Would have been great if the performance was there.
most of the board's power (75-85%) does not come from the CPU. The 5W tdp chip running only 2W higher at load, would have made a lot more sense if it would run at 2,5Ghz at 20W instead. Power/performance ratios are off.
I mean, a 24 core 1 Ghz arm processor like this, performs similar to a 12 core 2Ghz, or an 8 core 3Ghz arm CPU.
I bet the latter two CPUs would only consume 2 to 4watts more.
This design would make sense, if it can be scaled up by 10x. A 128-256 core arm CPU at 1Ghz does make sense.
Don't forget Intel Larrabee, that got scrapped for exactly the same reason, too low core by core CPU performance, and overall system power consumption was too high.
It would be a massive hit if it had 3 to 4 hyperthreads per core!
Announcement
Collapse
No announcement yet.
Benchmarks Of The 24-Core ARM Socionext 96Boards Developerbox
Collapse
X
-
This thing would be great as an automated test box. If you are developing on ARM, you might want to have 23 VMs running a continuous script to test all functions on whatever gets uploaded to your repo, each VM being a different Android version. And sure, if you are playing in ARM ecosystem all the time, this is nice.
But it's not really impressive on any metric. There are many low power cores, but not that many to be impressive. 24 is meh these days.
A53 cores are supposed to be so small, give me 255 of them for 140W instead if 24 cores at 33W. Then I might find use for it. Not when you can get real huge 32+ core full instruction CPUs that can emulate the ARM stuff much faster in real time metrics, not theoretical clock metrics.
Leave a comment:
-
Any proof this runs on 5 W? or could that be false advertising? ARM cores usually take 600 mW; 24 cores should make 14 W as reported by your test.
Leave a comment:
-
Originally posted by anarki2 View PostBwaha, here come the "desktop performance" ARM chips! I mean, it's just ridiculousHow could anyone take these promises seriously. The only bigger gimmick is the ARM servers lol. Most laughable concept ever.
I imagine four A76s would perform better than this board - but it wouldn't have the multi-core aspects that is surely the point of this board - to get developers creating better multithreaded server applications.
In the future, this company will likely release something faster, with a bigger target market in mind, for less money. Even if that turns out to be , for example, 32 A55s at 1.8GHz - it will run the previously developed software better. The board would likely be cheaper as well, as it will have a wider market.
And if you have a developer who costs you $10k a month, then adding a one-off cost of a couple of grand really isn't an issue for a serious business, if it speeds up that developer's work. Most likely you'd share a board between a few people in a team working on a common project as well. And lots of workloads don't need high IPC because of the I/O bound nature - and some of the benchmarks show this. You can also consider that this is a 24T CPU implemented using 24 single-thread cores, versus for example a 32T CPU implemented using 8 4-way SMT/CMT cores. Pros and Cons to each method.Last edited by sykobee; 03 September 2018, 08:19 AM.
Leave a comment:
-
Originally posted by coder View PostYeah, but $1200 for this? I'd rather use a Jetson TX2 for half the price. On any low thread-count workloads, the TX2 is way faster.
Think of it more as a scaled down model of a real 32 or 64 core ARM server. Yeah, its not going to run as fast as the real thing, but it can sit under your desk. Useful to folks who want to make sure their sw scales up to higher core count. Or in my case, I'd like to get my hands on one since it would be a good setup for debugging amd/nouveau on arm.
- Likes 2
Leave a comment:
-
Originally posted by Weasel View PostExactly this. People who just do "./configure && make" think they're "developing" but that's not the case how it's done at all. We don't compile from scratch all the time, only the few files that changed and then link them (which is cheap without LTO).
Only with LTO it gets compiled from scratch again, but that's only done once at release time (if you're even using it, most projects still don't). Which is not "during developing".
BTW C++ linking is not cheap. Or at least it used to be expensive, I have not worked on large C++ projects for years.
Lastly, seriously who cares how low power it is? Did you look at the price? You're not going to save $1000 in power bills with it.
Leave a comment:
-
You can already buy a Snapdragon 835 laptop, albeit it runs Windows 10. They're not even cheap because they're somewhat high end and cost of the CPU has little to do in this.
Snapdragon 1000 is next year?
Now, Windows on ARM sucks, the bootloader is probably locked to hell but the target is clear, business people and salesmen etc.
They run Excel and Outlook etc. and don't even change their wallpaper.
The benefit is going from train to hotel room to conference room without charging it (and having a 4G modem, later 5G though nothing prevents having 4G on an x86)
Having it all work on Linux might take more time : driver/kernel support, lawsuit against MS to unlock the bootloader.Last edited by grok; 30 August 2018, 04:47 PM.
Leave a comment:
-
Honestly, I can't really see an ARM Desktop computer being something tangible for at least 5 years. If it happens I would think it would be a joint partnership with AMD, something for us regular PC users. I don't know enough about this stuff. I will say this, why ARM for Desktop users? and further what is the point?
Bring something to market affordable and available for consumers besides tablets and smartphones and I will be all ears and thoroughly interested.
Leave a comment:
-
Ampere is mainly aimed towards cloud based stuff. Obviously has to be outrageously expensive.
Leave a comment:
-
Don't know how far Ampere 64-bit has come along but they are now designed around 32 cores @3.3Ghz with turbo and support 72-bit DDR4-2667, TDP 125w. 42 lanes of PCIe Gen3. I know their systems are running CentOS.
Leave a comment:
Leave a comment: