Read more.Quote:
Initial results are lack lustre: it is sometimes slower than the Snapdragon 8cx version.
Printable View
Read more.Quote:
Initial results are lack lustre: it is sometimes slower than the Snapdragon 8cx version.
Intel should have run Cinebench.
Sucks that the Windows scheduler isn't ready, but if the hardware's out, what else can you review?
If that issue is curtailing performance presumably it'll shorten battery life as well.
no Cinebench but can it run crysis?
I just actually lolled at how poor Intel and their partners are executing things right now...
Tbh, it does look like either a misreporting of the preferred core to Windows or the Windows Scheduler was FUBAR in these tests.
However, from my understanding of core architecture, the Intel Lakefield was being overly aggressive in moving the cinebench test between cores even though it was running at the max.
From their followup:
11-12 hours in use with a 42Wh battery; the Acer Swift 3 with R7 4700U got 12 hours from 49Wh, XPS13 with i7-1065G7 got 12.5 from 52Wh. Swift is a 14", other two both 13.3".Quote:
The Intel model uses a bit more power in almost every scenario, which is also noticeable in the battery runtime. Samsung advertises a lower runtime, and our results so far confirm that: The load test runs for almost 4 hours (without outdoor mode), 2 hours less than the ARM model. The Wi-Fi test at 150 nits is still running, but it should result in about 11-12 hours compared to more than 16 hours for the ARM model.
As much as I enjoy Intel failing at the moment, I do have to wonder how much of this is down to needing better drivers etc.... It kind of has the same sort of issue AMD zen cpu's faced on release written all over it imo.
I've also got this feeling that if this was in MS's Neo with windows 10x it might function 'correctly' and we'd see some different results.
As much as I'd like to see alternatives, especially with x86/x64 instructions, to ARM in the low power space there's also a side of me that would love to see Intel or AMD (I know AMD had an arm license, I'd assume intel does too) make an arm based cpu with a x86/x64 'coprocessor', a sort of hybrid cpu as you will, because I think something like that could work really well with Windows on Arm etc.
EDIT: at one point there were rumors of AMD trying to combine ARM and x86/x64 on to one chip... that could be interesting to imo.
Usual Intel: prime objective is not to tread on the toes of the higher-end CPUs, not to make the best product within some constraints.
This is a fairly new kind of product and so I expect that scheduler optimisation will be the key to getting it working within its potential.
Really though, I don't see that kind of optimisation as something that needs massive end user input to achieve as it can be tested, monitored and refined in house.
As such, this strikes me as yet another product released before the software is ready because "oh well, we will just OTA the bugs out with time". Software is never released when it is ready these days, it's always in a beta state and then patched in a hurry. Problem they don't see is that reviews are done on this early software and rarely re-done with patches applied. I'm looking at mobile phone reviews and not knowing what state the device is actually in compared to the original reviews. As a result, I am in a position where I do not have, and can not get, adequate information about a purchase and so do not part with my money.
I have a nine year old laptop, a three or more year old phone and so on. My PC is the only thing I have upgraded recently and that is because I control the software side of things (more or less) and the hardware potential is fairly well known.
I will not make a purchase when I can't adequately compare with the competition because reviews are not done on finished products. Flaws can not always be narrowed down to hardware or software and so you're placing a bet on whether the flaws with your chosen product can or will be fixed.
Pretty mediocre performance from this great white hope...and sorry, but i'm not buying the whole windows scheduler story.
This product would've been thoroughly tested at all stages, by both Intel and Samsung.
I'm sure there will be a lot of future optimization on this new tech....but as it stands - not good enough.
Keeping everything on the low power cores shouldn't hurt battery life too muchQuote:
Originally Posted by edmundhonda
Mixing instruction sets is asking for trouble - it gets much harder to shuffle work between cores when you're running different versions of software on each core. Going HUGE.little like apple is a far simpler way to do it, now that they've shown you can inflate an ARM core to those levelsQuote:
Originally Posted by LSG501
That doesn't work. If it encounters a scenario where it works only for x86, it'll just fail since you have no guarantee what core its running on.Quote:
Originally Posted by LSG501
Even x86 cores, for ISA parity Intel disabled Hyperthreading and AVX on Sunny Cove, since Tremont doesn't have them. Now imagine doing that for a completely different ISA, with one emulating it! It's an unrealistic suggestion.
"It supports a max 8GB of RAM, with 34GB/s bandwidth"
Not only does it support 8GB of RAM, it embeds 8GB of RAM with its own high speed, low capacitance connection to the core.
One question that I have is whether that 7W power rating includes the RAM.
It seems to me a great advantage that a small chip like that could include the RAM and the five processors.
Intel did say that their LTE modems were designed into the first three systems. Perhaps it is an option.
Intel also stated that they have a lte modem chiplet that can go onto that Lakefield chiplet stack, so I suspect we'll hear about a new part with that configuration next year.
Intel also partnered with MediaTek to supply a specified form of 5G modem next year. It would be interesting to know if the specified form is a Foveros chiplet to Intel's specification.