- People who don't follow the corrections and follow up are now expecting them to release it (you can especially find a lot of people who says "intel will release it this year" in reddit for some reason). Now whatever intel actually releases will feel like a let down.
- By all account Amd only planned to release their 24 core threadripper 2 at computex, but after intel demo they moved to the 32 core model directly (with all 4 unit activated instead of 3 out of 4). Intel pushed the bar it has to compete with higher (also, good reminder that competition benefits the customer). Pricing wise TR2 is not even EPYC so intel also has trouble there.
- Additionally, Amd / Global foundry whispered that with their new process getting those chips at 5 ghz barrier in the not so distant future is on the table. Nobody had that barrier in mind a few weeks ago, and now all we know is Intel had to cheat to look like it can, while Amd is on its planned course to reach it. And this right after intel's numerous failure to get their new miniaturisation process to be on schedule the last few years.
- Once Asus and Gigabyte spilled the beans on how intel did it (aka it was overclocked, and the cooling unit was gigantic), intel came to their booth and gave them 30 minutes to hand back the cpu, ending the demo. Which moved it from "intel does a tech demo of what could be" to "intel tried to pull one over us and got caught".
After the ME security failure, the meltdown/spectre "working as intended" joke, the multi-release of various chips to try and stop ryzen in a few months, we now have the classic "oh we're sorry we got caught" ... Could someone at intel PR and marketing get a grip and stop making so much stupid moves ?
Nobody would have cared if they had come out and said "this is overclocked, a demo to show what we aim to do in the near future". Instead we got ... that.
Remember, Computex is a PR event. It's a tech show, not an investors call. You can't "lose" there, it's people coming to drool over what you have planned next. On most of these things, I don't doubt Intel and Amd are neck and neck, or even Intel ahead. Somehow, they managed to make themselves look like the ones who can't follow. Utterly terrible.
It's not a problem anymore. The majority of the public never hold any company accountable on the long run. They now know that, and they adjusted accordingly.
Microsoft, as much as I hate them, has actually made a lot of of them.
And in general, actions speaks louder then words. And people do remember what they have done. Forgiving was never going to be a easy thing, especially when you audience are developers.
Companies and countries and personae know this. Its why ruling with authority happens. Its why those in power in Arabia feared the domino effect of the Arabic Spring. It can even be a reason why people swear.
Trying to avoid it (the domino effect) from happening is called damage control.
However, the best part is that now Intel cannot argue that overclocking is unintended as they bloody set 2.7Ghz chip to run at 5Ghz, albeit for like 5seconds. (I doubt the cooling would even survive a minute)
The chiller is rated at
1.7kW thermal while consuming
More common in EE contexts than anywhere else.
I was really surprised by this. Why would you treat your customers as if they are idiots like this and make them lose trust in you? Its not like the main people making purchasing decisions are "the public" either, they should be much more informed than the typical person buying a laptop:
>"Intel now has eight major customers that are responsible for 75% of its revenues. In 2016, Intel's three largest customers were responsible for 31% of the firm's accounts receivable. Intel might obey Moore’s Law, but the Pareto Principle (a.k.a the 80/20 rule) is a different story."
We do really miss you, Andy Grove. I can bet he wouldn't have let this insanity happen.
At 250W thermal, the 4 die, 7nm Threadripper AMD announced will have similar cooling issues, a dead give away was that Lisa Su completely omitted Cinebench results - despite those being prominently featured before.
Bottom line is that AMD and Intel have roughly similar products and can easily cover for any performance differences using the large financial margins high end CPUs enjoy. The game then becomes a marketing one, who can keep more of that margin by convincing it's fan base they have decisively won the match.
They had a separate 20+ phase power delivery board that could pump up to 2000W into that cpu...
250W << 2 horsepower
My conclusion is that the chips are roughly comparable in performance and we are witnessing a marketing war, not a performance showdown. Nothing you said seems to disprove that or even address that conclusion.
Why? Anyone with a brain knew this was carefully prepared and staged, and a nice muscle-flexing publicity stunt before AMD shows their new 32-core Threadripper.
I get it, it's slow season so journalists have to find news where there aren't any. Regardless of how did they achieve it, running 5GHz on all 28 cores is pretty neat! I am wondering if some 3rd party tries to push new TR to see how far will they go, so that a few "rocket scientists" among us can build similarly overclocked rigs.
Combine these two factors and AMD might have high-core-count server chips on TSMC's (much more on track!) 7nm before Intel even has decent desktop chips on 10nm.
I don't know if they sunk all their energy in the new fabs or not ..
Intel has AVX-512, which is in my opinion one of the greatest innvoation in the x96 instruction set for a long time. The problem is that Intel wants to introduce AVX-512 only very slowly into consumer CPUs (slowly beginning with Cannon Lake, which will only be available in small quantities for some time if rumors are to be believed). Even more: For some parts of AVX-512 (e.g. 4FMAPS and 4FMAPS which are very useful for deep learning), Intel seems to be willing only for special expensive accelerator cards (Knights Mill) to include them to segment markets even further.
Similarly it is often complained that Intel offers no ECC support for the Core i... series of CPU (IMHO this complaint is rightful). Again: Intel does offer it, but only in the much more expensive server/workstation CPUs (Xeon).
So in my opinion Intel is sitting on a lot of options and brains - but they seem only willing to sell this in the really expensive CPUs for much higher prices than lots of market segments are willing to pay.
I am also curious why would 4FMPAS be useful for Deep Learning when all the major performance improvements lately come from half 16-bit/quarter 8-bit-precision floating/fixed point math (you don't need precise boundaries).
This is a claim by Intel. See for example slide 140 of
Smaller FP sizes mean you can run more of these in parallel. Smaller FP sizes mean (much) larger roundoff error, fewer bits of precision kept.
how is AVX-512 innovative when it's essentially SSE with 4x bigger registers? am i missing something?
It is much more: Just to give examples:
- The introduction of the opmask registers to mask many AVX-512 instructions
- AVX-512 Conflict Detection Instructions (CD) enables lots of loops to vectorize that could not vectorized before.
Also the fact that now 32 SIMD registered can be addressed does not seem very innovative from the outside, but implies a deep change: Before only 8 (32 bit mode; bite 5-3 of MOD-REG-R/M field) or 16 (64 bit mode; additionally use REX.R field) SIMD registers could be addresses for deep instructional encoding reasons. This also holds when using a VEX prefix. So being able to use 32 SIMD registers requires a completely new prefix scheme (EVEX). This new scheme contains lots of new capabilities (source: https://en.wikipedia.org/w/index.php?title=AVX-512&oldid=841...):
- Expanded register encoding allowing 32 512-bit registers.
- Support up to 4 operands.
- Adds 7 new opmask registers for masking most AVX-512 instructions.
- Adds a new scalar memory mode that automatically performs a broadcast.
- Adds room for explicit rounding control in each instruction.
- Adds a new compressed displacement memory addressing mode.
But it's difficult to practically use AVX-512 when the turbo boost throttling prohibits breaking even with an equivalent AVX/AVX2 workload. I had a small project that would trivially scale up to AVX-512 register sizes. Despite doubling the vector width, it was actually much slower than the AVX/AVX2 version in practice -- just because it had such a low turbo boost ratio when running AVX-512 instructions.
This was also a problem with the first AVX implementations, so this is very typical: In the first generation, Intel makes such an instruction set extension available, so that one can write applications that make use of it (though they will usually not be faster, sometimes even slower). In the following generation this new feature is made fast, so that the newly written algorithms really get to profit from it.
I'm really not convinced by this list.
Intel having significantly more vector capacity than AMD in each core is just a design decision, not an innovation.
Adding a single-precision FMA instruction is also nice but not particularly 'innovative'.
And it's not like ECC is innovative either. Intel is sitting on options, but it's not sitting on ideas.
No, AVX-512 is not only about doubling everything and calling it a day. I think the main innovation in AVX-512 is making it an easier vectorization target for compilers.
Intel's 10nm HVM isn't coming until 2019. That has been confirmed in an investor meeting when they were pushed to finally give an answer. So Intel has options? yes, lower Pricing, new SKU to muddle things up, more tricks like these Overclocking to damage AMD buyers expectation. Show unrealistic Cinebench that now totally flavours them. More Intel Inside tactics, changes of LLVM / GCC general code to force slower compile on AMD CPU, bundle their SSD with their CPU and force any vendor to get off AMD.
Oh yes, lots of options. None of them are technically superior though.
And btw, lots of the best Intel engineers are no longer with them. Apple, Qualcomm, Samsung, Nvidia, Telsa are all hiring their talents out of Intel Campus. If TSMC had Fabs in US I would bet many Intel Fabs engineer would have joined TSMC as well.
The problem with this kind of statement is that many people by their PC/laptop just for surfing and office applications. Especially for the latter application even a 7 year old PC currently suffices. But it is in my opinion easy to come up with ideas how applications that really need this kind of speed will profit from AVX-512. Additionally keep in mind that people who run applications that benefit from more speed will much more be prone to buy new PCs than people who just do office and surfing.
Infinity fabric seems quite novel for desktops.
This is basically what Intel did.
After the fiasco with compilers and chips in the past, they have to go through rigorous approval for any marketing claims. This was 100% planned by Intel.
No ill intent to find here.
Note: Watrecooling is ok for everyday use, subambiend = condensation... i.e. competitive overclocking area.
It's at least an active chiller + heat transport equipment (which can be a watercooler but doesn't look like it)
After Bulldozer fiasco, I was skeptical about Ryzen. I would have never guessed that it will mark their comeback to high performance market. What is really surprising here is how bad Intel has failed. You would expect company with its own fabs, virtual monopoly on the market and loads of capital to be prepared for such a situation. Everybody thought Intel to slow down progress because of no competition. I guess very few people thought they are actually clueless and that's how the situation looks right now.
Processor market appears to be something to watch in the coming years.
Uhm, how? For casual to complex personal computer use, how is an AMD chip is different from Intel?
They really improved with Ryzen.
(It worked up to 1.2GHz, maybe 1.3GHz when they went to Tualatin based PIII - those also kicked the behind of higher clocked Willamette P4s)
Without their anticompetitive behavior, Intel would have lost their place back then already.
And of: https://news.ycombinator.com/item?id=17262510
“We overclocked this CPU to 5GHz” means all bets are off. You often have some headroom to overclock on stock voltage/cooling but always need to go above and beyond for more extreme overclocking. This particular setup apparently was using a very fancy 1700ish W chiller, and a 1300W PSU for the chassis itself.
Also, this is a Xeon part, so one you’d expect to see in servers or workstations. You don’t want that sort of thermals anywhere close to a data centre rack, and the sort of business with the budget to give you this sort of CPU wouldn’t want to have their IT department assembling custom coolers like this.
All of this assumes, of course, you can run the chip stably. That's more or less a given at stock speeds, operating within the vendor's operational specs. Once you start overclocking, all bets are off. Extreme overclocking (the sort that uses liquid nitrogen or this sort of chiller) never lasts long. Again, something an enthusiast overclocking for a few FPS on a gaming rig might be willing to live with, but a problem that's seriously disruptive in the sort of environment a Xeon is likely to be deployed.
(EDIT: Fixed chiller power and added stability concerns)
It was 1770W, not 1000W
and produces 1.7kW of cooling
If you need to overclock you don't get this and have to buy more expensive cooling and probably need to de-lit (remove the heatspreader) of your above 2000 dollar CPU (which can permanently damage it) to improve the thermals because Intel - unlike AMD - didn't solder their heatspreader. Even then, you're not guaranteed to get a chip which can run 5 ghz on all cores 24/7. The best 18 core HEDT from Intel I have seen to date was running 4.5ghz on all cores 24/7 with delidding and a custom loop.
The 28 core Xeon which this probably is based upon runs 2.5 ghz stock and can boost to 3.8 ghz on a single core (3 or 3.2 ghz on all cores I believe).
It’s like extrapolating from an athletes 100m sprint time to their Marathon time. In other words, taking peak performance and presenting it as something that can be sustained for long periods.
But yeah, this is absolutely the case with the Intel demo. Their demo proved that the chip can run Cinebench at 5GHz.
Cinebench is not the heaviest load and it finishes very, very quickly. If they showed Prime95 AVX with small FFTs running for a couple hours, that would count as stable :D
Edit: And here I learn to read the article first: It's almost definitely that they couldn't achieve it with stock air cooling based on the setup described in the article.
For comparison, a Ryzen 1800X does 0.49 watts per square millimeter and has a lot less square millimeters (probably less than half)
The entire system pulled over 2kW of power, a lot of that dedicated to cooling.
This isn't merely "you need aftermarket cooling" but rather "you need to dedicated serious resources to cooling this".
A really cool feature the new AMD CPUs have is that they have a very well implemented turbo, if you give it good cooling it will know how to exploit it and give you more power.
As a side note, is a weird feeling running a 4 core/4 threads CPU when now workstation CPUs could reach up to 64 threads
Overclocking almost always shortens life expectancy of the CPU.
What Intel in this demo did is strapped you behind the car and drove a 100 meters dash proclaiming - look he survived, he 'ran' that 100 meters run like a champ, it's possible to run this fast.
You can change pretty easily the maximum clock multiplier the CPU is allowed to use (on K CPUs at least) -- but doing so without proper cooling and proper power supply will just crash.
SOME 'good' bin CPU can be overclocked a fair amount without extra power/cooling, but it's always touch an go.
My workstation still has a 4600K i7 that has been running at 5ghz for probably 5 years, without even pushing the voltage. It's in fact a lot quicker (single thread) than a more modern system I built since. That was a "good bin"! the 4500 and 4600 were known for their overclock potential.
On the flip side, if you buy a car that has an engine that produces only 200 horsepower, but then add a large turbocharger that boosts the output to 400 horsepower, you're likely to run into problems. The cooling system might not keep up. The engine block internals might not be able to handle the extra pressure. The transmission or differential might not be able to survive the excess torque. Your clutch might slip.
Similar idea to a CPU. Transistors can only switch so fast. If you buy a CPU that Intel labels as a 5 Ghz CPU, Intel is guaranteeing those transistors can switch fast enough to maintain that clock speed. If you buy one that's only 4 Ghz and try to overclock it, there might be some transistors in that chip that can't switch that fast. Increasing voltage can help, but now you're creating more heat.
Just that it will be sold as lower?
You.. can be a millionaire.. and never pay taxes! You can be a millionaire.. and never pay taxes!
You say.. "Steve.. how can I be a millionaire.. and never pay taxes?"
First.. get a million dollars.
Now.. you say, "Steve.. what do I say to the tax man when he comes to my door and says, 'You.. have never paid taxes'?"
Two simple words. Two simple words in the English language: "I forgot!"
How many times do we let ourselves get into terrible situations because we don't say "I forgot"?
Let's say you're on trial for armed robbery. You say to the judge, "I forgot armed robbery was illegal."
Let's suppose he says back to you, "You have committed a foul crime. you have stolen hundreds and thousands of dollars from people at random, and you say, 'I forgot'?"
Two simple words: "Excuuuuuse me!!"
- Why bother pushing for higher core count? Just leave it to the marketing department to convince customers 4 cores are enough.
- Why bother pushing for 10nm? Just leave it to the marketing department to talk about Intel's fancy 14nm++++++++++.
- Why bother pushing for higher frequency? Just leave it to the cooling solution vendors to bring in their 1400W aircon.
These are exactly what Intel actually did to its home users.
If you look at the whole drama, the most shocking parts are -
- They didn't deliver anything new when they know full well that AMD has been busy pushing for higher core count for the last 12 months.
- No one from Intel's management team is going to step down for such PR disaster.
Customers are lucky to have AMD.