Intel still doesn't know what is causing Core i9 desktop CPUs to crash

Maybe electromigration is showing his ugly head again, but please let's not forget why it happens, pushing the silicon too damn much. The decision of factory settings without a reasonable target (power, voltage, temps, amps, whatever) from Intel just to look better on benchmarks now is spanking them, even when gen 12, Alder Lake P-Cores with a power target actually were pretty good, but not so tempting from the benchie numbers.
 
HardReset said:
Why we would want to get rid of x86-64? FPU side already has basically no x86 (x87) stuff because x86-64 compilers only create SSE2 or higher FPU code (x87 will work but compilers never created such code). Only major problem is "only" 16 GP and FPU registers but it can be solved with new instruction set.

Only integer side still uses x86-based instructions but again, it too can be solved with new instruction set that will some time replace x86-64.
Click to expand...
it's a messy, complex, instruction set with lots of side cases and oddities. If you get away from Windows (well, even they are looking at ARM now but...), Linux distros give you 90% of your software on even the fringe platforms like Alpha, PA-RISC, when the had Itanium distros, etc. that they support. And it's closer to 98-99% software coverage on ARM and probably on MIPS (that's not counting running things under emulation, which x86/x86-64 emulation on ARM is quite good now.)

I'm running a Coffee Lake desktop, Tiger Lake notebook, and had a Ryzen 3450U notebook before that; it's all x86-64. But I'd be missing out on nothing if it was an ARM, I could run steam and wine games in x86/x86-64 emulation and get plenty of performance, with literally eveyrthing else being native.

But I had an Acer Chromebook 13 with a Tegra K1 (quad-core ARM + 1 "little" core for the BIG.little setup; and a roughlty GTX650-equivalent GPU), I threw Chrubuntu on there and that thing got 22 hours battery life under real use; 12 hours if I ran video encoded and whatever and kept the cores at 100% load. It ran a full Linux desktop VERY well, the video encodes were fast (I had a Sandy Bridge desktop at that point and the Chromebook cleaned it's clock). I'd LOVE to get an ARM notebook for my next one! The GPU "could" be a weak point, but apparently the Qualcomm GPUs are direct descendents of the whatever mobile Radeon was in development when ATI sold to AMD, and have excellent driver support in Linux (Mesa Gallium's "Ardreno" driver, Ardreno is a anagram for Radeon..) and apparently quite good in ARM Windows too.)

If I have a desktop, I'd have lower power consumption, quieter, and less heat production. Same in a notebook but that also gives you the benefit of massive battery life increase.

(Side note: A few vendors make ARM workstations. So someone asked "This chip you're using, the GPU is not very strong. What do you do about that?" "It's a workstation, it has PCIe slots, we threw a 4090 in it." Maybe Windows doesn't have them yet, but Linux Mesa Gllium drivers cover AMD and Intel (kind of perverse to have and ARM CPU with Intel GPU but someone tried it and it worked!), and Nvidia has had Linux ARM drivers for many years. If Windows for ARM doesn't, the Linux and Windows Nvidia drivers are reportedly like the same core with just an OS-specific wrapper, so presumably they are most of the way there toward gettinga Nvidia Windows ARM driver if they wanted to.)
 
Last edited:
hwertz said:
it's a messy, complex, instruction set with lots of side cases and oddities. If you get away from Windows (well, even they are looking at ARM now but...), Linux distros give you 90% of your software on even the fringe platforms like Alpha, PA-RISC, when the had Itanium distros, etc. that they support. And it's closer to 98-99% software coverage on ARM and probably on MIPS (that's not counting running things under emulation, which x86/x86-64 emulation on ARM is quite good now.)
Click to expand...
Who actually cares anyway. All CPUs are basically RISC internally, it barely matters what sets are used outside CPU. Windows is another thing but Windows can support others too.
hwertz said:
I'm running a Coffee Lake desktop, Tiger Lake notebook, and had a Ryzen 3450U notebook before that; it's all x86-64. But I'd be missing out on nothing if it was an ARM, I could run steam and wine games in x86/x86-64 emulation and get plenty of performance, with literally eveyrthing else being native.

But I had an Acer Chromebook 13 with a Tegra K1 (quad-core ARM + 1 "little" core for the BIG.little setup; and a roughlty GTX650-equivalent GPU), I threw Chrubuntu on there and that thing got 22 hours battery life under real use; 12 hours if I ran video encoded and whatever and kept the cores at 100% load. It ran a full Linux desktop VERY well, the video encodes were fast (I had a Sandy Bridge desktop at that point and the Chromebook cleaned it's clock). I'd LOVE to get an ARM notebook for my next one! The GPU "could" be a weak point, but apparently the Qualcomm GPUs are direct descendents of the whatever mobile Radeon was in development when ATI sold to AMD, and have excellent driver support in Linux (Mesa Gallium's "Ardreno" driver, Ardreno is a anagram for Radeon..) and apparently quite good in ARM Windows too.)

If I have a desktop, I'd have lower power consumption, quieter, and less heat production. Same in a notebook but that also gives you the benefit of massive battery life increase.
Click to expand...
Why ARM? There is basically no difference between ARM and x86-64 on power consumption. As software support is much better on x86-64, I hardly see any reason to switch into ARM. You could have less heat output on desktop using AMD instead Intel.

Again, ARM is NOT "low power" and x86-64 is NOT "high power" by design. It's just what you want to have. And traditionally x86-64 has been directed on high performance and ARM has been for low power devices. But it does not have to be that way.
 
HardReset said:
Who actually cares anyway. All CPUs are basically RISC internally, it barely matters what sets are used outside CPU. Windows is another thing but Windows can support others too.
Click to expand...
All that junk wrapped around the RISC core increases power consumption, and there's really no way around it. (Well, there is, Intel and AMD used to have their chips die shurnk more than any ARM you could buy, when die shrink automatically saved some nice amount of power. That helped make up for it. But you now have ARMs being made with 5nm and 3nm processes too so that won't work any more.)

HardReset said:
Why ARM? There is basically no difference between ARM and x86-64 on power consumption. As software support is much better on x86-64, I hardly see any reason to switch into ARM. You could have less heat output on desktop using AMD instead Intel.

Again, ARM is NOT "low power" and x86-64 is NOT "high power" by design. It's just what you want to have. And traditionally x86-64 has been directed on high performance and ARM has been for low power devices. But it does not have to be that way.
Click to expand...
I beg to differ; ARMs are consistently considerably lower power than x86-64 setups. You can get some low-power Intel/AMD models that approach ARM in power consumption, but then you look at the speed and find they are a total slug compared to the ARM they are being compared to. You get the highest-performance ARMs, and they use a lot of power but much less than the Intel/AMD CPU with comparable performance.

That said -- if RISC-V wanted to take over, fine with me; MIPS? I doubt it, it's just not a high performance design, but sure. A resurgence of PowerPC? I used Linux on one years ago and it was nice. Those Chinese-built SPARCs get on the market? I doubt it'll be a thing but sure. Something else? I'd just like good cost, good performance, and (especially in my notebooks) good performance per watt. ARM is getting enough focus in the current market for hardware besides dev kits, one-off deisgns, and embedded stuff, to actually come on the market, thus my focus on ARM.

I'm NOT saying it's impossible! I'll be perfectly happy if AMD and Intel get that performance per watt up there while keeping nice performance! But, it'll be pretty difficult for them to do so; especially since many of the same tricks AMD/Intel can use to reduce power consumption, so can ARM vendors (the ARM vendors are probably already using those tricks, so that leaves room for Intel/AMD if they aren't using "every trick in the book" yet.)
 
Last edited:
hwertz said:
I'll be perfectly happy if AMD and Intel get that performance per watt up there while keeping nice performance!
Click to expand...
News flash! They already have!

AMD has proven time and time again in benchmarks published on this very site showing how they're getting within a hair's breadth of the performance of an Intel chip all while using vastly less power. 175 Watts vs nearly 400 Watts is nothing to sneeze at.
 
trparky said:
News flash! They already have!

AMD has proven time and time again in benchmarks published on this very site showing how they're getting within a hair's breadth of the performance of an Intel chip all while using vastly less power. 175 Watts vs nearly 400 Watts is nothing to sneeze at.
Click to expand...
Oh yeah AMD is definitely better... The 3450U I had, compared to the Tiger Lake I have now (i3-1115G4)... the Tiger Lake's per-core performance is nice, it gets about 2/3rds th CPU performance despite being a 2C/4T compared to the 3450U's 4C/8T; and about 2/3rds the GPU performance. But the i3 is like 35W TDP while the 3450U was 15! (And these chips came out within a month of each other, so it's not like I'm comparing significantly newer versus older chips here.)

I got too good a price on this Coffee Lake desktop to pass up, otherwise I'd definitely have gone AMD on it; and if I don't get an ARM notebook next, that'll definitely be an AMD (unless Intel radically turns things around in the interim power consumption wise.)
 
hwertz said:
All that junk wrapped around the RISC core increases power consumption, and there's really no way around it. (Well, there is, Intel and AMD used to have their chips die shurnk more than any ARM you could buy, when die shrink automatically saved some nice amount of power. That helped make up for it. But you now have ARMs being made with 5nm and 3nm processes too so that won't work any more.)
Click to expand...
How that junk increase power consumption if it's not used? That old junk is more and more driven on microcode and will continue to do so. That's pretty good way around it. So good that it rarely makes any difference.
hwertz said:
I beg to differ; ARMs are consistently considerably lower power than x86-64 setups. You can get some low-power Intel/AMD models that approach ARM in power consumption, but then you look at the speed and find they are a total slug compared to the ARM they are being compared to. You get the highest-performance ARMs, and they use a lot of power but much less than the Intel/AMD CPU with comparable performance.

That said -- if RISC-V wanted to take over, fine with me; MIPS? I doubt it, it's just not a high performance design, but sure. A resurgence of PowerPC? I used Linux on one years ago and it was nice. Those Chinese-built SPARCs get on the market? I doubt it'll be a thing but sure. Something else? I'd just like good cost, good performance, and (especially in my notebooks) good performance per watt. ARM is getting enough focus in the current market for hardware besides dev kits, one-off deisgns, and embedded stuff, to actually come on the market, thus my focus on ARM.

I'm NOT saying it's impossible! I'll be perfectly happy if AMD and Intel get that performance per watt up there while keeping nice performance! But, it'll be pretty difficult for them to do so; especially since many of the same tricks AMD/Intel can use to reduce power consumption, so can ARM vendors (the ARM vendors are probably already using those tricks, so that leaves room for Intel/AMD if they aren't using "every trick in the book" yet.)
Click to expand...
How many x86-64 CPUs are really designed to be low power ones? Basically almost every high performance x86-64 is designed to be "all around" cores, not just low power ones. So far there are no ARM cores that get even close x86-64 performance unless using tricks like integrated DRAM. AMD Zen4C is nothing else than Zen4 that do not need to reach around 6 GHz but around 3 GHz clock speeds. That gives 35% area reduction. And 4C is NOT designed to be low power one. It's just Zen4 for lower clock speed. So yeah, if AMD/Intel really want to make low power x86-64, it's not ISA that prohibits that.

Right. There have been many competitive ISAs but still x86-64 is king. That tells more than enough: it's not about ISA, it's about how you design CPU. That's why I cannot understand your ARM bias.

What are those tricks? Apple is using best node available, Apple is also using integrated DRAM and still struggle against x86-64 on performance side. At least Intel begins to use integrated DRAM and there goes one advantage. Node advantage may remain but if that is also gone, then what ARM has?
 
You must log in or register to reply here.

Similar threads

Daniel Sims
Intel provides recommended voltage settings for 13th and 14th-gen CPUs suffering crashes
Replies
2
Views
131
Skjorn
Skjorn
zohaibahd
$10,000 cooler designed with AI keeps Core i9-14900KF chilly at 7.5 GHz
Replies
15
Views
158
SilverCider
S
Scorpus
  • Locked
Intel is relaunching the Core i9-14900K... wait, it's the 6.2GHz Core i9-14900KS
2 3
Replies
55
Views
2K
erickmendes
erickmendes

Latest posts

Back