$107 for 2c/2t capped at 1.9GHz is horrendously overpriced. You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage. Seeing as the only reason you'd buy a 2-threaded CPU nowadays is for ultra-small and embedded devices, I really don't see this (or the Pentium for that matter) to sell well.
$107 for 2c/2t capped at 1.9GHz is horrendously overpriced. You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage. Seeing as the only reason you'd buy a 2-threaded CPU nowadays is for ultra-small and embedded devices, I really don't see this (or the Pentium for that matter) to sell well.
$107 for 2c/2t capped at 1.9GHz is horrendously overpriced. You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage. Seeing as the only reason you'd buy a 2-threaded CPU nowadays is for ultra-small and embedded devices, I really don't see this (or the Pentium for that matter) to sell well.
Few things play into this - x86 compatibility, which is worth something, and the fact that no manufacturer is buying this in quantities of 1000 - so they are probably paying closer to $70 per unit.
Few things play into this - x86 compatibility, which is worth something, and the fact that no manufacturer is buying this in quantities of 1000 - so they are probably paying closer to $70 per unit.
Both Linux and Windows 10 on ARM can run x86 programs sufficiently. Granted, not super efficiently, but if you're expecting a lot of performance of a sub-2GHz dual core, you may want to rethink your priorities anyway.
To my understanding (and maybe I'm wrong), these are BGA models, which the average consumer doesn't have access to buying directly. So, what's the point of listing the price as $107 when it won't ever be sold at that price?
Both Linux and Windows 10 on ARM can run x86 programs sufficiently. Granted, not super efficiently, but if you're expecting a lot of performance of a sub-2GHz dual core, you may want to rethink your priorities anyway.
To my understanding (and maybe I'm wrong), these are BGA models, which the average consumer doesn't have access to buying directly. So, what's the point of listing the price as $107 when it won't ever be sold at that price?
ARK pricing (the listed $107 is from ARK) is based on a thousand unit purchase. Companies like Dell/Synology/Buffalo/QNAP/etc that are putting these in small NAS devices, routers and stuff are probably buying 50 to hundred thousand. Linux is obviously fine - ARM compatibility on Windows 10 is left to be desired.. not only are drivers mostly a no-go, various applications suffer performance penalties on the current emulation layer (half the Surface Pro X demonstration was Adobe talking about building dedicated ARM apps for the device). Obviously it works to some degree but it still provides Intel with leverage over the ARM market to charge more. The Core M devices were a good example of this - they were selling 15w laptop processors for like $450 because at the time they knew they had no ultrabook competition. Now that ARM is somewhat supported and AMD has competitive products they are readjusting their 15w pricing but they are still the leader at that wattage - which still gives them a degree of freedom of pricing.
$107 for 2c/2t capped at 1.9GHz is horrendously overpriced. You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage. Seeing as the only reason you'd buy a 2-threaded CPU nowadays is for ultra-small and embedded devices, I really don't see this (or the Pentium for that matter) to sell well.
Uh.....no.....you can't. The IPC on these things is at least double the best currently available arm uarch (which are not used in sbc). Typical arm uarch have IPC of around 1/3 to 1/4 of this.
Uh.....no.....you can't. The IPC on these things is at least double the best currently available arm uarch (which are not used in sbc). Typical arm uarch have IPC of around 1/3 to 1/4 of this.
Uh... yes, you can, and you are heavily exaggerating how much faster x86 is. x86's performance really only stands out once you start including instructions like AVX or SSE4, and for CPUs of this performance tier, you're probably not going to be running applications that need those instructions.
Also, you can get ARM platforms with 6+ cores pretty easily, which would easily give you more total compute power than a 2c/4t x86.
EDIT:
Another thing to keep in mind is ARM tends to be better in terms of performance-per-watt, which is typically the top priority for this level of performance.
$107 for 2c/2t capped at 1.9GHz is horrendously overpriced. You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage. Seeing as the only reason you'd buy a 2-threaded CPU nowadays is for ultra-small and embedded devices, I really don't see this (or the Pentium for that matter) to sell well.
Uh... yes, you can, and you are heavily exaggerating how much faster x86 is. x86's performance really only stands out once you start including instructions like AVX or SSE4, and for CPUs of this performance tier, you're probably not going to be running applications that need those instructions.
Also, you can get ARM platforms with 6+ cores pretty easily, which would easily give you more total compute power than a 2c/4t x86.
EDIT:
Another thing to keep in mind is ARM tends to be better in terms of performance-per-watt, which is typically the top priority for this level of performance.
You do not need AVX or SSE4. ANY application will have dramatically better IPC on modern Intel/AMD chips vs. even something like a snapdragon 855. Current x86 chips use much bigger cores that achieve more ipc at the cost of higher power consumption at a given clock rate and higher chip size per core. Ignore the ISA for a minute and just look at the differences in the current uarch pipelines on each side. They are optimized for different things.
Yes arm chips do counterbalance this by having higher core counts to reach similar levels of performance. But this makes them less versatile since they still lack the single threaded performance of their x86 counterparts. Chips with higher core counts using smaller cores are almost always more power efficient but less versatile since many applications simply can't be multithreaded well depending on the workload. This combined with x86 compatibility is why consumer grade computers still overwhelming use current x86 chips.
If you look at similarly priced sbc boards (since you said those are faster) their chips are laughable. A53 cores running at 1 GHz are literally going to be 5x slower than these cores. Even with 4 cores they will not perform close to these. Nor should you expect them to since those boards have a fraction of the TDP. You can buy sbc boards that perform better but they cost A LOT more. So your original statement remains incorrect. These are priced very appropriately for what they are. You could prove me wrong on this by posting a link to a $100 arm sbc that performs similarly to this. But I haven't been able to find anything even close.
These chips are being primarly used on low end consumer laptops where they excel due to high single threaded performance and x86 compatibility. If someone is trying to put these in a embedded platform and does not need the extra power then yes they should use a lower power arm chip. But that's not the niche these are aimed at so I don't even know why we're talking about that in this thread....
You do not need AVX or SSE4. ANY application will have dramatically better IPC on modern Intel/AMD chips vs. even something like a snapdragon 855. Current x86 chips use much bigger cores that achieve more ipc at the cost of higher power consumption at a given clock rate and higher chip size per core. Ignore the ISA for a minute and just look at the differences in the current uarch pipelines on each side. They are optimized for different things.
Yes arm chips do counterbalance this by having higher core counts to reach similar levels of performance. But this makes them less versatile since they still lack the single threaded performance of their x86 counterparts. Chips with higher core counts using smaller cores are almost always more power efficient but less versatile since many applications simply can't be multithreaded well depending on the workload. This combined with x86 compatibility is why consumer grade computers still overwhelming use current x86 chips.
Remember the target applications here. We're talking phones/tablets (which are optimized to be multi-threaded, and aren't expected to run anything computationally complex), netbooks, embedded devices (where responsiveness isn't important), and robots (which are application-specific and therefore can be programmed to maximize the CPU's resources).
If you look at similarly priced sbc boards (since you said those are faster) their chips are laughable. A53 cores running at 1 GHz are literally going to be 5x slower than these cores. Even with 4 cores they will not perform close to these. Nor should you expect them to since those boards have a fraction of the TDP. You can buy sbc boards that perform better but they cost A LOT more. So your original statement remains incorrect. These are priced very appropriately for what they are. You could prove me wrong on this by posting a link to a $100 arm sbc that performs similarly to this. But I haven't been able to find anything even close.
Like I said before, the workloads you'd be expected to run on these devices will offer comparable performance. Also, cite your sources.
The Amlogic S922X is better than a lot of the CPUs listed in the benchmarks I listed above and can be found in SBCs for well under $100, such as the ODROID-N2.
These chips are being primarly used on low end consumer laptops where they excel due to high single threaded performance and x86 compatibility. If someone is trying to put these in a embedded platform and does not need the extra power then yes they should use a lower power arm chip. But that's not the niche these are aimed at so I don't even know why we're talking about that in this thread....
Yes, these chips (including the ARM ones) are also used in low-end consumer laptops - don't forget Chromebooks, which run very smooth even on the outdated Cotex A15 CPUs.
You do realize that comet lake does not use atom cores (silvermont/goldmont) right? THOSE are more in line with high end arm processors IPC. But coffee lake (and by extension comet lake) are much higher IPC.
Remember the target applications here. We're talking phones/tablets (which are optimized to be multi-threaded, and aren't expected to run anything computationally complex), netbooks, embedded devices (where responsiveness isn't important), and robots (which are application-specific and therefore can be programmed to maximize the CPU's resources).
But that is not the target application. This is the first sentence of the article: "You can spot two new laptop processors as of today"
Yes, these chips (including the ARM ones) are also used in low-end consumer laptops - don't forget Chromebooks, which run very smooth even on the outdated Cotex A15 CPUs.
This is just not true. Anyone who has actually tried to use one of these chromebooks can confirm how ridiculously slow they are compared to the x86 options. Anyone I know that has tried to use one for more than 5 minutes has wanted to throw it out the window.
You do realize that comet lake does not use atom cores (silvermont/goldmont) right? THOSE are more in line with high end arm processors IPC. But coffee lake (and by extension comet lake) are much higher IPC.
You do realize not everything I linked to were Atoms, right?
As far as I'm aware, the only significant differences between Comet Lake are the faster cores, beefier cache, and more instructions. Only 2 of those things affect IPC. But considering the applications expected to run on these systems, none of that is going to make a big difference. The average workload is going to be things like web browsers and media playback, which ARM is perfectly capable of.
But that is not the target application. This is the first sentence of the article: "You can spot two new laptop processors as of today"
I assure you, the CPUs I'm questioning (not the entire list of them) will be used in all of the applications I mentioned, except maybe not robots. Also, netbooks are laptops. So yes, they will be seen in laptops, but that doesn't mean that's the only place you'll find them. Many of the Intel NUCs for example are using mobile CPUs, but they're obviously not laptops.
This is just not true. Anyone who has actually tried to use one of these chromebooks can confirm how ridiculously slow they are compared to the x86 options. Anyone I know that has tried to use one for more than 5 minutes has wanted to throw it out the window.
I have used one of those Chromebooks and it runs perfectly fine. No stuttering, no lagging, no struggling to do anything. This is a model from I think 2015 and it's still running fine. For the typical power user, it isn't sufficient, but power users aren't going to buy a Chromebook (which is why this model isn't mine). You need at least 4 cores and 4GB of RAM for an OS like that to run smoothly. Remember - Chrome is pretty RAM intensive, and a lot of ARM-based devices up until very recently don't give you much memory.
You do realize not everything I linked to were Atoms, right?
Most of them were. Which means you thought they were relevant somehow if you decided to include them. One of them was a 1.4GHz haswell celeron. Which actually helps prove my point because in that benchmark it was neck and neck with the snapdragon 855. The lowest comet lake cpu here should be about 60% faster than that (35% higher clock rate, 20% higher average IPC). And considering the snapdragon 855 is considerably faster than the Amlogic S922X (1.8GHz vs. 2.8 GHz, cortex A73 vs. A76 cores) that you linked the gap will be even larger. Likely 2-3x. You cannot expect a 5w chip to compete with a 15w. ARM chips do enjoy a slight efficiency lead in the low end but you can't seriously expect them to be able to compete with x86 chips using three times the power.
schmidtbag:
the only significant differences between Comet Lake are the faster cores
So then.....I was right? The entire point of this argument was that you said that they weren't.....
schmidtbag:
But considering the applications expected to run on these systems, none of that is going to make a big difference. The average workload is going to be things like web browsers and media playback, which ARM is perfectly capable of.
Oh I see. So now you've changed your position from "You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage" to "they're fast ENOUGH, you won't notice the difference".
Why can't you just admit that you made a factually incorrect statement? Is it really that hard? Why waste all of this time being stubborn when you KNOW that you're wrong? You've looked at the same data I have so you have to know that you're wrong by now. Why keep changing your argument to keep this charade going?
schmidtbag:
I assure you, the CPUs I'm questioning (not the entire list of them) will be used in all of the applications I mentioned, except maybe not robots. Also, netbooks are laptops. So yes, they will be seen in laptops, but that doesn't mean that's the only place you'll find them
What do you base this on? Intel has historically aimed their 15w U series mainline chips at laptops. Their marketing material for comet lake also shows them being aimed at laptops.
schmidtbag:
Many of the Intel NUCs for example are using mobile CPUs, but they're obviously not laptops.
Yes the whole idea of NUCs is essentially to put laptop hardware in a desktop box. But that's not exactly an embedded system/application now is it?
schmidtbag:
I have used one of those Chromebooks and it runs perfectly fine. No stuttering, no lagging, no struggling to do anything. This is a model from I think 2015 and it's still running fine. For the typical power user, it isn't sufficient, but power users aren't going to buy a Chromebook (which is why this model isn't mine). You need at least 4 cores and 4GB of RAM for an OS like that to run smoothly. Remember - Chrome is pretty RAM intensive, and a lot of ARM-based devices up until very recently don't give you much memory.
Yeah I'm sorry but I don't believe you. Maybe your standards of fast are a bit different than mine. But anyone here can walk into any best buy and try out the rockchip/mediatek chromebooks compared to the x86 offerings and see how much slower they are. Back when I worked for an MSP clients would get so mad at them they would start yelling at them and beg us to get rid of them. Lots of youtube videos exist also showing direct comparisons. An easy way to do this is to just open a few apps and switch back and forth. It will stutter on the arm chips, guaranteed.
Most of them were. Which means you thought they were relevant somehow if you decided to include them. One of them was a 1.4GHz haswell celeron. Which actually helps prove my point because in that benchmark it was neck and neck with the snapdragon 855. The lowest comet lake cpu here should be about 60% faster than that (35% higher clock rate, 20% higher average IPC). And considering the snapdragon 855 is considerably faster than the Amlogic S922X (1.8GHz vs. 2.8 GHz, cortex A73 vs. A76 cores) that you linked the gap will be even larger. Likely 2-3x. You cannot expect a 5w chip to compete with a 15w. ARM chips do enjoy a slight efficiency lead in the low end but you can't seriously expect them to be able to compete with x86 chips using three times the power.
They are relevant, because they're not beating out ARM. You're the one who made blanket statements that x86 is always faster, but, you keep moving the goal posts. Speaking of which, frequency doesn't affect IPC, which is something you keep bringing up. So, your cherry-picked results aren't proving your point.
It's not hard to find a test where x86 dramatically outperforms ARM, but my point is x86 isn't the unanimous win that you think it is, especially once you account for other factors other than raw performance.
Also, since we're mostly just talking about performance here, wattage isn't worth comparing to (and you should be happy I'm saying this since that works in x86's favor). Certain processors can certain perform tasks more efficiently than others. A higher wattage isn't necessarily a bad thing - AVX dramatically cranks up the heat but it's also very efficient. But ARM and x86 are so drastically different that comparing them in terms of wattage can't really be done. This is especially relevant since ARM has the ability to completely power off unused cores, which skews results.
Efficiency gets even messier once you start meddling with clock speeds. An Intel CPU at 3.5GHz has substantially better performance-per-watt than if it were clocked at 1.0GHz. Meanwhile for ARM, the efficiency tends to plummet when overclocked past ~2GHz (depends on manufacturer).
So then.....I was right? The entire point of this argument was that you said that they weren't.....
wtf? How desperate are you for a win right now? First of all, no, my argument actually has very little to do with what these CPUs were going to be used in. But even if that was the case, this discussion could be summarized as follows:
Me: This Celeron (keep in mind, that's the only CPU I've been referring to here) will be used in netbooks and
You: No, it's going to be used for laptops!
Me: Netbooks are laptops
You: So you admit it's for laptops! Aha! I'm right!
See how ridiculous you sound?
Oh I see. So now you've changed your position from "You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage" to "they're fast ENOUGH, you won't notice the difference".
No, I'm not. Look at ALL OF THE BENCHMARKS I linked to and not just the 2 or 3 graphs that suited your agenda. Notice how web browsers (which again, is what a lot of these CPUs are going to be doing for 90% of their life) have competitive performance. It's not just "good enough".
Also, where did I say anything about wattage? How is that helping your argument at all?
My point here is the applications where x86 is faster are irrelevant. Nobody in their right mind is going to buy one of these low-end CPUs and expect to do workstation tasks or heavy computing. If you only account for tasks people will actually be running, ARM is very competitive, which is what you seem to be disagreeing with (despite your lack of sources). It's not necessarily faster, but it's not the small fraction of speed you insist it is, and it isn't unbearably slow either.
Why can't you just admit that you made a factually incorrect statement? Is it really that hard? Why waste all of this time being stubborn when you KNOW that you're wrong? You've looked at the same data I have so you have to know that you're wrong by now. Why keep changing your argument to keep this charade going?
How ironic of you to say that. Unlike you, my argument has never changed (which the entire time was "for the same price, wattage, and performance of a Comet Lake Celeron CPU by itself, you could get a complete ARM SBC computer") and I actually have data to back up my claims. You have moved the goalposts several times and never linked to any graphs.
What do you base this on? Intel has historically aimed their 15w U series mainline chips at laptops. Their marketing material for comet lake also shows them being aimed at laptops.
Yeah I'm sorry but I don't believe you. Maybe your standards of fast are a bit different than mine. But anyone here can walk into any best buy and try out the rockchip/mediatek chromebooks compared to the x86 offerings and see how much slower they are.
Where did I say Rockchip or Mediatek? Just because you're shooting for the cheapest crappiest model, doesn't mean they're all bad. Some of the Samsung models are solid.
Sigh......look let's make this simple. This is the exact statement you made that I originally responded to:
"You can buy an entire ARM-based SBC more powerful than that for the same price and same wattage"
That is exactly what you said word for word. I asked you for proof since having been in the market for such a system recently I knew that was false. You linked a board that according to the benchmarks that you posted is considerably slower than any of these comet lake chips. Therefore that statement you made is factually incorrect. End of story. If you respond at this point I will just keep repeating this (I've done so already) until you respond to this point appropriately.
You: No, it's going to be used for laptops! Me: Netbooks are laptops You: So you admit it's for laptops! Aha! I'm right! See how ridiculous you sound? No, I'm not. Look at ALL OF THE BENCHMARKS I linked to and not just the 2 or 3 graphs that suited your agenda. Notice how web browsers (which again, is what a lot of these CPUs are going to be doing for 90% of their life) have competitive performance. It's not just "good enough". Also, where did I say anything about wattage? How is that helping your argument at all? My point here is the applications where x86 is faster are irrelevant. Nobody in their right mind is going to buy one of these low-end CPUs and expect to do workstation tasks or heavy computing. If you only account for tasks people will actually be running, ARM is very competitive, which is what you seem to be disagreeing with (despite your lack of sources). It's not necessarily faster, but it's not the small fraction of speed you insist it is, and it isn't unbearably slow either. How ironic of you to say that. Unlike you, my argument has never changed (which the entire time was "for the same price, wattage, and performance of a Comet Lake Celeron CPU by itself, you could get a complete ARM SBC computer") and I actually have data to back up my claims. You have moved the goalposts several times and never linked to any graphs. "hurr durr U is only for laptops" Do I seriously need to do all the research for you? Like really, this is getting pretty annoying how arrogant you're being: https://up-board.org/wp-content/uploads/up-xtreme/Datasheet-UP-xtreme.pdf https://www.mouser.com/pdfdocs/pico880.PDF https://www.aaeon.com/en/p/pico-itx-boards-pico-kbu4 https://www.cincoze.com/data/files/201907/Datasheet_DI-1000.pdf Note the marketing terms used for some of these. These are not laptops and they're not based on laptop motherboards. I made sure to only include products that offer U series CPUs. Where did I say Rockchip or Mediatek? Just because you're shooting for the cheapest crappiest model, doesn't mean they're all bad. Some of the Samsung models are solid.