Ram & eMMC costs / power drain

[ible]

don't be too dramatic. things will work out.

 

[rygD]

Thanks for adding the list of why various stuff wasn't included. I will try to edit the wiki myself in the future. I like that some of what I said was cleaned up so that normal people can make sense of it, and it sounds more professional. Maybe we could do that with the rest of what I said.

I will try to find some time to check out the wiki this weekend.

 

[elvissteinjr]

The time for those discussions passed many months ago.

The initial questions were asked many months ago, ending there with "probably 2GB, but let's wait until before parts will be ordered". I can see that everything points at 2 right now, but if they're pin-compatible (are they?) what would be wrong then?

I personally don't care much about the results of this, but I'm genuinely wondering why you're so concerned about people discussing things so far as to keep telling people to shut up in a public forum.
Now forgive me for this one, as it could probably nicer worded but:
Who are you to tell us not to talk about this? Don't you trust ED enough to do the right thing™ even under the evil evil influences from here?

Regardless of anything, the curiosity about the actual power drain beyond spec sheet maths remains. And there's nothing wrong with that.

 

[TrashyMG]

I can see that everything points at 2 right now, but if they're pin-compatible (are they?) what would be wrong then?

Power consumption and perhaps secondly performance is what most people are concerned with. With more capacity comes more current drain, also generally you need to run the memory at a higher clock speed on larger capacity memories to get the same performance as the lower capacity memory.

To me the concern is just a marketability thing, as more looks better to potential customers, but I know that 2GB is a decent amount of RAM for an ARM Native Linux solution. I guess it's just a matter if marketability is worth the potential drawbacks or weather the drawbacks are even that big of a deal... Real tests would be nice.

 

[comradekingu]

Since when do you need higher clockspeed to get the same performance as lower density memory?

Did you mean add more voltage to maintain speed?

Adding more memory at the same speed (which adds cost/infeasibility as you go big) is a net benefit to performance as you can buffer more things, and don't have to juggle memory.

For what 4GB costs, it adds a tremendous amount of perceived value.

 

[TrashyMG]

Since when do you need higher clockspeed to get the same performance as lower density memory?

There is generally is a latency penalty with DRAM when increasing the size of the memory array, it's just inherent in how DRAM works. I do have to say there may not be that noticeable of a latency issue between 2 and 4GB of RAM.

 

[comradekingu]

Oh, but its also possible to buy memory with stricter timings. It is a matter of cost regardless of what total amount you are at. Cost scales with capacity though.

 

[Exophase]

There is generally is a latency penalty with DRAM when increasing the size of the memory array, it's just inherent in how DRAM works. I do have to say there may not be that noticeable of a latency issue between 2 and 4GB of RAM.

More DRAM in this case probably means more DRAM chips and more chip selects to address them, either inside a packaged DRAM module or DIMM or something or between multiple ones and handled by the SoC.

You need somewhat longer traces to accommodate this and that adds access time but I think it must be very minor compared to the core DRAM read operation and all the stuff that goes on in the memory controller. AFAIK rule of thumb is somewhere in the neighborhood of 70 ps transmission time per cm of copper. So it's very small compared to the tens or even hundreds of ns that a full memory access can take. Hence why PoP memory never materialized a latency benefit despite being right on top of the SoC.

There's also maybe some extra logic time muxing the added chip selects but I think that'd also be comparatively small.

I doubt the memory controller even tries to accommodate these things, it probably just assumes some safe maximum and the SoC layout requirements dictate trace lengths not exceed some amount.

 

[ible]

just as confirmation, ED says 2GB.
(also, it wasn't asterisked anymore on the pyra technical specifications page...)

 

[Mr_Loon]

Got an answer from ED relating to Ram costs, hope to hear soon about eMMC :

Well, RAM won't change, price is about 15 EUR for the 2GB, for the eMMC we're now checking the prices, and especially the MOQ (maybe we can offer different sized versions).

Added a few supplementary questions to the OP :

5) What is the base size for Pyra Ram? i.e. 1 x 2GB or 2 x 1GB etc.?

6) Is it possible to disable ram to save battery ?

7) What is swap performance like when system Ram use is 100%+?

8) How difficult would it be to make a 'Premium' Pyra with 4GB of Ram (& possibly a larger eMMC size?)?

 

[rygD]

While it is known that I support the idea of considering more RAM, maybe we should save that for an upgrade. We can use this to push ED to get on that quicker if we hold back for now. (Hmmm, no devil smiley.)

 

[Mr_Loon]

Had an update from ED in this thread about eMMC costs :

4GB : €3-€4
8GB : €3-€4
16GB : €6-€7
32GB : €10
64GB : €20

Updated OP with this info.

 

[Grench]

Had an update from ED in this thread about eMMC costs :

4GB : €3-€4
8GB : €3-€4
16GB : €6-€7
32GB : €10
64GB : €20

Updated OP with this info.

Because clearly discussing this in one thread is never enough...

Again, you're quoting wholesale part costs. By the time all factors get added in, the -retail- price winds up being closer to 250% of those part cost differences. That 64GB part may cost 20EUR per on a reel of 1600 units wholesale. Short board production runs will have higher failure rates than large scale production. So, assume 5% or so failure rate that, with the costs of testing to find it, adds about 10% to the component costs of the board. ED has to mark up the end units - likely more than you would see as 'fair' in order to cover the costs that you're not counting for maintaining stock, returns & repairs, advertising and profit.

By the time it's 'done' that 20EUR part's actual cost to the end user winds up being closer to 50EUR or 10% of the total estimated retail of 500EUR. The 10EUR part, though, would wind up being closer to 25EUR or 5% of the total estimated retail of 500EUR. The 6,5EUR part would be about 15EUR or 3%. I'm not sure what ED estimated for the component cost proportion for the eMMC within the overall unit, but something tells me it's more on the 3-5% side of total unit costs than the 10% range.

The net of it is, though, that these wholesale prices you're throwing around are interesting, but are not remotely reflective of how much the end user would end up paying to get it - and there seems to be a psychological barrier at the ~500EUR price point.

"But but I want a premium version that has 8GB of RAM and 128GB of eMMC and the wholesale costs for those parts only adds 50EUR and..." Any time you split a product stream it adds costs. Those pick & place machines have to be stopped, a different reel loaded, different parameters set, a few boards for testing placement produced and measured - the labor costs of just changing one component can really add up. Also, the prices you are posting are wholesale with 1600+/- units on-reel prices broken down to per piece price. That Minimum Order Quantity of 1600 at 20EUR each is 32000EUR up front. The minimum order quantity would need to be in the 1500+ unit range of 'premium' units. Until or unless the Pyra's 'normal' unit sells in quantities that can justify additional diversification (beyond the 4G modules that are already running into MOQ questions), talking about making a 'better' version cannot do anything but hurt sales.

So, these messages you keep adding asking for a 'premium version' just for you wind up doing far more harm than good. It implies that the production unit is somehow inferior to what it could/should/would/will be. Trust ED. He will make it as good as it can be within the limitations of production costs, retail target pricing and community needs.

Get the Pyra produced, get it some sales. Then in a year or two maybe a souped up processor board can be made IF sales quantities can support it - which would still have to overcome the MOQ of ~1500 units +/-.

 

[SNESFAN]

Wow

 

[Grench]

Wow

Too much reality?

 

[SNESFAN]

Too much something for sure

 

[levi]

Maybe I've missed something, but I thought the point of this thread was to determine the cost/power tradeoff of different RAM and eMMC solutions for all of us. I've not seen MrLoon suggesting he'd like to pay the extra for a short run of super-Pyra boards. Kingu on the other hand has, in the other thread, though he does seem willing to pay a distinct premium for the advantage.

 

[FaeMinx]

Too much something for sure

Have you forgotten where you are?

 

[Grench]

Maybe I've missed something, but I thought the point of this thread was to determine the cost/power tradeoff of different RAM and eMMC solutions for all of us. I've not seen MrLoon suggesting he'd like to pay the extra for a short run of super-Pyra boards. Kingu on the other hand has, in the other thread, though he does seem willing to pay a distinct premium for the advantage.

8) How difficult would it be to make a 'Premium' Pyra with 4GB of Ram (& possibly a larger eMMC size?)?

 

[Mr_Loon]

Because clearly discussing this in one thread is never enough...

IMHO adding the answer to a question asked in the original post seems reasonable, to each their own....

Again, you're quoting wholesale part costs.

I'm quoting the costs given by ED when asked the question 'what do the parts cost?', I assume they are wholesale costs, it would be more useful to this discussion if ED had answered with a cost based on final retail price, but he chose not to (understandably so). Feel free to speculate on what those costs / impacts may be, your figures seem fairly ridiculous to me for many reasons, forgive me if you actually know what you are talking about.

EDIT : From someone who does know what they are talking about, from ED in the 'still alive' thread

Maximum failure rate for us is 5%, everything else is covered by GC (though it will probably be a lot less).

This means my calculation is based on the fact that we pay for 1000 but only receive 950 working ones.

So, in case of the eMMC that means:
8GB = 4 EUR = 4000 EUR for 950 = 4,21 EUR
16 GB = 7 EUR = 7000 EUR for 950 = 7,36 EUR
32 GB = 10 EUR = 10000 EUR for 950 = 10,52 EUR

If I keep he planned margin of 15%, then the prices would be:
8 GB: 4,84 EUR
16 GB: 8,46 EUR
32 GB: 12,10 EUR

So using 32GB eMMC instead of 8GB will increase the end price by 7,26 EUR.

"But but I want a premium version that has 8GB of RAM and 128GB of eMMC and the wholesale costs for those parts only adds 50EUR and..."

Not quite, the difference between asking for something and asking about something may be a little too subtle, let me try and explain.

I'd like to see an open discussion about the possibility of a premium version with 4GB of Ram and 64GB eMMC, the wholesale costs add up to €25.

Before I suggested this was a good idea there are few things I'd need to know.

• Is there demand for a premium version?
• What would be the impact on retail price?
• Would people be willing to pay this increased price?
• How much complexity would it add to the manufacturing / sales / support / repair process?
• Is ED able / willing to handle this added complexity?

Why I think it might (depending on answers to above) be a good idea :

• I suspect a lot of people would be willing to pay a premium price.
• Fast storage & ram add good value to a computer system.
• If demand & willingness to pay a premium price was sufficiently high then the cost of the base unit could be lowered, thus hopefully unleashing more Pyras into the wild. Pretty sure that's how it works with a lot of electronic products.