Topic: ARM processors with Mali Octacore - ignored by mining coders, why? - page 2. (Read 1575 times)

 Closer to 200 to match a Vega 56 - mine is pulling about 1960 hash with very little work attempted at tuning it (I HATE WATTMAN!), and others have reported higher.

What facts you've presented ZERO Facts! while I've told you I can show every review of RX Vegas  on the internet Gaming / Productivity / AND Mining even taking into consideration the lowest power consuming vega of them all mining a Vega with it system CPU at Load you get 300W+, if you change the power limite to 70% downclock the gpu and with an Intel dual core pentium doing nothing you might get the figure you are mention, a Vega 64 don't even dream about it.
 
Again what math are you even using Serisouly it's just a simple division !  and what aload of misinformation you are giving and I'm being nice.
Here bellow a fully optimised Vega 56 rig !


The Vega rig is consuming almost a 800W or a 1000W I forgot which which !! and this is a fully optimised rig

Here is another one 3xRX Vega 56 @ 600W without the CPU Mining ! (no BIOS Mod tho)

I mean the math is easy 6000w/350w = 17 cards unless you are really downclocking and setting a very low power limite unless you are not mining with CPU unless you are using 6 Cards or more per rig unless you are using platinum power supplies you won't increase the efficiency, if you do tho all this ! and with only Vega 56 No Vega 64 you might reach an optimal 180w per card or so! and then you might get to the 30 cards, this is A LOT of IFs


What tiny display ? it's not about the size it's about the computational power, most of modern smartphones have 1440p screens heck some have 4K screen and they run them no problem, also those gpus can power normal display via HDMI or WDMI no problem ! and yes the Adreno GPU on the Snapdragon runs OpenCL the full OpenCL 2.0  while the X1 from Nvidia runs Cuda actually it has 256 Cuda cores
Proof
https://www.qualcomm.com/news/releases/2017/01/03/qualcomm-snapdragon-835-mobile-platform-power-next-generation-immersive
https://en.wikipedia.org/wiki/Adreno
http://www.nvidia.com/object/tegra-x1-processor.html



FP 64? No no one compares cards on FP 64 Unless they have very specific use for them, it's either 32 full precision or half precision 16, if the card is not BUILT and optimized for double precision it sucks at it
a good read for you my friend : https://arrayfire.com/explaining-fp64-performance-on-gpus/
If you think floating point calculation are a good measurement single and nowadays half point calculation, and please feel free to prove me wrong with some counter examples as long as the cards are not bottleneck-ed with something else such as VRAM capacity or speed, You are not going to compare for example an HD 6990 vs RX450 that doesn't make sense at all.




I agree totally. Nothing to add here, aside the fact the 1080 TI or even the 1080 are not good example because of the latency issues with the GDDR5X memory

My "less than 300 watts" IS THE ENTIRE SYSTEM as I have already stated that's "AT THE WALL" on a system that has NOT been optimised for power consumption.
That figure is NOT "only considering the GPU load".
It's also a very NON-OPTIMAL setup in general - ONE non-optimized GPU in an old high-power-draw SINGLE CARD system.
It would be nice if you would pay attention and quit trying to prop up your strawman "THE GPU eats 250+ watts to mine with" argument with lies while ignoring presented FACTS.

The FACT is that 6 KW will power a lot closer to 30 Vega GPUs including total rig consumption - and possibly MORE than that - than your TOTALLY BS 17 claim.
If that "130 watt" TDP figure I've seen claimed on one "how to build it" site is correct, that would put a 4 card Vega rig right close to 600 watts - for *40* Vega GPUs in 6 KW including total system draw - while still respecting the widely reported issue of trying to get more than 4 Vegas mining Monero in a single system at high hashrates.

Does the 835 have a GPU that's useable for anything other than driving a tiny display?
Does it support OpenCL or CUDA?
If not, it IS useless for mining.

Comparing floating point capabilities is MEANINGLESS for cryptocoin work, as floating point does NOTHING to help cryptocoin work in any way shape or form.
There is a reason any RX 470 blows away the R9 280x on most cryptocoin work, even though the R9 280x does more than DOUBLE the FP64 work of any RX 470 - Floating Point is MEANINGLESS.
Again you insist on ignoring FACTS to try to prop up a strawman argument.


 Nvidia cards in general are more efficient - but there are exceptions, they're a tossup at best on ETH mining, they just aren't competative at all in Monero mining where Vega blows away the hashrate of ANY Nvidia card by a wide margin and has better hash/watt when both sides are well optimized, and outside of cryptocoin MINING as such the Vega 56 can very close to MATCH the keyrate of the GTX 1080 ti on Distributed.net RC5-72 work while soaking a little less power (much better on keyrate/$ when you can find the Vega 56 at close to MSRP, tossup on keyrate/watt).

 In many cases though, even where Nvidia cards are more efficient, they are NOT "much more efficient".

Everyone knows that Nvidia cards are much more efficient right now.
Your 300W load is only considering the GPU load. And like I said before even if you consider such figure You cannot reach the 30 you've mentioned that's simple math.
20Hash on monero is totaly possible an 835 in theory, neither the 835 or the X1 have 2MB of cache per core they have 2MB per 4 cores. but you are not factoring the GPU Side of thing
Like I said before the cost from random supplier is expensive, but if you can build your own board it can be interesting. It is hard to justify such a project considering that not only the miners and software is not efficient

as for  
There are always different version of these SoC for different markets for example when Nvidia or Qualcomm produce a SoC for embeded machines these SoC do not have said functionalities, and Again your forget each time about the GPU side of these SoC please check out the SDA835 (APQ8098) for example ! as for floating point calculation it's a correct way of comparing hardware especially when it comes to different platforms, since it doesn't not require platform optimisation and whatsnot.

Gaming reviews have NOTHING to do with mining usage.
Try looking at the various build-up reviews for the Vega FOR MONERO MINING USAGE.

For a specific related example, most of the reviews I have seen put "high load" power usage of a GTX 1070 ti in the 180-220 watt range - yet the MOST EFFICIENT mining point for those cards is 106 watts give or take a couple, and even folks that "push" them for higher hashrate rarely go higher than 145-150 range because the gains above that are TINY compared to the increase in power usage (my specific 1070 ti mining cards are currently set to 105 watts to keep total system draw under 6 amps on the system they are in).

Also, the Vega 64 pulls a lot more power IN GAMING USAGE than the 56 does, the 56 was deliberately designed by AMD to have a much lower TDP limit

 YOU need to check your ASSUMPTIONS, that's where you are messing up.

 Floating Point operations are 100% WORTHLESS for cryptocoin usage, which is 100% INTEGER operations.
 Another BAD ASSUMPTION you make, not a mistake on MY part.
 Would you also care to explain how "LTE16x" cell phone interface helps mining? Just for ONE example of the "not useful stuff for mining" on a Snapdragon 835.
 
 My "under 300 watt" measurement was made AT THE WALL on a Brand power meter while the system was actively hashing at 1950+ hash on Monero mining - and that was on a system that is NOT "power optimised" well, as I've never figured out how to get bloody Wattman to do a lot of the stuff I can do routinely in Afterburner like UNDERVOLT (Vega 56 cards LOVE to be undervolted, they tend to clock HIGHER with some undervolt as it lets them stay below the TDP easier). It is running a severely overkill Gold-rated power supply (Seasonic X-850) because that's what I had available when I put the system together, but the system is based on a FM2 motherboard with am AMD A10-7890k (which is NOT a low power APU) with the iGPU running the graphics for Win10, an HGST 3TB hd (system is also doing BURST mining) so the actual power draw of the SYSTEM as a whole would probably be about 275 watts (That model of PS usually pulls about 92% efficiency in the 30-50% load range).
 220 watts draw for the GPU is a PESSIMISTIC estimate, as I'm pretty sure the rest of the system is pulling 80-100 watts total NOT 50-60.

 Presuming an optimistic 2 watts, can a Snapdragon manage 20 hash on Monero?
 8 cores at 2 Ghz (ballpark average, I saw the "big/little" core split) in theory should manage more, but how much CACHE do they have on them - Monero wants ballpark 2 MB of CPU CACHE MEMORY per thread to run efficiently (and I can't find a spec anywhere that shows the amount of CPU cache on a Snapdragon).

 Then figure in the COST of the things - even *IF* they can mine efficiently, is it worth the COST of the things for whatever hashrate they achieve?
 THAT is the primary reason pretty much any SoC setup gets ignored for mining - even if it IS efficient in hash/watt, the sheer COST makes the time to achieve ROI end up being measured in YEARS.

Every single review on the internet rates RX Vega power consumption between 250 and above 300W for the an Overclocked 64 at full load, so unless you are really lowering your power limit to at least or around 70% I really want to see your system consuming under 300w (GPU AND CPU full load).  so even if we take a 350W conservative number by adding (system power, one system per 5 cards and divided)so we are on the same page. 6000/350= 17 cards not 30!!! so please check your math

Also other mistake you are doing is that you are assuming that the only usefull part of SoC is the CPU which is not, the easy way is obviously to use the CPU only but like I mentioned before a Tegra X1 from 2015 can achieve 1TFlops 16FP. there is no room for comparison here if The ARM SoC are fully optimised for mining software wise they'll beat RX Vegas in terms of efficiency, but the problem would the cost of making such a system that would not be worth it

A Snapdragon 835 in your smartphone consumes 2W to 2.5W, when I used 5W to have a marging even considering the boards power consumption for the nodes; so no

High end smartphones have never been profitable for mining - they cost way too much for the performance. Cheap smartphones like $20 Moto Es give way more performance per dollar for mining.

I have also tried mining earnhoney with a Pi 3. It is profitable when it runs, but it proved too unstable to be worth running.

My current single Vega 56 system eats a hair under 300 watts for the COMPLETE SYSTEM (that's probably 200-220 for the Vega) while mining Monero at about 1960 hash/sec - and I've not had the time to work on power optimisation, and Wattman makes it a PITA to do so IMO (I had it working with Afterburner for a while, but had a power outage and haven't been able to get it working again with AB).

 I dunno where your "15 Vegas for 6 kilowatts" is out of, should be closer to 30 even with POOR power optimisation - and I've seen claims of managing close to 1900 with 120 watts out of a Vega 56 with a "flash it with a Vega 64 BIOS then well-optimise it" setup.

 250-300 watts for a Vega 56 is WELL above it's rated TDP, though they run cool enough that you might be able to get there if you push TDP hard enough - but a lot of that wattage will be waste heat, they run faster when undervolted and don't need tons of power to max out their core clock.


 Can an ARM even mine ETH at all?
 I would not hold out a lot of hope for it on ZEC, they're a LOT lower performance than the i7 hardware some folks used to CPU mine ZEC in the early days and the hashrates on THOSE was low double-digits.
 Monero - they might have a prayer against a Vega if they can manage better than 10 hash per watt used ...... hmm, given all the "extra stuff" on the Snapdragon 835 that doesn't DO computing, how much DOES that thing eat - and how bloody much does it COST?

At one point (over a year ago), it was profitable to mine with cheap ($20 or less) smartphones, but now it won't return enough to break even in a reasonable time.

Ah yes this, a cluster of 1000 of those the only problem with this is that raspberry uses old SoC from 4 years ago or so If I'm not what I call efficient by any mean, but if a similar solution with let's say a Qualcomm Snapdragon 835 or a Tegra X1 or something similar could do wonders and beat out a out RX Vegas in specific operation since an RX vega consumes between 250 and 300W around 20 of those could consume the same amount of power as a thousand and this without counting the power consumption of the boards cpus and other components, I'd say 15 Vegas is a more realistic number vs a cluster of a 1000 of those so that's a ratio of around 67 SoC per Vega card.

Now the question that should be asked is, is a Vega 56 is more powerfull than 67 X1 for example, the X1 is 1Tflops of peak 16FP calculation where as the Vega around 21 Tflops which means that the Cluster has more potential than the Vega cards, but the software to take fully advantage is not there.


And the big elephant in the room is that  I'm not factoring the costs, obviously the amount of work and the price of component is in favor for the classic mining rig with vega cards. so Tl;dr, the ARM route can be more powerefficient if the software is there, but it is right now just more expensive and not fully optimised to be the better solution for mining. but for other use cases it is much more interesting.