think he knew the possible implications of offering a device with 4GB of RAM that wasn't completely tested at that time. Please correct me if I'm wrong.
Please read the statement you're replying to, especially if your reply is "don't get the device this forum is about":
Herd mentality!? You don't have any use for more than 2GB of RAM, so you assume that anyone who could have is irrational? This should be a open platform, not something tailored to you specifically. The fact you could even decide if you wanted a 2GB or 4GB version is fantastic, and you now talk about it as if it was the stupidest decision made by ED and as if it was made only to cater to irrational people.
Yes, his testing were great and showed that typical use rarely gets the RAM filled. But the Pyra is an open and niche product and everyone uses it differently. With the mobile version, I wouldn't be surprised if using Android VMs become normal, and modern Android can be a RAM hog. Also, it's OMAP5432. The *0 version uses LPDDR2 RAM instead of DDR3.
All in all, I think ED knew what problems offering a 4GB version could bring with it and wouldn't have done it if he thought he couldn't handle it.
Oh boy. Seems ED was right. It is going to be a colorful week...
Come on. If you see 2GB for 715 and 4GB for 745 what are you going to get ? Even if you don't need them how much can they hurt ? (yeah, battery)
It is not surprising that people ask for 4 GB given the option, what I find surprising is that the price of the memory chips is not known at the time the
price is offered. But hey, nobody is perfect, prices change, etc.
Many of you here like old computer games, right ?
Do you like old computer books too?
Back in 1984 there was "Micromania: The Whole Truth about Home Computers" by Charles Platt.
You'll find there the "Sixth Law of Computers":
Computer memory and male genitalia have one thing in common: everyone says size isn't important, but no one quite believes it.
Even if you just glance over
some may have chosen a GPD Win with 4GB RAM over a Pyra with only 2GB. Rational or not, it could lead to fewer sales, and the Pyra needs to ship a certain number to be profitable to produce at all.
Pyra wins by versatility, with every use-case it can cater for. If it's a mobile handheld *and* could be a desktop or living room computer when I come home, then it's more valuable. If it's easy to attach extra storage, that could make buying a separate NAS unnecessary - making the Pyra worth paying more money for, etc.
I've mentioned it before, but I paid for a 2GB GSM model because that 30EUR saving is 30% off a predicted price of a new CPU board down the line, which I could buy with more ram once my usage is better understood. I'm hoping that for the time being before a replacement is available, software won't be common that needs much more. In the worst case I can always buy the same board again with 4GB RAM, if it looks like something better isn't going to come, although that will end up costing me around 60 EUR more, so it's all a bit of a gamble.
I suspect ED saw that a 1GB part-to-part replacement for the 512MB RAM chips existed and may have spot-checked the prices for evaluation samples and likely assumed that the quantity markup would be proportional - and has since found out that is not the case.
So, the choice is either substantially more expensive (for the 4x1GB) or somewhat more complicated (2x2GB Samsung - untested) than he'd thought. He can't easily hit the 4GB purchasers up for another 50+EUR to cover the difference. So, logically he's pursuing the untested 2x2GB Samsung option - which luckily won't be nearly as bad as HNS had included traces and pads for that contingency. ED is following a very logical path in this and I'm confident the project will get through it. However, I'm also a realist and those changes require time and money to implement.
Still, the entire situation could have simply been avoided by moving forward with 2GB RAM for the first release then coming out with a 'maxed out' version of the OMAP 5432 (Thank you @Agricola for the correction - I keep wishing that we had the 5430) in six moths to a year or so as a short-run limited edition upgrade board with maximum RAM (8 GB?) & maximum eMMC (256 GB?) sizes and a premium board price (250 EUR?).
Something intriguing would be to work in true parallel - since there are so few of them, move forward with the 4x512MB chips = 2GB 2nd round prototype boards with the 32GB eMMC to fulfill the prototype orders and strap on the few 2GB orders to that run - then close out the 2GB as a purchasable option. Meanwhile do the tests of the Samsung modules and test for producing the 4GB units as the 'primary production run'. It would mean that all of the 2GB orders would more or less end up with 'prototype' SoC boards (not cases), but it might make production sense.
Just a quick clarification: of course everything works parallel.
We ordered the samples a while ago while working on other things.
Nikolaus picked up the sample RAM at GC yesterday and already populated the CPU boards with them today, so tomorrow he can test if that all worked fine.
So if all goes well, we will know this week whether the Samsung Chips work fine.
This isn't a delay of 40 days, we had a lot of other things to work on in parallel, like changes to the charger circuit (so both USB ports can be used for charging), etc.
That is great news! Thank you!
Separate names with a comma.