No one could have ever guessed resting on your laurels and price gouging customers because you were the only game in town would ever come back to bite them. /s
Isn’t this how every business works? If there is no competition, and if there are people willing to pay for the price they set, how is that a fault of the business?
That said, I think Intel has a long history of anti-competitive behavior and we need law makers with a spine to address and punish this type of behavior.
I don't think anyone at Intel has been patting themselves on the back for their 10nm failure. This wasn't their plan - cannon lake was supposed to be released in 2016, but now they'll maybe have volume production by the end of this year.
Intel can coast on corporate inertia for some time. They weathered the Netburst era alright from the Dells and Apples of the world being married to Intel.
I'm personally excited for actual competition in the CPU space. It seems like this entire decade has been in the doldrums with CPU growth. Someone needs to Conroe the market again.
"
Chris Gordon, a rookie cop in the Big Apple with an interesting sideline—hunting demons. But after rescuing a beautiful girl from a demonic attack, he finds life stranger than he ever thought possible"
I'm guessing that isnt it.
And there's a city in Texas??
Edit: Corner the market? That's the opposite of what you mean though...
And from the consumer end at least, that marriage may be coming to an end. Apple is probably moving to ARM over the next few years starting at the low end and with Microsoft trying to get Windows running well on ARM chips, Intel may be “stuck” with just selling high margin server chips.
Whether being stuck selling high end chips is a bad thing, I’m not sure.
The performance drop after their next security mitigation is supposed to be bigger than this price drop.
The release of their security workaround is being held back until something like one day after Ryzen 3000 release, presumably to cheat the benchmark numbers at the expense of users' security.
Do they then run the risk of somebody publicizing the security issues early? For example, if somebody already likes AMD better and gets angry at Intel for ignoring security risk.
Or is this additional mitigations for already public security issues?
Their costs are not going to come down anywhere near AMD's for products which can be sold for a similar price, at least with regard to their current product lines. AMD has really hit it out of the park from engineering on up to strategy.
If the performance-competitive AMD CPUs were priced at 15% below Intel, that would be reasonable. But... it looks like you'll pay about 45% less for an AMD equivalent to an Intel CPU.
Intel is still winning in single threaded work, last time I checked. AMD is winning in performance per dollar, and only because Intel gouges so damn much.
"The compilers", in plurality? They obviously control their own Intel C++ and Fortran compilers, but is there evidence of them "controlling" any others?
It's not like GCC or LLVM are blind to uarch differences.
Intel has their own compiler(s) but at least on the server side where everything is compiled using gcc or LLVM, both are optimized for both AMD and Intel. Was AMD behind a year ago in this regard? Yes. Even Windows just finally got more optimizations.
On the other hand, the benchmarks with AMD in the lead are obviously using compilers that 'exist' so really the compiler advantage Intel 'may' have is already taken into account. Additionally, the last few years of security issues has shaved off an additional 3% to 20% off of Intel's performance (compared to the 2-5% AMD lost).
The effect is that Intel has been loosing from both directions allowing AMD to be outpacing them in many ways.
I was pointing out that it takes more than an equal or better product at an equal or lesser price, due to many things baked into the current state of computers / servers.
Last I checked benchmarking tools are assembled by compilers instead of by hand so the supposed Intel advantage there is already baked into the results.
GCC has `-march` and `-mtune` flags for arch-specific optimizations. Just because the compiler is agnostic doesn't mean it can't have, or welcome, arch-specific optimizations.
One thing to note is that AMD performed their Intel CPU benchmarks without any security mitigations in place. Actual performance numbers in some applications might be higher than what was advertised.
I thought this business has technology cycles. It's just that AMD has missed a few and now Intel is at the mid-stage of it's current architecture. Wouldn't they be naturally back on top in a few years while AMD retools a new architecture family? I don't get the "intel is done for" reasoning. Cutting pricing on a product-line that is in mid to late life-cycle is expected in most other businesses including technology.
There are other arguments wrt Intel not being fabless that I get but those are not the same thing.
From my understanding the problem isn't with the architecture, but with manufacturing. AMD is about to release 7nm CPUs manufactured by TSMC, meanwhile Intel only recently sorted out its 10nm issues and 7nm is scheduled for 2021 if everything goes well.
Density is measured differently by different fabs, i.e. you can't do direct nm comparison, same as you can't do direct MHz comparison between different CPU vendors. Supposedly Intel 10nm and other 7nm are roughly on par.
My understanding is that Intel's manufacturing lead is basically gone now, and with Zen 2 so too will their uarch lead (IPC, AVX throughput, ...) disappear.
IIRC, Intel’s 10nm is measured differently than TSMC’s 7nm, so they are actually more or less at the same node. By some measurements, Intel 10nm is actually smaller than TSMC 7nm. I think some significant part of AMD’s manufacturing advantage also comes down to the fact that their “chiplet” architecture is proving to be easier to scale than Intel’s less modular designs.
> AMD is about to release 7nm CPUs manufactured by TSMC, meanwhile Intel only recently sorted out its 10nm issues
You can tell who the AMD fanboys are because they keep repeating this line even though it gets pointed out every time that the different fabs measure density differently and that Intel's 10 nm process is on par with TSMC's 7 nm.
Nowhere is his comment he said Intel 10nm is better or worse than TSMC 7nm. Not sure who is being fanboy here. At the time of my comment there are three comments repeating the same fact of how fabs measure density differently. I would say Intel fanboys is getting defensive.
Except TSMC is also on track with their 5nm plans for volume production in 2020 while Intel has been stuck on 10nm for years with "promises" of 7nm ...
No, Intel is not mid-cycle; they are still selling 2015 tech because they have failed to deliver on their roadmap for the last four years. In theory it's possible that they could have a great leap forward and catch up but it's more likely that it will take 3-4 more years to dig out of the hole.
I agree with this. About eight years ago, when AMD premiered their "bulldozer" architecture, it was clear that even their best technology was years behind Intel.
Now the situation has flipped, and Intel is years behind.
That's hardly satisfying. Price for an i5-8500 is higher than it was a year ago. They chocked the market while providing low supply and now are trying to fight back with laughable discounts. Thanks but no thanks Intel. My next upgrade will be an AMD.
52 comments
[ 3.8 ms ] story [ 108 ms ] threadThat said, I think Intel has a long history of anti-competitive behavior and we need law makers with a spine to address and punish this type of behavior.
I'm personally excited for actual competition in the CPU space. It seems like this entire decade has been in the doldrums with CPU growth. Someone needs to Conroe the market again.
They were cheating, though. https://www.cnet.com/news/intel-to-pay-amd-1-25-billion-in-a...
Google gives John Conroe (author)
" Chris Gordon, a rookie cop in the Big Apple with an interesting sideline—hunting demons. But after rescuing a beautiful girl from a demonic attack, he finds life stranger than he ever thought possible"
I'm guessing that isnt it.
And there's a city in Texas??
Edit: Corner the market? That's the opposite of what you mean though...
Edit2: Words
The processor that nearly killed AMD.
It wasnt until sandy bridge and bulldozer that things got really bad.
Whether being stuck selling high end chips is a bad thing, I’m not sure.
The release of their security workaround is being held back until something like one day after Ryzen 3000 release, presumably to cheat the benchmark numbers at the expense of users' security.
Or is this additional mitigations for already public security issues?
I'm really looking forward to EPYC Zen-2 parts.
Intel controls the compilers. The compilers can do special things on Intel and make better code.
Just having a faster processor doesn't matter if most compiled code is slower on your processor...
It's not like GCC or LLVM are blind to uarch differences.
On the other hand, the benchmarks with AMD in the lead are obviously using compilers that 'exist' so really the compiler advantage Intel 'may' have is already taken into account. Additionally, the last few years of security issues has shaved off an additional 3% to 20% off of Intel's performance (compared to the 2-5% AMD lost).
The effect is that Intel has been loosing from both directions allowing AMD to be outpacing them in many ways.
I was pointing out that it takes more than an equal or better product at an equal or lesser price, due to many things baked into the current state of computers / servers.
Of course it was riddled with (outdated) lists of packages that either
A: ran worse than with GCC (this was pre-Clang/LLVM days)
B: Didn't build at all
There are other arguments wrt Intel not being fabless that I get but those are not the same thing.
My understanding is that Intel's manufacturing lead is basically gone now, and with Zen 2 so too will their uarch lead (IPC, AVX throughput, ...) disappear.
You can tell who the AMD fanboys are because they keep repeating this line even though it gets pointed out every time that the different fabs measure density differently and that Intel's 10 nm process is on par with TSMC's 7 nm.
Now the situation has flipped, and Intel is years behind.