Intel Launches 11th Gen Core Tiger Lake
anandtech.com
anandtech.com
Anecdote: a family member and software dev just bought the latest macbook pro coming from about an 8 year old macbook pro. Apparently he doesn't perceive much real world performance difference between the 2 machines. He said single core synthetic benchmarks were higher, but not by much.
I'd love to see the new processors and RAM in a 2015 enclosure. Makes me wonder whether someone has attempted to hack it together.
I think much of the increase comes from going from dual core to quad core and perhaps not as much the single core gains, but still, I'd be curious what kind of software he's deving to not see much improvement.
If he's doing compiling or web-dev transpiling kinda stuff, I'm really surprised he's not seeing better gains.
Felt definitely faster than the 7200u I keep using, though I didn't get much of a chance to benchmark both. I also noticed that, despite having a much weaker battery (some 36W/hr mess), it managed to pull 6hs on a single charge, on merit of dropping to 1.2Ghz as fast as it could (though it quickly climbed to 3.4Ghz if required).
So results can be a little mixed. It’s clearly pulling more weight at lower clocks but lacks the high frequency performance of intel’s hella refined at this point 14nm process.
And that’s what makes Tiger Lake interesting. It’s all the good stuff Ice Lake has going for it, but now Intel is able to take off the training wheels.
It really depends on the workload. My old MacBook Pro feels the same as the newer model 98% of the time that I'm using it, but I really appreciate the extra RAM and double the core count when I need it.
Reducing build times from 30 seconds down to 15 seconds doesn't sound like much when you're not pressing the compile button very often, but it really does help improve my engagement and focus.
The newer graphics cards are also much better at handling large and high-resolution external monitors. The difference isn't pronounced if you're just using the built-in laptop screen, but start using multiple external 4K+ monitors and the faster GPU starts to shine.
1. If they previously had less than 4GB of memory and were often swapping
2. If they previously had a spinning HDD and moved to an SSD
3. If they previously had a really low end CPU like something outside the Core range, and moved to an i5 or better
Beyond that, all the other stuff is either a marginal performance gain or a convenience feature. The only other "must have" thing from the last few years that I wouldn't want to lose is a high res monitor, but that falls into the convenience category for me and not performance.
Power users like your family member got quantifiable benefit from performance upgrades over the years, but if they were already running decent hardware like a Macbook Pro, they didn't get any life altering upgrades. Taking 15 seconds to compile vs 30 is definitely faster, but you still have to sit and wait each time you push the button, so any habits that you learned from the 30 second pause will probably continue.
Examples:
https://github.com/Tecate/bitmap-fonts
If you're on 4K you can simulate how they look on 1080p with 2x integer nearest neighbor scaling.
The by current standards normal sharpness is nice too (1080 screens aren't really blurry, they're pixely).
I basically went back to 1080p because of this.
Eye resolution is ridiculously high. There's a reason companies started marketing 'retina' displays.
Small fonts will not always fall into discrete pixels which is especially true if the resolution is not very high. Which requires you to do antialiasing or mangle the characters.
Higher resolution displays allows you to render them more faithfully.
This improves both motion quality and latency. When poorly designed software makes you wait multiple frames, halving (or better) the frame time makes a noticeable difference.
I think jitter is more important than FPS and until software has caught up (likely never) I would rather cap the frame rate at 60 FPS and experience that somewhat consistently rather than have such disparate performances.
It’s the same thing with retina/HiDPI. It’s amazing when everything is retina, but all the random bits and pieces of UI, let alone entire softwares, that appear blurry make for a worse experience than a strictly low dpi UX.
However, you are definitely correct when it comes to GUI productivity applications. Drawing text in boxes is apparently much harder than rendering an immersive world in real time. The implications are too disturbing to consider, so I do not.
https://github.com/GNOME/vte/blob/master/src/vte.cc#L10543
It's another harmful effect of mobile-first design. A trivial power saving gives measurably improved battery life, but the horrible latency is ignored.
...if you're playing with v-sync :-P. I got a (rather expensive too) 165Hz monitor recently and the main benefit is to lessen the lag that Windows' forced composition (which also does v-sync) has. But in games i always played with v-sync turned off (i do not mind the tearing, unless it only happens at a fixed position i never really notice it) and the responsiveness was already very high. And on Linux and older versions of Windows where i could disable the compositor, i already had high responsiveness on the GUI too (if anything i feel like even with the 165Hz monitor, the Windows 10 desktop is still not as responsive as Windows 7 and previous versions of Windows were with the compositor disabled).
Though a large issue with that is how modern monitors work. I also have a CRT connected to an older computer here which can do 120Hz and the motion feel is considerably better to the point where i wonder what the hell was wrong with people in the mid-2000s to switch away from CRTs when in pretty much every other area except size and weight, the CRTs were superior - especially at the time (better image quality, better and true dark colors, better responsiveness/no 'response time', higher refresh rates, higher resolutions, variable resolutions with no scaling artifacts, etc). In fact a reason i decided to try a 165Hz monitor was because i recently connected that CRT to a machine that can do 120Hz and have some games to run at 120fps and wanted that experience for my main PC too.
(sadly i do have a feeling that if i hadn't used a CRT the last few years i'd be more enthusiastic about my new monitor since i do not think the currently technology can do much better - at least without going to something like that gigantic $3k OLED monitor though even that would annoy me for its size)
High refresh rate monitors tend to support freesync or Gsync. For minimum latency disable vsync, enable free / gsync. Only enable vsync if your fps is above the monitors refresh rate. Alternatively you can enable a framerate limit just below the refresh rate.
Even supports VR. '80s Cyberpunk matrix here I come!
pulls out a virtual gun to kill a process dead
Take the same idea and put it into Unreal Engine. Add further OS interactions like files, maybe a floating virtual window for web browsing etc.
The UX has to be done very well though otherwise you are a prediction by TV show Community:
The very, very first thing I did at my last company after migrating to a retina Macbook Pro was go into our monitoring systems and configure it to render the graphs at 2x size and then scale them down on the client.
A lot of software, especially websites, have annoying waits for Vsync that are not needed in them.
When you update styles that triggers a reflow or repaint, then you can introduce a wait for VSync in your website thats stalls flow until it is finished.
If you add 3 frames of delay at 60fps then it's 50ms of delay, and only 25ms at 120fps. It doesn't sound like a lot, but that 25ms can be the difference between feeling totally fluid and not, and a lot of websites are even worse than that.
Software can be IO bound, Network bound, CPU bound, memory bound.
Modern software is Vsync bound!
4. If they previously had an early generation of Core CPU like a Sandy Bridge that suffered great performance hit due to vulnerability, and then moved to an Ice Lake, or better, an AMD...
These days, it's more like if they previously had less than 12 GB of memory and were often swapping. Having a few dozen tabs open isn't that uncommon and interactive websites (Twitter, Gmail, Google Docs) can use an enormous amount of RAM.
https://www.cpubenchmark.net/compare/Intel-Celeron-N2830-vs-...
It took 700ms or so for my server process to even start on a r.pi - I'd noticed a brief delay on my laptop, but it was so awful on the pi that I ended up spending the whole weekend profiling and optimizing. Anyway, come Monday I pulled the changes onto my fancy laptop and they made a noticable difference in subtle responsiveness. Everything just felt faster. I probably wouldn't have even noticed if I didn't spend that time programming on the pi.
Facebook has "slow internet thursdays" or something, where once a week developers can opt in to experience their internal facebook dev environments in the same way it feels for people on slow connections in poorer countries. I think that sort of thing would be good for all of us. Programming in constrained environments makes our software work better everywhere.
There are several race conditions in many webpages related to waiting for video or images to load.
Tangentially my late 2018 i5 MBA feels sluggish compared to my 2014 MBP. Simple things like switching windows are noticeably slower and janky. Unfortunately my 2014 MBP has an no longer supported NVidia GPU by Chrome/Firefox/Photoshop so I may have to upgrade but I know it will basically feel no different than what I have except I'll have the touch bar which I'm not really interested in.
I believe there is an option in the Performance pane to throttle the CPU in Chrome.
Some games supposedly have been optimised similarly by running parts of it on a C64 or Amiga. I cant recall the game or studio but John Carmack comes to mind.
And heat up, unfortunately.
I'm completely fine with it getting loud and toasty when I push it, but the 5300M and 5500M in the 16" MacBook Pro ramp their memory clocks up to max as soon as two displays are on, even when idling. This results in a constant ~20W power draw that destroys the battery life, has the fans constantly audible, and makes the top case above the keyboard uncomfortably hot.
Apparently similar issues in desktop AMD cards have been addressed via driver updates, but considering whoever is responsible for updating Mac GPU drivers never actually does it, I'm not holding my breath.
Really? i7-9750H (from 2020) scores more than twice as high in cinebench single threaded than i7-3615QM (from 2012)
https://hwbot.org/benchmark/cinebench_-_r15/rankings?hardwar...
https://hwbot.org/benchmark/cinebench_-_r15/rankings?hardwar...
Dollar cost per compute, Watt-hr/compute, latency, io bandwidth, running on real world workloads. The process size is a red herring.
Any increases in single node performance are not red herrings and should be called out. This is excellent news.
I recently upgraded my monitor from a Dell U3014 to an OMEN 27i (it's gaudy and "gamer" oriented, but it was the best I could get in an afternoon at a physical store). That's an upgrade from a 60 Hz monitor to a 165 Hz monitor, and a significant (though not numerically measured so far) drop in input delay, some tens-of-milliseconds.
This has been the biggest improvement in "snappyness" that I've experienced since I moved to solid-state storage 10 years ago.
I'd say the same thing about upgrading from an i7-6800K (2016 Broadwell-E) to a Threadripper 3960X. Barely a difference (my main work is a ~250k loc c# solution, also a similar medium-ish sized Angular app).
Honestly, I think this speaks to the quality of Apple's hardware and engineering (8 years ago), combined with the lack of serious improvement from Intel lately.
That said, one thing I've noticed when doing IT and desktop support is that, unless someone's hardware is woefully inadequate, they won't notice a huge change going from slower to faster, but they will notice faster to slower after they've been using it for a week or two.
As one of may examples, set up an SMB3 connection to a network-attached storage device and then write a short program that uses two or three threads that walk a big director calling stat and lstat on every file. For large complex file hierarchies, Catalina (unlike Mojave) will start stuttering the entire UI and media playback.
"rclone" with --transfers=3 can bring a 2019 mbp to its knees, like Windows 3.1 writing to a floppy disk.
I'm in a not dissimilar position with my (older) personal laptop and my (newish) work laptop. Two things I notice:
* Battery life. My work laptop lasts longer on battery because it uses less power to do the same amount of work. Not really a problem 99% of the time when I'm plugged into an outlet. * Compile times. Work laptop is much faster at compiling, but this is really only noticeable on cold-builds. Incremental builds are fast enough in both cases.
We’re actually planning on running that gear longer and moving people with more memory needs to newer gear with more memory — it’s all about the maintenance cost curve. We can get much more memory now with the same density, which offsets the CPU suckage.
I just upgraded my 13" MBP from a 2017 i7 to the 2020 i7.
Single core benchmarks are basically the same but it's also running at 2/3 the clock speed, so much cooler. Plus two additional cores.
I also notice windows starting up faster than it used to before, but I know its because of the SSDs and caching. These 20% and 30% YOY improvements are simply not worthy to invest in because the returns are negligible for me.
AMD did a really smart thing with their processor designs. Instead of doing a large monolithic multiprocessor like Intel (which may be able to eek out some extra performance due to optimized wire placements), they use smaller chiplets. This results in requiring no manufacturing errors on a much smaller area of the die (which becomes exponentially less probable as the area increases). Even if AMD was using Intel's 10nm process, their yields would be better because of their modular designs that get soldered together after the lithography. Intel can still bin out some errors (the 10300-10900k are all the same chip with different parts turned off due to manufacturing errors), but the less modular design likely makes this less efficient.
That's my understanding, anyway -- correct me if I'm wrong.
That logic seems backwards. If you have poor yields you tend to favor higher-margin products (i.e. datacenter CPUs). This is a laptop CPU intended to sell at significantly lower $/mm2.
As far as chiplets: multichip packages are an ancient idea, and the industry goes back and forth on them. Both Intel and AMD shippsed multi-chip solutions way back in the day, the current age of integration is actually the anomaly. You win on yield but lose on package costs, and the decision as to which to use depends on the specifics of the market you're trying to target.
Certainly AMD would prefer to ship single-chip solutions and pocket the savings, but they can't. Likewise Intel accepts some loss of scaling because of the need to share die designs across the product line.
I say this having no idea what Intel's defect rate is right now, and I acknowledge fusing of bad sections can mitigate this a bit.
Since these are mobile processors your understanding is incorrect. What you've described is what AMD did on the server & desktop side. But on mobile, to hit the power efficiency targets, AMD stuck with the single monolithic die. And it's a decently large one, at that: https://www.anandtech.com/show/15381/amd-ryzen-mobile-4000-m...
So no AMD wouldn't yield better on Intel's 10NM+, not in this case. In the desktop usage they would, as their desktop CPU dies are way smaller than their mobile CPU dies. But the mobile die sizes are fairly equivalent between AMD & Intel.
> The fact that Intel's only new releases on 10nm are <=6 core low power products suggests their yields are still poor.
Their argument was that, had they followed AMD's approach, they would be able to produce desktop and server chips despite being yield-limited to "<= 6 core" per chiplet.
Wow. Is this a relatively new phenomenon, or something that's been the norm in processor mfg for a long time? I know that the Ks were usually just binned/higher tested chips so they'd unlock those and charge a premium for them, but the idea that they're disabling parts of the die and separating the chip offering that way is crazy to me. Are the parts being disabled redundant or are they reducing the instruction set space? I don't know anything about how any of that would work.
So no, nothing new under the sun.
wait, it's all just binning?
it always has been.
;)
https://en.m.wikipedia.org/wiki/Product_binning
Iirc, early Celeron were often just Pentiums with defective (and disabled) cache
Something that's different between software and hardware such as chips are that the chips that have things disabled are done so because those disabled parts actually don't work. If you only yield 2 out of 4 cores, it's easier to just make it a dual-core CPU rather than throw the chip away.
https://www.zdnet.com/article/upgrade-your-radeon-hd-6950-to...
https://www.notebookcheck.net/All-core-4-3-GHz-at-28-W-Intel...
Intel's 4 core parts getting within 3% of AMD's 8 core parts, 20% lead in single core perf.
This is not exciting, unless I’m entirely missing the exciting part of this, which is wholly possible.
[1]: https://www.notebookcheck.net/Every-AMD-Ryzen-7-4800H-laptop...
I think AMD are pretty safe from Intel for a while.
That’s not to say they won’t be threatened by other parties, but it’s clear Intel still thinks it’s fabs are it’s biggest problem.
(I.e. were I Lisa Su I’d be more concerned about the new ARM chips coming over the hill from Nvidia).
A nice tweak sure but also irrelevant as a consumer. It just results in one less component on the motherboard, it doesn't change any functionality.
> much faster clock speeds
Relative to what? Last year's 4c/8t 14nm i7-10510U has a higher boost frequency than any of the chips announced today, and it's also a 15 W TDP CPU. The 4c/8t i7-1185G7's 3ghz base frequency is nice I guess but barely higher than last year's 4c/8t i7-8569U at 2.8ghz base - and both are 28W TDP. Probably, anyway - the 11th gen is listed in the '12-28 W' category but as it's the highest base freq of the listed quad cores I assume it sets the top end of that range.
If there's a power improvement here, which is possible, it's not clear in the spec sheet, and I think nobody is buying Intel's marketing claims at this point. So we won't have reasons to be excited here until reviews hit, if a reason to be excited for this exists at all.
> greatly improved graphics
This is an great upgrade, but only for Intel. It's playing catch up to the Ryzen 4000-series, which means it's not exciting on its own for consumers. Exciting in that competition drives price wars & overall stronger ecosystem, but on its own? Super boring.
Ice Lake was released with higher IPC and lower clock (due to 10nm issue) but performs well compared to older higher clocked CPUs thanks to IPC. Now Tiger Lake is released in higher clock, why not exciting?
For iGPU, both AMD and Intel is mainly limited performance by memory bandwidth, I'm curious whether Intel releases a SKU with eDRAM.
I notice they're still going for their smallest core count chips first. No desktop, no HEDT, no Xeon. Nothing more than 4 cores. That's about half of their business that's still stuck on 14nm, and what should be their highest margin SKUs.
I think it's interesting/exciting because it looks like they're starting to get higher quality results out of their 10nm node. Ice Lake was decent, but it was limited on clocks and was clear they were trying to get something viable on 10nm out the door. Almost proof of concept for all the 10nm stuff they'd been talking up and failing to deliver on for years.
But these Tiger Lake parts actually looks like higher quality 10nm running at much better frequencies. It's Intel finally delivering on their 10nm promises.
The Xe graphics are also fairly interesting. Ice Lake Gen 11 graphics was an improvement over older integrated parts, but Xe is another solid step forward. As a laptop consumer I'm always happy to get more graphical oomph so that multiple high resolutions displays can be comfortably run when I'm docked into an at-home setup (which is kinda always in this COVID world).
All in all, none of this is mind blowing, but Intel has been so... whelming... these last handful of years, that I think it's fair to be a little excited when they've finally taken a legitimate step forward.
Intel announced 10th Gen CPU lineup with a new logo.
Meanwhile AMD is shipping 16 core CPUs to enthusiasts and 24/32/64 core CPUs to high-end desktop users. Getting 56 extra cores is something to be excited about; your 8 hour render is now a 1 hour render. Your 15 minute build is now a 2 minute build. Making your 8 core processor do one synthetic benchmark 10% faster is comparatively very "meh". People are expecting each Intel launch event to be something revolutionary like Zen 2, but Intel just isn't doing that. So people are consistently underwhelmed even though a computer with a 10th or 11th generation Intel chip is going to be quite adequate.
(As for nVidia, I am also not amazingly excited. The 30XX series feels like a refresh of the 20XX series; yeah they're cheap, but if you already have a 20XX your world is not going to change dramatically.)
https://semiaccurate.com/2020/09/02/intel-doesnt-actually-la...
I'm not good at English, What means this sentence?
Intel 10nm desktop CPU's are what I'm actually interested in.
Will have for 3rd party benchmarks to determine the real performance improvements. The single-core and GPU are certainly the highlights, but it's odd that even the top-end i7 chips still retain the 4 physical cores. AMD's 4700U is already at 8 physical cores.
I hope this delivers. I'd hate to see Intel go down in flames.
But it's entirely missing from this announcement. Seems like a regression.
For laptops (which are far more likely to thermal throttle), this is a huge deal.
That said, also agreed with everyone that core count isn't absolute and we should be looking at the clock frequency, etc. One metric I'm curious to know more is sustained performance - e.g. if the chip can only hit its turbo for <30secs, then that just isn't very good...
EDIT: ah, these are laptop chips, that's the catch.
Most systems won't come anywhere near that
[1] https://ark.intel.com/content/www/us/en/ark/products/208664/... Intel appears to have eliminated the concept of a "standard" TDP and now it sounds like the laptop maker can set it anywhere within the range of 12-28W.
[2] https://ark.intel.com/content/www/us/en/ark/products/201888/...
I have the i5-10210U which turbo boosts up to 4.2GHz. Intel lists its TDP as 15 watts, but if I stress test it in s-tui[2] it spends several seconds at 47 watts before being throttled due to thermal constraints.
1. https://www.notebookcheck.net/Intel-Core-i7-10810U-ThinkPad-...
With a lineup of chips aimed at fanless ultraportables, pay attention to their performance compared to the upcoming Apple Silicon chips, or more importantly, the Qualcomm chips that Microsoft and Samsung are putting in their fanless ultraportables. Intel needs to head off the rise of Windows on ARM.
I also find it interesting that the TDP maxes out at 50W. Is Intel entirely focused on the medium/low end or is there a problem with doing higher TDP chips using the "SuperFin" architecture?
In fact that’s been their strategy for a while from what I understand.
Is this:
-- a major improvement in power efficiency, heat?
-- major gain in speed (and who is the consumer concerned with this)
Seems like everyone is more concerned with GPU capability as the bottleneck for whatever application these days.
A bit, yeah
> heat
just enough to make a laptop 0.5mm thinner in order to have it PROCHOT throttle a lot of the time
> major gain in speed
Not noticeable by anyone. That's been the case for years, since everything Intel has is an improvement upon Haswell/Skylake.
> GPU
For anyone who cares about GPUs, these processors will be paired with some nVidia or (unlikely) AMD chips, so they'll be unused, just like most Intel IGPs.
Please note the lack of shipping dates and the positioning of these at "premier consumer" laptops meaning they can't produce enough still -- who the heck wants an 1500 USD IdeaPad? What's the point when the X1 Carbon Gen 7 is 1200 USD? You can spin that as "but that's previous generation" as if it meant anything at all. On the business side (14nm CPUs they can make plenty of), Intel hasn't produced anything new for three years now, they just change the labels so they announce something "new".
Ok?
Not a great time to be Intel.
https://images.anandtech.com/doci/16063/474551355-Intel-Blue...
I don't see any coherent fabric blocks — what looks like large cache sections to my eyes are entirely inside the "Core" blocks. Maybe they're just not drawn correctly. Here are some much older i7s, for example:
https://www.cs.uaf.edu/2009/fall/cs441/proj1/russell/images/...
https://www.notebookcheck.net/fileadmin/_migrated/pics/Core_...
Right, that was really my main question: besides the labeling (which is of course approximate), the positioning sure looks like each core would have its own "favorite" section of the cache. And being shared, it seems like this could make for some interesting performance behaviors.
Ah, interesting, this topology was introduced with Sandy Bridge. I'm just out of date: https://www.anandtech.com/show/3922/intels-sandy-bridge-arch...
https://cdn.wccftech.com/wp-content/uploads/2020/08/Intel-Ti...
Looks like ya, the L3 cache is "inside" the core blocks.
I have heard others with relatively recent laptops say they see the CPU go into power/thermal limiting if they barely exercise it with a few seconds of compilation. The power limit can be worked around with utilities, the thermal one not so much...
Ultimately, Intel have the resources to compete with Nvidia. If they want to they'll have to set realistic expectations though (Intel's revenue is more than AMD's, Nvidia's, TSMC's combined - it takes more than money though)
...still not as bad as what happened to Atmel's logo, however.
https://www.statista.com/statistics/272595/global-shipments-...
We can dream :p
I'd love it all in one package, fast single threaded performance when needed, fast multithreaded compiles, and ability to experiment with and benefit from AVX-512 specific workloads
Granted, workstation component pricing. Figure ~$650+ for a C621A chipset motherboard, and ~$2000 for the CPU without the massive (> 2TB) memory support.