rooted Posted February 19, 2019 Posted February 19, 2019 N2 sbc-bench results (overclocked):http://ix.io/1BrG
sfx2000 Posted February 19, 2019 Posted February 19, 2019 18 hours ago, balbes150 said: With such a chip, RPi will become big price. Not Snapdragon... Think RISC-V for the Pi folks... https://riscv.org/membership/4531/raspberry-pi/ They've been clear that the recent releases for the Broadcom VC4 is pretty much done... Just saying...
NicoD Posted February 19, 2019 Author Posted February 19, 2019 12 minutes ago, rooted said: N2 sbc-bench results (overclocked):http://ix.io/1BrG It seems something went wrong with cpuminer. System health while running cpuminer: Time big.LITTLE load %cpu %sys %usr %nice %io %irq Temp 00:41:05: 1992/1992MHz 7.04 22% 0% 21% 0% 0% 0% 36.3°C 00:41:25: 1992/1992MHz 5.38 20% 0% 20% 0% 0% 0% 29.7°C 00:41:45: 1992/1992MHz 4.14 0% 0% 0% 0% 0% 0% 29.2°C 00:42:05: 1992/1992MHz 3.24 0% 0% 0% 0% 0% 0% 29.0°C 00:42:25: 1992/1992MHz 2.61 0% 0% 0% 0% 0% 0% 28.7°C 00:42:45: 1992/1992MHz 2.15 0% 0% 0% 0% 0% 0% 28.8°C 00:43:05: 1992/1992MHz 1.82 0% 0% 0% 0% 0% 0% 28.5°C 00:43:25: 1992/1992MHz 1.59 0% 0% 0% 0% 0% 0% 28.4°C 00:43:45: 1992/1992MHz 1.42 0% 0% 0% 0% 0% 0% 28.2°C 00:44:05: 1992/1992MHz 1.30 0% 0% 0% 0% 0% 0% 28.1°C 00:44:25: 1992/1992MHz 1.21 0% 0% 0% 0% 0% 0% 28.2°C 00:44:45: 1992/1992MHz 1.15 0% 0% 0% 0% 0% 0% 27.9°C 00:45:05: 1992/1992MHz 1.11 0% 0% 0% 0% 0% 0% 27.8°C 00:45:25: 1992/1992MHz 1.08 0% 0% 0% 0% 0% 0% 27.7°C 00:45:45: 1992/1992MHz 1.05 0% 0% 0% 0% 0% 0% 27.6°C VS a healty one. System health while running cpuminer: Time big.LITTLE load %cpu %sys %usr %nice %io %irq Temp 23:20:03: 1800/1896MHz 6.60 14% 0% 13% 0% 0% 0% 47.6°C 23:20:24: 1800/1896MHz 6.79 100% 0% 99% 0% 0% 0% 50.2°C 23:20:45: 1800/1896MHz 6.85 100% 0% 99% 0% 0% 0% 50.4°C 23:21:06: 1800/1896MHz 6.90 100% 0% 99% 0% 0% 0% 50.7°C 23:21:26: 1800/1896MHz 6.93 100% 0% 99% 0% 0% 0% 50.8°C 23:21:47: 1800/1896MHz 6.95 100% 0% 99% 0% 0% 0% 51.1°C 23:22:08: 1800/1896MHz 6.96 100% 0% 99% 0% 0% 0% 51.3°C 23:22:29: 1800/1896MHz 7.03 100% 0% 99% 0% 0% 0% 51.5°C 23:22:50: 1800/1896MHz 7.02 100% 0% 99% 0% 0% 0% 51.6°C 23:23:11: 1800/1896MHz 7.02 100% 0% 99% 0% 0% 0% 51.8°C 23:23:31: 1800/1896MHz 7.01 100% 0% 99% 0% 0% 0% 52.0°C 23:23:52: 1800/1896MHz 7.01 100% 0% 99% 0% 0% 0% 52.2°C 23:24:13: 1800/1896MHz 7.00 100% 0% 99% 0% 0% 0% 52.4°C 23:24:34: 1800/1896MHz 7.00 100% 0% 99% 0% 0% 0% 52.5°C 23:24:55: 1800/1896MHz 7.06 100% 0% 99% 0% 0% 0% 52.7°C In 7-zip it now runs at 533% core use vs 543% last time, while it's still 6% faster. But it's also about 8C cooler with the OC while running multicore 7zip. Did you change anything else except the OC(fan)?
NicoD Posted February 19, 2019 Author Posted February 19, 2019 22 minutes ago, sfx2000 said: Think RISC-V for the Pi folks... https://riscv.org/membership/4531/raspberry-pi/ Your words a while ago... "I don't think the board makers would do their own chips based on RISC-V - the costs to fab a chip are very high, and most of the board vendors in the Armbian space are low margin.." You might have to eat your words one day. I don't know if they've got plans for it, not quickly I assume. It's not ready at all, but the help from the foundation is very good. But one thing doesn't mean the other(in supporting RISC-V doesn't mean quickly adapting) But I'm very interested to see. I would love it...
rooted Posted February 19, 2019 Posted February 19, 2019 It seems something went wrong with cpuminer. System health while running cpuminer:Time big.LITTLE load %cpu %sys %usr %nice %io %irq Temp00:41:05: 1992/1992MHz 7.04 22% 0% 21% 0% 0% 0% 36.3°C00:41:25: 1992/1992MHz 5.38 20% 0% 20% 0% 0% 0% 29.7°C00:41:45: 1992/1992MHz 4.14 0% 0% 0% 0% 0% 0% 29.2°C00:42:05: 1992/1992MHz 3.24 0% 0% 0% 0% 0% 0% 29.0°C00:42:25: 1992/1992MHz 2.61 0% 0% 0% 0% 0% 0% 28.7°C00:42:45: 1992/1992MHz 2.15 0% 0% 0% 0% 0% 0% 28.8°C00:43:05: 1992/1992MHz 1.82 0% 0% 0% 0% 0% 0% 28.5°C00:43:25: 1992/1992MHz 1.59 0% 0% 0% 0% 0% 0% 28.4°C00:43:45: 1992/1992MHz 1.42 0% 0% 0% 0% 0% 0% 28.2°C00:44:05: 1992/1992MHz 1.30 0% 0% 0% 0% 0% 0% 28.1°C00:44:25: 1992/1992MHz 1.21 0% 0% 0% 0% 0% 0% 28.2°C00:44:45: 1992/1992MHz 1.15 0% 0% 0% 0% 0% 0% 27.9°C00:45:05: 1992/1992MHz 1.11 0% 0% 0% 0% 0% 0% 27.8°C00:45:25: 1992/1992MHz 1.08 0% 0% 0% 0% 0% 0% 27.7°C00:45:45: 1992/1992MHz 1.05 0% 0% 0% 0% 0% 0% 27.6°C VS a healty one. System health while running cpuminer:Time big.LITTLE load %cpu %sys %usr %nice %io %irq Temp23:20:03: 1800/1896MHz 6.60 14% 0% 13% 0% 0% 0% 47.6°C23:20:24: 1800/1896MHz 6.79 100% 0% 99% 0% 0% 0% 50.2°C23:20:45: 1800/1896MHz 6.85 100% 0% 99% 0% 0% 0% 50.4°C23:21:06: 1800/1896MHz 6.90 100% 0% 99% 0% 0% 0% 50.7°C23:21:26: 1800/1896MHz 6.93 100% 0% 99% 0% 0% 0% 50.8°C23:21:47: 1800/1896MHz 6.95 100% 0% 99% 0% 0% 0% 51.1°C23:22:08: 1800/1896MHz 6.96 100% 0% 99% 0% 0% 0% 51.3°C23:22:29: 1800/1896MHz 7.03 100% 0% 99% 0% 0% 0% 51.5°C23:22:50: 1800/1896MHz 7.02 100% 0% 99% 0% 0% 0% 51.6°C23:23:11: 1800/1896MHz 7.02 100% 0% 99% 0% 0% 0% 51.8°C23:23:31: 1800/1896MHz 7.01 100% 0% 99% 0% 0% 0% 52.0°C23:23:52: 1800/1896MHz 7.01 100% 0% 99% 0% 0% 0% 52.2°C23:24:13: 1800/1896MHz 7.00 100% 0% 99% 0% 0% 0% 52.4°C23:24:34: 1800/1896MHz 7.00 100% 0% 99% 0% 0% 0% 52.5°C23:24:55: 1800/1896MHz 7.06 100% 0% 99% 0% 0% 0% 52.7°C In 7-zip it now runs at 533% core use vs 543% last time, while it's still 6% faster. But it's also about 8C cooler with the OC while running multicore 7zip. Did you change anything else except the OC(fan)? My air-conditioning was on and blowing on the device.cpuminer crashed I believe at the end
sfx2000 Posted February 19, 2019 Posted February 19, 2019 3 hours ago, NicoD said: Your words a while ago... "I don't think the board makers would do their own chips based on RISC-V - the costs to fab a chip are very high, and most of the board vendors in the Armbian space are low margin.." RPi is going to do what they see as best for their overall objectives.... Their current platform is at the best end of what they can do with the broadcom VC4 - no further headroom for them. With 33M plus devices sold, one can assume they have some resources to choose their next steps - ARM or RISC-V....
TonyMac32 Posted February 19, 2019 Posted February 19, 2019 37 minutes ago, sfx2000 said: With 33M plus devices sold, one can assume they have some resources to choose their next steps - ARM or RISC-V.... And with 33M sold, they don't have to answer the hard questions on their forums... I actually just mentioned this to @chwe the other day, whether or not RPi would go RISC-V. The reality is, any SoC change destroys their little ecosystem, so why not make it a flashy change? They externally put up a front about open source and such, so why not use a fully open-source core? As far as years to implement, well: https://www.extremetech.com/computing/285856-western-digitals-risc-v-swerv-core-now-available-for-free Perhaps not quite as long as all that after all if they've been working with a bigger partner...
balbes150 Posted February 19, 2019 Posted February 19, 2019 8 hours ago, NicoD said: "A lot of clueless people (me included) think/thought heat would be rising. It doesn't, it just radiates in all directions while only hot/warm air rises. As such having the huge heatsink at the bottom isn't that much of an issue (personally tested with ODROID HC1 which uses exactly the same design of having the SoC on the bottom PCB side directly attached to a giant heatsink below" "Combining this huge heatsink with a cheap plastic cover creates a full enclosure able to efficiently dissipate the heat to the outside unlike RPi inspired 'cooling solutions' where the SoC is on the wrong PCB side and also inside the enclosure making efficient heat dissipation nearly impossible." 1. Arguments that" it is possible to turn a radiator upwards " - aren't discussed. If desired, you can do anything. This solution was not offered by the developers of HK immediately during the development (and in the description of the use), so this use case is not the merit of the developers of HK, which is already confirmed by the facts described earlier. 2. A huge radiator is not a plus, but a minus for developers, because developers, instead of doing something good (correctly calculated design), decided the issue of cooling blunt and not effective increase in costs. With an unjustified increase in the weight of the entire structure, the hidden costs incurred by buyers are increased (the manufacturer always includes them in the final price). For example, transportation and postage costs for the delivery of final products to consumers, the cost of production is not the right radiator, etc.. 3. It's funny to read the opposite reasoning, in the beginning argue about the possibility of turning the radiator up (and by the way, only in this position the radiator will benefit from heat radiation), and then talk about the ease of connecting additional devices. It is interesting to see how the user will work with the display, which is connected face to the table. 4. Given the fact that all the numbers about the temperature, frequency, number of cores used their modes, etc., forms a closed binary code for "secret algorithms", I am skeptical about all the results, except one - the real work. All artificially invented tests are a measurement of a" spherical horse in a vacuum", which is not difficult to" teach " the correct answers. PS test Results under air conditioning - a good example of how you can (if desired) to organize any "correct" tests.
balbes150 Posted February 19, 2019 Posted February 19, 2019 Quote: ------------------------ Whoa... careful there. Hot air rises, “heat” does not rise. As you yourself say, heat radiates. Heat leaves a hot object in straight lines as if it were light shining from the object. When it collides with something, that something gets the transferred energy (physicists in the audience please forgive me). A tiny amount will transfer to air, hence the inefficiency of using air as the transfer medium... but air is cheap, plentiful, non-electrically conductive, non-corrosive, and easy to move along the radiating surface. Thus a heat sink in any orientation will heat the things around it, and air in contact with hot surfaces will absorb some heat... becoming hot air and then expanding and rising (because it is less dense). So as long as that heated air can flow away, you will end up with a convective flow and a constant supply of cooler air (that gets heated and flows away, etc). Don’t put your heat sink in an upside down bowl, and you should be fine. ------------------------- Sorry, I could not resist, it's just a masterpiece of stupidity and ignorance of the basic laws of physics ... For reference. If the radiator "looks" at the surface of the table, the radiation is reflected from the surface of the table (all solids have a reflection) and returns back to the radiator (only a small part, where the radiator ends, will be reflected in the environment (i.e. will leave from N2). Only a very small part of the energy is transferred to the top of the table surface and tries to heat it. As soon as the top of the table is heated, it starts to heat the air above it and .... this heated air rises back to the radiator. The better the reflection of the table, the greater the back radiation. Used black radiator, so it is much better to get back the heat (in the form of radiation) than gives. For this reason , all heat exchangers, which are used to heat the water by the sun (radiation) are specially painted in black , which would be better to take energy (pay attention, do not give, namely to receive energy). Radiation works only from surfaces that "look" into the environment. That is, all the surfaces of the lamellas on the radiator emit energy into the environment only by a very small surface (on the tops of the lamellas). The rest of the surface looks at each other and radiates energy back to each other (i.e. does not do any good to remove heat from the system). Only a smooth plate reflects the entire surface. By the way, just a reflection from the radiator, which is directed towards the PSB Board, "helps" to heat the components on it. Deep lamel are designed to increase the area of air contact with the radiator, but they work only with good air movement past them (with active air supply). When natural air circulation Delta recoil does not exceed 10-20 W\g, and when forced to 200-250 W\g.
NicoD Posted February 19, 2019 Author Posted February 19, 2019 5 hours ago, TonyMac32 said: As far as years to implement, well: https://www.extremetech.com/computing/285856-western-digitals-risc-v-swerv-core-now-available-for-free Perhaps not quite as long as all that after all if they've been working with a bigger partner... Nice, I can't like posts today, so consider your post liked @balbes150 I agree that we disagree Putting the heatsink direcly on the table without rising it up a bit is indeed not good practice. But I still think it's not bad to have it there. It reaches 52C maxed, so it's more than sufficient, and 52C isn't a danger to the board at all. And to keep the top free for hats and so is also a good argument by @tkaiser. Let's keep it at this, I don't want to become a lower-heatsink fundametalist Cheers
chwe Posted February 19, 2019 Posted February 19, 2019 a bunch of off-topic here: Spoiler 6 hours ago, TonyMac32 said: 6 hours ago, sfx2000 said: With 33M plus devices sold, one can assume they have some resources to choose their next steps - ARM or RISC-V.... And with 33M sold, they don't have to answer the hard questions on their forums... those numbers spread from 20M to 30M (33M is new to me) depending where you look.. There aren't many reasons to go to risk for them (except Broadcom glues a Risk core to a VC5 ). Their VC4 ecosystem will break no matter what comes next.. 3b+ is end of pipe for the VC4. They still fail to make a bunch of their stuff 64bit compatible (or aren't willing to do it). They can't go to Rockchip cause rockchip offers their SoC also to its competitors.. Something they never want.. You don't get the RPi SoC on free market.. And since more or less every new board offers at least USB3 and/or GbE - they need something which can at least partly compete with them means real GbE not 'by electrical characteristics' (that was a funny discussion btw.. ). I like the SoC on the bottom approach from multiple boardmakers, when done correctly and not always under high load.. It allows an passive cooling for small boards (where a standard mounted cooler, not such a tiny sticking cooler, wouldn't fit). For 'farms' like @wtarreau one the bottom cooler approach also allows a nice cooling. If I need active cooling I'd rather go for an X86 based solution than an ARM.. Sorry the reason I use SBCs is cause they don't make any noise.. IMO if they do, there aren't many reasons to use one.
rooted Posted February 19, 2019 Posted February 19, 2019 PS test Results under air conditioning - a good example of how you can (if desired) to organize any "correct" tests. The first test was done with the AC off.The location of my devices happens to be where an AC vent blows from the ceiling, nothing I can do about that really...
sfx2000 Posted February 20, 2019 Posted February 20, 2019 20 hours ago, TonyMac32 said: And with 33M sold, they don't have to answer the hard questions on their forums... Because they don't have to - they have a working solution that works for their objectives... For now... 20 hours ago, TonyMac32 said: I actually just mentioned this to @chwe the other day, whether or not RPi would go RISC-V. The reality is, any SoC change destroys their little ecosystem, so why not make it a flashy change? They externally put up a front about open source and such, so why not use a fully open-source core? They're going to do what they're going to do... because they can afford to do so... "Only Nixon could go to China" - one must understand the dynamics of the time... and of the current dynamics with Broadcom/Cypress and ex-Avago - the Broadcom chip is convenient for them as long as they don't need to spend so much... until they need to - and for Broadcom/Avago, it might not be worth it for them. The VC4 based chips are close to end of development - the VC4 is 32bit only, maxed out at 1GB RAM, and is stuck on a 40nm process dictated by the layout of the VC4... Going to 28nm is a new tape out for VC4, and Broadcom likely isn't willing to go there for an old VPU when VC5 is the future... Don't get me wrong - RPi and Broadcom have done well to get a lot of mileage out of a basic VC4 chip - where they are right now - Cortex-A53 at 40nm is about as fast as it can go at 1.4GHz, and turbo'ing VC4 to 400MHz from 250MHz... Not so much different than Apple going to x86 after being heavily invested in PowerPC until it basically stopped being competitive... and that's where RPi and the VC4 chip is right now, they've run out of room... So perhaps a hard cutover - they can port their platform over to RISC-V, and look like a strong player...
rooted Posted February 20, 2019 Posted February 20, 2019 @NicoD I ran the BMW Blender benchmark and the results are. Time: 50 minutes 28 seconds Maximum temperature: 69.2° C
balbes150 Posted February 20, 2019 Posted February 20, 2019 16 hours ago, rooted said: The first test was done with the AC off. The location of my devices happens to be where an AC vent blows from the ceiling, nothing I can do about that really... I'm not accusing you of forgery. This was an example of how many different factors (both often implicit and difficult to track) affect events. Maybe I'm not looking hard enough, N2 is WoL ? 1
NicoD Posted February 20, 2019 Author Posted February 20, 2019 5 hours ago, rooted said: @NicoD I ran the BMW Blender benchmark and the results are. Time: 50 minutes 28 seconds Maximum temperature: 69.2° C Thank you very much. That's a lot faster than I would have imagined(20 minutes to my previous fastest or 28% faster). I imagined around the hour. Here you see the huge difference in lpddr3-lpddr4. Also you see that it takes a long time to heat up the heatsink fully. I think no airco at all now. So for maxed out bursts of 15/20 minutes it seems fine to run without a fan. But continuous use maxed out will probably need a (small) fan. But this is only when cpu use is always the full 100%. Many software can't utilise the full 100%. For me this will be a render beast. Now I see that the complete package of ram+SoC will be miles ahead to my NanoPi M4. The Rock Pi 4b could do with the same OC as the M4 of 1.5Ghz-2Ghz. That one is almost as fast stock (1.4-1.8Ghz) as the M4 with OC(for render tasks). Also thanks to the lpddr4. I can't wait to get it in my hands. So many test I'd like to do with it. How long does it take for the heatsink to reach it's max temp, How much power does it consume, how's the single core performance in booting software, what OC's are all possible,....... I also wonder if the use of HDMI made the temperature rise. Thank you so much @rooted
NicoD Posted February 20, 2019 Author Posted February 20, 2019 9 hours ago, rooted said: @NicoD I ran the BMW Blender benchmark and the results are. Time: 50 minutes 28 seconds Maximum temperature: 69.2° C At what frequencies did you run this test? 1.8-1.9Ghz or 2-2Ghz? That's rather important to know.
rooted Posted February 20, 2019 Posted February 20, 2019 At what frequencies did you run this test? 1.8-1.9Ghz or 2-2Ghz? That's rather important to know.At the stock frequencies, it would throttle if overclocked I'm guessing.I was quite pleased how quickly it rendered the scene.I also checked the gimp functionality, it works really well and renders plug-ins such as iWarp on large images quickly. The performance of the N2 doesn't disappoint, especially considering the retail price. I was very sceptical when I learned it was Amlogic based but they really upped their game with this package, it will be a popular SoC in the coming months.
NicoD Posted February 20, 2019 Author Posted February 20, 2019 54 minutes ago, rooted said: I also checked the gimp functionality, it works really well and renders plug-ins such as iWarp on large images quickly. Many gimp effects are single core. And that the S922X does good too. 56 minutes ago, rooted said: The performance of the N2 doesn't disappoint, especially considering the retail price. I was very sceptical when I learned it was Amlogic based but they really upped their game with this package, it will be a popular SoC in the coming months. I started to think it was overclocked since it was this fast. I also doubted about the clockspeed/performance. But indeed it seems to be real this time. (I would have loved an Odroid C2 @ 2Ghz 2 years ago, but at 1.75Ghz it did very well) It's a bigger performance gain than I imagined for real live tasks that need ram+cpu. I want it so badly, this is going to be so useful. It's got all I need, super fast multi-core, fast single-core and fast ram. I'll look for a 12V power bank for it so I can use it everywhere.
rooted Posted February 20, 2019 Posted February 20, 2019 I'll look for a 12V power bank for it so I can use it everywhere. It accepts 7.5V ~ 20V according to the specs, powering with a battery pack shouldn't be an issue This range makes it perfect for automotive use, such as a CarPC. 2
balbes150 Posted February 20, 2019 Posted February 20, 2019 10 hours ago, NicoD said: For me this will be a render beast. Now I see that the complete package of ram+SoC will be miles ahead to my NanoPi M4. The Rock Pi 4b could do with the same OC as the M4 of 1.5Ghz-2Ghz. That one is almost as fast stock (1.4-1.8Ghz) as the M4 with OC(for render tasks). Also thanks to the lpddr4. How many have shown with Blender your rk3399 ?
NicoD Posted February 20, 2019 Author Posted February 20, 2019 44 minutes ago, balbes150 said: How many have shown with Blender your rk3399 ? Blender results BMW render @ 1080p Odroid N2 @ 1.9Ghz-1.8Ghz Bionic 50m28s NanoPC T3+ @ 8x1.4Ghz Armbian Bionic 1h10m25s The NanoPi M4 @ 1.5Ghz-2Ghz Armbian Bionic 1h13m28s 1.4Ghz-1.8Ghz FriendlyElec Bionic 1h28m13s RockPi 4B @ 1.4Ghz-1.8Ghz Bionic 1h17m22s Odroid C2 @ 1.75Ghz 1104Mhz ram Bionic 2h10m21s 1.5Ghz 912Mhz ram Bionic 2h35m10s Rock64 @ 1.5Ghz Bionic 2h55m56s The difference is huge. The N2 is in a league of it's own here. You can only compare 64-bit with 64-bit OS's with Blender. And Stretch performs a bit worse than Bionic. 7zip is an ok test for cpu only. Blender is good to see the performance for most daily use(cpu+ram). 1
balbes150 Posted February 21, 2019 Posted February 21, 2019 10 hours ago, NicoD said: The difference is huge. The N2 is in a league of it's own here. Do you really see a big gap here ? Interesting figures are obtained. The NanoPi M4 @ 1.5Ghz-2Ghz Armbian Bionic 1h13m28s 1.4Ghz-1.8Ghz FriendlyElec Bionic 1h28m13s RockPi 4B @ 1.4Ghz-1.8Ghz Bionic 1h17m22s If I'm not mistaken : NanoPi M4 -> DDR3 RockPi 4B -> DDR4 The transition from 1.4 \ 1.8 to 1.5\2.0 for NanoPi M4 (DDR3), shows a significant difference , almost 15 minutes. So if you move the model from DDR4 to the same mode, we get about the same improvement. So the figure for RockPi 4B will be 1h17m - 15m = 1h2m It's only 11 minutes (That's less than 20%) worse than N2 (which has 4 super-fast cores). Yes, I know that you can try to raise the frequency of the N2, but it will require active cooling and vaunted stupid huge passive radiator will not allow this to be done without the addition of a fan. More interesting nuances. In the Blender test, has the following numbers. The temperature is almost 70 degrees, with an ambient temperature of only 22 degrees and a hidden semi-active cooling system from the air conditioner (i.e. these are very good conditions for the passive cooling system). This is just 50 minutes of running all the cores with maximum load (this is my assumption, because I do not know exactly whether I loaded Blender all cores work at 100% or not) and the standard frequency of 1.8\1.9 . And now look at the graphs from the official temperature test N2. There at an air temperature of 35 degrees, the temperature of N2 at a maximum load of 73 degrees. That is, only 3 degrees more than in the Blender test. Very amusing comparison. In this case, the Delta for the operation of the passive cooling system in the tests is obtained 70-22=48 and 73-35=38. That is, the difference between good conditions and complex 10 degrees (this is a significant value for passive cooling systems with such input parameters). I.e. imagine that you are easier to do after a hard work at 35 degrees than at 22. And if you look closely at the official chart, you can see the strange sawtooth behavior of the temperature. The oddity is that with a uniform maximum load and the onset of the thermal balance (the energy input is equal to the consumption) there is a significant temperature fluctuation for the passive cooling system. Possible reasons - either there is a cut-off when the temperature reaches critical, the system briefly and not noticeably for you, reduces the load, or the cooling system should instantly improve (for example, rapid heat removal by increasing the flow past the air radiator, which should take away the heat). In passive systems, there can be no sharp (instantaneous) fluctuations in air flows. For those who have brains, you can draw your own conclusions. IMHO s922 only enters the market and it is necessary time that this product "ripened" and began to be fully supported and work. rk3399 available for a long time and is well established. And it is important that the" old " rk3399 is not much inferior to the latest s922. I think Rockchip is not sitting still and working on new projects and maybe soon we can expect the appearance of improved versions of rk3399. Given that RK already has a good (and well-established) base in the form of rk3399, we can expect that we should expect very interesting events in the next few months.
NicoD Posted February 21, 2019 Author Posted February 21, 2019 50 minutes ago, balbes150 said: So if you move the model from DDR4 to the same mode, we get about the same improvement. So the figure for RockPi 4B will be 1h17m - 15m = 1h2m It's only 11 minutes (That's less than 20%) worse than N2 (which has 4 super-fast cores). Yes, I know that you can try to raise the frequency of the N2, but it will require active cooling and vaunted stupid huge passive radiator will not allow this to be done without the addition of a fan. You can't really make this reasoning. The RockPi 4B is at stock and the N2 is at stock. The RockPi must use a fan otherwise it overheats like crazy. If you want to make such a claim, then the N2 must be overclocked too. And that will make it go towards 40minutes (my pc doesn't do much better) So the gap is real. The bst M4 result is an overclocked result. Also with active cooling or it'd burn out. I do love the RK3399's, and they are more versitile that the S922X. But the combnation of lpddr4 and the S922X has a lot more horsepower than the RK3399s.
tkaiser Posted February 21, 2019 Posted February 21, 2019 Funny place here... All the RK3399 'benchmark' results collected so far were done with CONFIG_HZ=1000 (good for a responsive UI in Android and Linux, not so good for normal computing or server tasks) while everywhere around CONFIG_HZ=250 is the default. Is this affecting Blender or not? Anyone tested so far (latest ayufan images switched to CONFIG_HZ=250)? Anyone into Active Benchmarking instead of just firing up the next round of kitchen-sink benchmarks in fire-and-forget mode collecting more numbers without meaning? The challenge with 'overclocking' the N2 is not trottling but stability/reliability as can be easily seen with the cpuminer tests that are sufficient for reliability testing (see N2 notes here). The DVFS OPP are defined in a closed firmware BLOB and the whole thing happens on the Cortex-M3 inside S922X. N2 has no LPDDR4 but DDR4, memory performance depends not just on type of memory but on DRAM initialization (done with BLOBs on both RK3399 and S922X). TL Lim said Rockchip will provide a new BLOB to make use of faster LPDDR4 access on RockPro64 in 2018 but I don't know whether this has already happened or not Memory benchmark scores depend on stuff like dmc or CONFIG_HZ since how would you explain better memory performance when switching from CONFIG_HZ=1000 to CONFIG_HZ=250 on RK3399 (previously known as difference between RK's 4.4 kernel and mainline though it's just different CONFIG_HZ defaults). Same with executing a memory benchmark on a little vs. big core Memory bus width is different between S922X and RK3399 (32-bit vs. 64-bit) but whether this will result in faster execution depends on a lot of factors (application in question, DRAM initialization BLOB, kernel settings, kernel tunables, see dmc memory governor with RK3399, and so on) Added some updates to https://github.com/ThomasKaiser/Knowledge/blob/master/articles/Quick_Preview_of_ODROID-N2.md#updates 1
rooted Posted February 21, 2019 Posted February 21, 2019 Test for yourselves, Hardkernel has 4 N2 on the bench ready for testing. https://forum.odroid.com/viewtopic.php?p=246987#p246987@tkaiserI will build the kernel with 1000hz enabled soon and run sbc-bench again. 1
tkaiser Posted February 22, 2019 Posted February 22, 2019 Since the last round of Moronix madness didn't include a single RK3399 device I decided to give two rounds of silly kitchen-sink benchmarking a try using the 'reputable' Phoronix Test Suite: https://www.cnx-software.com/2019/02/13/odroid-n2-amlogic-s922x-sbc/#comment-560987 For those thinking the N2 results would be better when running bare metal... I doubt it since of course I checked with sbc-bench first: here's running sbc-bench on 'ODROID Bench' inside a container: http://ix.io/1BEM and there's the results Dongjin (@tobetter) shared with me few days ago (look at the timestamp, Hardkernel tested already weeks ago so they were pretty clear about which benchmark results to publish and which not): http://ix.io/1BsF The ODROID-N1 numbers are representative for each and every RK3399 device out there running at 2.0/1.5GHz with CONFIG_HZ=250. Scores might improve once Rockchip provides new DRAM initialization BLOBs especially on those boards using LPDDR4. 1
NicoD Posted February 22, 2019 Author Posted February 22, 2019 30 minutes ago, tkaiser said: Since the last round of Moronix madness didn't include a single RK3399 device I decided to give two rounds of silly kitchen-sink benchmarking a try using the 'reputable' Phoronix Test Suite: https://www.cnx-software.com/2019/02/13/odroid-n2-amlogic-s922x-sbc/#comment-560987 You're funny. You complain that others create random numbers that are made (just out of curiosity) and put on a forum like Armbian. And now I catch you commit the same "crime".
TonyMac32 Posted February 22, 2019 Posted February 22, 2019 He also hung a lantern on it, so to speak:. "silly kitchen sink benchmarking". He's illustrating how numbers can be used to selectively represent reality.Sent from my Pixel using Tapatalk 1
Recommended Posts