Intel demonstrates a 28-core processor running at 5GHz
engadget.com
engadget.com
Interesting how only when AMD starts shipping something competitive with their chips, does Intel roll up its sleeves and get busy on breaking new ground. There almost seems to be a correlation ...!
Now I am not completely sure they are just sitting in a warehouse though. I understand Volta is currently in mass production and will launch around July/August.
Last year I went looking through consumer-class CPUs looking to upgrade. They are more or less all a very unimpressive 4 core, 8 thread, 3.0 ghz. Maybe they have a tiny bit more cache. There is absolutely no reason to upgrade.
Progress on consumer-class CPUs this decade has been utterly anemic. The only one I've seriously been wowed by is the Ryzen Threadripper. Now all of a sudden, Intel is readying better CPUs. Entirely coincidental, I'm sure.
In a way, that was good thing, though. For the first time in my memory, a five-year-old PC is not a piece of junk I am just too lazy to get rid off, but a machine fully capable running current software (at least after install more RAM and an SSD). (Within reasonable limits, of course, I am not talking about high end games or such.)
The fact that CPUs did not get faster (not much, at least) also meant that software could not afford to get slower, which greatly extended the useful lifespans of older machines.
No, it's not the mid 90's anymore where we were tripling performance every time we upgraded. There's still progress.
I went from a Nehalem i7 3.20GHz extreme hex core, to a kabylake i7 7700k 4-core, and it's a night and day difference.
In every one of my use cases my 7700k outperforms the older CPU.
A Sandy Bridge 4-core, 8-thread, 3.0 ghz CPU from 2011 would be something like a i7-2960XM. It's actually 2.70 ghz base frequency, but 8MB cache and 55W.
A comparable modern CPU (same cores, threads, base frequency, cache) is a i7-8559U. This has a TDP of 28W.
The old processor would scale up to 3.7 ghz while the new one scales up to 4.5. And the new processor has 1.5 times the memory bandwidth. And I expect the GPU is a lot better.
The new processor probably has somewhat better instruction throughput at the same clock rate. All that said, I agree this is not the 90s and we're not getting 10x or more improvements in speed over the course of 7 years here... The new processor _is_ better; just not enough better to be worth paying $400 for unless you're getting a new computer anyway for other reasons.
(Note that these are both "mobile" processors, fwiw; I assume that's the comparison point if we're talking consumer-class, because consumers don't really buy desktops.)
In order to maximize their yield and make consumer processors an economic reality, Intel tests every single chip that goes off the assembly line and cut the "fuses" of the regions that don't pass quality control. A perfect chip becomes an i7 (or an i9 or even sometimes a Xeon) while the less perfect chips become i5s and the ones with the most errors become i3s with most of their cores disabled.
As other posters have noted: that 2600K CPU part you bought in 2011 is the Intel Extreme line, which are the top of the line i7s that have not only passed QC on their cores but have also been "burned in" or tested for stability at the higher end of possible operating frequencies for that generation. You're literally comparing the best of the best from 7 years ago to stuff that just barely missed the dumpster.
For example, out of the dozens of desktops I've built for my friends, family, and me, all of the -K processors that I've used are still alive today - including an i7-965 Extreme that's almost ten years old now. My experience with i3/i5 is much worse.
AMD's Zen arch is an 8 core die, and they get very high yields out of it. Even the downmarket parts mostly or entirely had 8 working cores before fusing off.
Intel's big 28 core parts are (at least historically) a huge single die, with very very low yields. They charge a huge dollar amount for the parts because they cannot afford not to — the manufacturing costs are very high. There is some speculation that this 5GHz part is actually a bridged smaller die design, like AMD's Zen platform. Intel's bridge design is called "EMIB," if you want to google that speculation further.
No. The 2700K was a high end consumer CPU that could be bought for about $300, and didn't need a crazy expensive motherboard or socket. It wasn't extreme (also note the lack of an "X" in the model name).
(FWIW, TR is a very upmarket part compared to a 4-core SB and you will pay a large premium for that, especially for the motherboard. However, even an 8-core Ryzen 1700X would be a huge upgrade over a 4-core SB and the 4- and 6-core Ryzens look very affordable.)
They don't go up to that insane frequency, but still. There's plenty of opportunity to shop around if you want really beefy systems, which is a very refreshing state of affairs compared to just two years ago.
The Epyc 7501, 7601, etc, are all 32 cores and 64 threads.
Presumably they were too expensive (manufacturing, power, cooling, etc.) since they're not ubiquitous. IIRC the "barrier" to going any higher was the speed of light across the chip: running faster than 5GHz would require the output to be causally independent from the input!
Moore's law doesn't include capital investment as a factor.
Initially, it could be mostly ignored as it was swamped by available profits (at least for microprocessors, less so for memory). More recently, I believe it's been matched by increased volume of semiconductors used throughout the global economy.
But at some point we'll reach saturation. And costs to shrink feature size will continue to increase exponentially. So it simply won't make financial sense to continue.
(Unless non-silicon, non-electrical, or radical patterning technology hits a breakthrough)
"Intel stated in 2015 that the pace of advancement has slowed, starting at the 22 nm feature width around 2012, and continuing at 14 nm.[20] Brian Krzanich, CEO of Intel, announced, 'Our cadence today is closer to two and a half years than two.'"
https://en.wikipedia.org/wiki/Moore%27s_law
"Intel disclosed in a regulatory filing that it is slowing the pace with which it launches new chip-making technology."
https://www.technologyreview.com/s/601102/intel-puts-the-bra...
Moore's law for GPUs, on the other hand, hasn't slowed down yet because it's not as close to the wall.
https://blog.openai.com/ai-and-compute/
https://www.pcgamer.com/nvidia-ceo-says-moores-law-is-dead-a...
I am just afraid Intel will once more use its deep pockets and questionable business practices to crush AMD just when things were starting to get interesting again.
EDIT: Typo
This sounds extraordinarily doable to me. I'd be very shocked if this was anything but an off-the-shelf SKL-SP die with consumer fusing.
[1] And given that it's a server part, this calibration is inherently more conservative than you see in desktop stuff.
Also, while not AVX-intensive, CB does use some AVX, which usually draws more power (and thus limits clock more) than pure integer or even ordinary FPU computation.
Jumping from 3.8 GHz single core turbo to 5 GHz 28-core turbo is a big jump in power.
Those boxes on stage sure don't look air-cooled to me. The plumbing seriously resembles that of a phase-change setup, which would actually make sense here as an extremely low operating temperature would help tame the leakage power tremendously.
In summary though, it ain't gonna be cheap to buy or cool.
it is simply not practical to boost freq from 2.5 base/3.5 turbo to all cores 5Ghz.
If this had a different core count from the 8180 I would agree but I wouldn't expect Intel to do a completely separate design just for a model that will need to be highly binned regardless.
That or Intel has made some serious magic happen at 14nm++++++++++.
[1] https://www.anandtech.com/show/9390/the-amd-radeon-r9-fury-x...
will be interesting to watch what AMD is going to offer using their new 12nm toy. ;)
Possible IPC enhancements from architecture optimizations of Zen2 + better clocks = first time since over a decade that AMD beats Intel in single threaded workloads? Due to their better SMT implementation (?) they seem to beat Intel in multithreaded scenarios already with same core count and around the same clock (at least in Cinebench).
Though I don't think phase-change is really suitable for a consumer CPU. I would think it's just going to be bundled with a big integrated water-cooler (a bit like AMD's FX-9590 CPU which was a 220TDP part)
300W??
http://www.legitreviews.com/wp-content/uploads/2017/09/overc...
Try 800W.
They also had the 28 core Cascade Lake on the showroom floor with their partners with normal water blocks.
https://videocardz.com/newz/asus-dominus-spotted-with-28-cor...
Air cooling is however quite likely out of the question if you want to keep your hearing.
I think air cooling is 100% out. Using the current generation LFA3647 Xeons, you pretty much need dual 8180 kind of setup to get Cinebench score of around 7500, that is 410w TDP for two processors. With the same 14nm, squeezing such computing power into a single chip is not going to lower the TDP. We are about to see the hottest (300-400W) Intel processor ever released.
Perhaps we should be more worried that it's only 5.something times faster than an 8yo CPU with fewer than half the cores.
https://s.aolcdn.com/hss/storage/midas/c18d803c00f096b8d7b57...
Before you get too excited, the current generation of 28-core xeon running at 2.5G costs you over $10k each [2]. I won't be too surprised if Intel sets the price of a 5G 28-core Xeon at $15-20k.
That being said, having more choices is always good. Can't wait to see how AMD is going to respond to this using their Epyc line of chips.
[1] https://www.anandtech.com/show/12893/intels-28core-5-ghz-cpu... [2] https://ark.intel.com/products/120496/Intel-Xeon-Platinum-81...
Or is this like AMD's Threadripper any EPYC, and they are duct taping together multiple CPUs in a single socket? Eg, 2 x 14 cores CPUS connected with QPI or EMIB ? 4 x 7 cores? I wonder if AMD patented this technique..
IIRC, according to Intel's slides, AMD used glue.
In my experience, cores getting blocked for instructions is not commonplace. They can block waiting for data if the computation is memory-bound.
Edit: These are the specs for the [Skylake 8180](https://ark.intel.com/products/120496/Intel-Xeon-Platinum-81...). These chips are nowhere close to 5GHz though.
Normal RAM can handle 1.35V typically and often goes up to 3200 MT/s (on AMD systems) or 3600 MT/s (on Intel systems). That's easily +20% to +35% more memory bandwidth. With lower latency and better overall specs to boot.
And if your 3d render has a pixel-error on frame 50, delete the .png file and rerender frame 50. No biggie.
As for me I'm just annoyed by the fact I have never seen a laptop that supports ECC (as for now it seems there actually are a couple of models e.g. ThinkPad P70). I personally don't really care about the RAM bandwidth, I'm still using DDR2 anyway, but I would love to use ECC.
If your 3D rendering has any pixel errors, replace the offending hardware. If your RAM is bad in one place, you need to replace it, ECC or not.
The RAM chip itself would do any computation if needed instead of the CPU, and if it needs to give interrupts to warn about error conditions, it could output it as an electric signal you can hook up to anything you want yourself (including to the CPU in some way, like via USB or something lower level, or otherwise a warning light or beep or so)
There can't be many people in the world who will pay the 10k-20k price tag. How many people have the pull with their employers to justify it?
The production run must be insanely low or i just don't understand the marketplace around high end chips.
If you're a cloud provider, it also makes sense to have very large capable machines to increase your density.
Power consumption in datacenters is typically a significant cost of total hosting. (I pay the same as what a server costs in power consumption per year, per server)
So if you can have double the compute, with the same power draw, it's worth double the price in the first year alone, typically servers have a 3-5y life-cycle. So you can extrapolate that to "a server that is double the cost but the same power draw can be worth 5x as much as a slower CPU". but that's the cut-off..
For many servers the flat rate fee of: the license of anything that runs; the motherboard, the drives, the memory, the PSU, the cabling, the space, the power consumption means that it can be economical to egregiously spec out on CPU; if that's the bottleneck which is the case in a lot of virtual or build environments.
To put hardware prices to perspective, think how much it costs to run even a small development team. If hardware makes them more productive and/or happier, most of the time you should go for it. Sad too see expensive people stuck with mediocre tools.
Another poster suggests the TDP is over 300W.[1]
For a home computing device component, that's quite the pricetag. With the amount of heat it generates, maybe Intel could market it as a combination processor and space heater?
our current 2 x 14C workstations are $10-12K - and we're just a lowly no-pull devs in a BigCo not known for any lavishness toward low level rank and file.
How did we manage to break 28 Core all running at 5Ghz barrier when we were previously hold up even 8 Core running at 4.4Ghz at the same time.
No. Electron "developers" may say that. The vast majority of us, do not in fact say that.
I've yet to actually see any argument directly from someone involved, in favour of electron apps besides "it was just or a web app" or "it's quicker/cheaper this way".
If I throw a bucket of house paint over your Honda, that's also quicker. That doesn't mean the result is really what anyone would consider 'acceptable'.
But, I think my pedantry does not dissuade from your actual point; making a modern MacOS program requires that you have a macOS machine- It also precludes learning a language that typically only runs on that platform (I know swift is cross platform but none of the windowing libraries that you'll use are).
Making tailored software per platform is tedious, making the same program multiple ways so that it can be native is just time consuming if nothing else.
Electron is an ugly way of fixing that; but it does the job of fixing it- Qt is an ideal solution in theory but it's obviously too much of a burden on developers to adopt.
I only hope it's from the second half of 2013: https://www.macrumors.com/2018/06/04/macos-mojave-supported-...
Eh? No it doesn't.
> When you discover you can no longer develop on your 2012 machine
The cutoff would be either 2008 or 2009 depending on line.
16 GB DDR4 cost $200 now. This is not low-end, this is reaching premium levels. I got an entire mid-range smartphone for $165 last month.
Seriously why have the video float? If I’ve scrolled down it’s probably to read - like wtf is the intended user interaction goal there?
For one it is opt out, which isn't allowed. It also claims that you need to accept to continue using the site, which isn't allowed, then it's countless steps required to opt out (if you even can, I got lost and accidentally allowed everything a few clicks in).
This is probably worse than just pretending GDPR didn't exist. This is a deliberate and disrespectful fuck you to every single visitor.
Example: https://www.anandtech.com/show/12875/intel-announces-the-cor...
If this 28 core monster really is released I'm curious what it will cost. The 18 core i9 costs ~2000 dollar, the 28 core Xeon with much lower clocks (but with ECC etc.) ~8000.
The power requirements will be even more interesting looking at the 18 core numbers when overclocked.
It is also not illegal to require visitors to clarify to which data processing they consent to (GDPR only says that if you require consent for processing that is not necessary to provide the service, then that does not count as consent for GDPR purposes).
As I interpret it if you don't actively checked that means that you don't opt out from it. Which really makes sense considering how the sole purpose of the entire feature is to confuse and tire people into submission.
The whole thing is opt-out, since it clearly states that pressing "Ok" (the only button that dismisses the dialog) is accepting tracking.
Which is good becase that saved me a chunk of my life on Engadget who clearly have no respect for the readers.
1) Manage Options
2) (Body font sized link) Manage
3) (Body font sized link) Show
4) Scroll down
5-100) Click each toggle individually (select all button is currently broke) (It is intentionally ambiguous which is the 'correct' selection)
101-150) Do the same on other page
151) Click done
152) Click OK (hoping it's saved, it doesn't feel like it has)
This is much harder than it is to opt-in. I'm pretty sure this against the spirit of the GDPR.
Even if it isn't against the GDPR. It's a dick move.
Intel's 18-core i9 CPU is still impressive, but that's last year's news. Today at Computex, Intel SVP Gregory Bryant demoed a 28-core processor running at 5GHz. It's the first time we've seen a single socket desktop CPU cram in that many cores, and it's certainly powerful. In Cinebench, it achieved a score of 7,334 on stage, something that's only bested by multi-processor systems. Better yet, Bryant says it's not just a concept, as Intel plans to sell the 28-core CPU by the end of the year. Time to start saving.
That's what I've been doing for non-compliant sites.
- If your browser doesn't have this option yet, write its developers, product managers, or CEO, voicing your most humble request that they try do something that does work, at least with some pages some of the time, to try to proactively automatically agree to GDPR and cookies. (Or automatically disagree with them, discard cookies, whatever.)
I promise you that it is possible technically to do something that works on at least 1 page, that other web sites could mimic the interaction of and have it work on them too.
Sites won't all support browsers' solutions but if browsers can automate something, then some web sites can adapt to it. (By using the same interaction, names, etc - i.e. legally clicking for the user, or so that the web site can't tell if the browser sent that automated per the setting chosen by the user, or if the user manually clicked. Even if they can only automate ones that closely match certain element names etc it will still be a step ahead.)
But they realise that almost nobody actually wants the tracking, and that does not align with their interests.
They are currently just trying the waters, see what they can get away with, most certainly already have a backup plan they will enable at the first sign of trouble.
The GDPR is very different : it does not prevent offering legit features by itself. So I often happen to leave a site if I'm not satisfied by their GDPR implementation and they try to force tracking on me, I won't sell my privacy that cheap if I'm given the choice. That's something I couldn't do with cookies and I just was accepting their notice blindly.
By the way there even is a simple standardized way to know if I give my consent for tracking me or not without asking - the Do Not Track HTTP header field. That's a pity Microsoft has screwed it by turning it on by default. I wish GDPR would just require every website to respect this flag.
I'm pretty sure almost every reaction and e-mail that I have got due to GDPR is against the spirit. I had never thought it would be allowed, much less massively adopted, to "opt in" 100% of your existing userbase by fleshing out your existing Terms of Service and that's that. I thought it would require a new contract/agreement with the user to be signed/clicked "yes" to. I swear by far most have just somehow pulled me into their new "more privacy aware" system.
# of cores: 28
# of threads: 56
Hyper Threading FTW!
AMD had a 5Ghz processor in what, 2014? Clock speed isn’t everything...
https://www.anandtech.com/show/12893/intels-28core-5-ghz-cpu...