What this says to me is don't bother trying to super overclock a i9-9940X, Intel has already skimmed off the cream from that chip run.
It seems kind of scummy, but I guess otherwise your rare stable overclock chip might end up in some boring business server where it will always run at stock clocks. I do note that Intel still wants nothing to do with overclocked chips in their warranty department, even when they did the overclocking themselves.
It feels like when you learn that Ticketmaster also owns most of the scalping services and your chance of getting a good ticket is zero unless you're willing to go to the scalpers.
Basically the good stuff is reserved for the rich because they are rich.
>It feels like when you learn that Ticketmaster also owns most of the scalping services and your chance of getting a good ticket is zero unless you're willing to go to the scalpers.
If anything, this situation is better than Ticketmaster's, because at least the producers are getting the money. With Ticketmaster, they're pocketing all the money and the performers are getting nothing. If Intel isn't binning these, a third party (probably someone affiliated with an OEM) would be doing this and pocketing the difference.
>Basically the good stuff is reserved for the rich because they are rich.
Are you also upset that you can't get luxury goods for cheap because they're "reserved" for the rich?
I have come around to the belief that economically efficient behavior can be more ethical. When I have something I want to get rid of, I have tried giving it away, but I learned by experience that people who are first to grab something aren't necessarily those who will make the best use of it or who need it the most. So it may be better for society to sell or auction it even if that has significant overhead. Raising more money isn't even the point; it's a side effect.
Most would call it completely fair, and a proper incentive to bring similar products to market.
It seems to me that greed is a problem affecting those who want it all, particularly luxury items, while complaining that they should be exempt from paying any cost.
It's (a little bit) scummy, because they're selling them on auction, which has a decent chance of flogging them for orders of magnitude more than they'd ordinarily sell them for, with the added bonus that they don't have to honor any warranty stuff since they're going to be clocked to hell.
> It's (a little bit) scummy, because they're selling them on auction, which has a decent chance of flogging them for orders of magnitude more than they'd ordinarily sell them for,
It seems to me that demand is far higher than supply, thus the auction ensures that those who really want them will have a realistic shot of getting one. That's far better than having to resort to buying them from price-gouging scalpers.
Buying them at inflated prices from Intel rewards Intel for being able to make these at all. If they were able to make more of them, they'd be rewarded more.
Paying the inflated prices to scalpers would reward scalpers for hogging the supply. That would be far worse. If I'm going to pay an inflated price for something, I'd rather use it to reward the person making it than some speculator trying to wrench some profit out of a tight market.
> Buying it at inflated prices from Intel effectively rewards them for their inability to produce more.
For this assertion to make any sense you'd need to believe that they don't benefit by increasing sales of a product line, particularly when they increase prices.
Hardly, since the "rejects" will just be sold as the regular model(s).
In the old "lottery" system the rejects would end up on the secondary market, making Intel's final revenue effectively the same as if extreme overclocking wasn't a thing.
Nobody is forced to take part in the auction. There are other CPUs that can be chosen from. If an auction participant calculates it pays for them to buy these even an inflated prices they are not really ripped off, it's just free market.
Also, binning of CPUs is a fact of life. CPUs don't all come the same from the production line. It allows you to buy bit less capable CPU for much cheaper if you wish so which, I would assume, you didn't take into consideration when you posted your thoughts.
I don't see the auction itself as scummy. They're absolutely free to do that. It might be a bit scummy that the auction isn't open to everybody. Only a select group of OEM vendors will have access to this. That may also mean it's harder to acquire for benchmark tests.
Furthermore, no warranty could even mean they can get away with selling processors that won't work reliably at the listed speeds. If they were to do that, that'd definitely be scummy.
That wouldn't make that much sense at that number given that such a large percentage of recent chips have been good enough to make it to 5.0 ghz+ according to these statistics recorded by Silicon Lottery: https://siliconlottery.com/pages/statistics -- no stats for i9-9990XE but there's enough to get an idea.
It wouldn't be that shocking if delays keep hitting on the new product line that they just skim the top 30% of chips and call it a "new" product though, which would be both really funny and almost immediately noticed. But overall this kind of thing is inevitable given the giant difference between what these kinds of processors run at stock and what people who are eager to buy a top of the line processor are actually going to want to run it at.
Or, more charitably, they are integrating that. In some industries, these high binned chips are a crucial competitive advantage, and I don't see any reason why they shouldn't validate the performance of each chip they sell (at least in HEDT, other product lines I can see having little or negative return from an approach like that).
Maybe they could even just make it explicit for all their products: market them by die type, configuration, and peak stable base clock.
Also notable for a huge increase in TDP from the next-slower model (255 W vs 165 W) - and it has four fewer cores.
> Other details about the chip that we have learned include that it will have a listed TDP of 255W, which means the peak power will be higher. Motherboard vendors will have to support 420 amps on the power delivery for the chip (which at 1.3 volts would be 546 watts), and up to 30 amps per core.
That is insane, like the worst of the Pentium4 NetBurst era when AMD was competitive at the high end and Intel decided to try to clock their way out of the problem. Granted this is a one-off chip to grab headlines and not a long term micro-architecture commitment.
But between this and the fiasco when Intel announced a 28-core 5 GHz chip[0] (without mentioning the 1800 W industrial water chiller needed to cool it), it's starting to sound desperate in its attempts to deflect attention from AMD's EPYC and Threadripper high-core-count chips.
550+ watts is insane. That is spaceheater territory. Residential outlets start tripping at 1500-2000 watts. How soon we will have to run even basic gaming machines off multiple outlets.
I remember reading/hearing about a ~2000w PSU about 10 years ago that needed to be run off 2 circuits. Fortunately, SLI never became that popular, and even 1000w PSUs (while readily available) are overkill for most gaming setups, even with 2 GPUs.
It's not so bad if you think of it more like a folding desktop that can occasionally be transported from place to place. The total weight is a lot less than a mini-itx chassis and a standalone 20" class monitor... But not by much.
I've never tripped an outlet on less than 2500 watts, and that was an old electrical network... Did I get lucky or do standards vary in different countries?
Most circuits trip on current, because I^2R heating is what causes fires in house wiring, 15A is relatively common, to get power you multiply by voltage which varies by country, in US that gives you 1800W and UK it’s 3600W. There are a ton of circuits which aren’t 15A though.
If you live in UK, you might note that electric tea kettles work fine, but in the US they do not (they’re too slow). That’s the biggest practical difference to my mind, very few other portable appliances need that kind of juice, except the biggest space heaters. Big, fixed appliances get wired differently.
If you’re handy with electricity and don’t mind fracturing a few rules, European kettles work just fine on 240V 60Hz power. A British 13A plug in the kitchen is a nice conversation piece too.
Note: because of the US “split phase” system, the neutral wire won’t be neutral anymore but rather 120v above ground. The kettle bases I’ve used are well designed, but do be careful.
Specifically to being careful: put a GFCI upstream of the kettle outlet. Which should be there anyway because it's a kitchen, but even more important with a phase-to-phase setup.
It never occurred to me that the voltage difference between America and Europe/Asia would cause a corresponding differences in watt output given the same amps... But yes indeed, US kettles boil much slower.
Other appliances that require a lot of juice include vacuum cleaners, electric hobs, space heaters, portable ACs/dehumidifiers. So having something trip a living room circuit pulling just 2000 watts would drive me nuts in the long run
I live in Japan where we have a paltry 100V. In our apartment it's made by for by having lots of (20A) circuits - 3 for the (small) kitchen alone [which uses gas for the stove anyway]. Total of 14 breakers for a small 3 bedroom + living + kitchen apartment.
In 3 years we've only tripped a breaker once - running a microwave, toaster oven and dishwasher all off the same outlet at once.
You could imagine it driving you nuts, in my experience trips are rare (once every few years?) unless you love space heaters. Electric ranges (hobs?) run on their own dedicated 240V circuit. As a tradeoff, the plugs are small and convenient; I can fit a phone charger in my pocket, plus the phone and cable—true in Europe, but not the UK. I can also get a power strip with 8 outlets and it won’t take up much space.
A couple days ago I tripped a circuit breaker for the first time in many years. It turns out that the refrigerator shares a circuit with the rest of the kitchen, so when I ran a rice cooker and deep fryer off the same circuit, when the compressor turned on it tripped.
Older houses have interesting wiring. I've even seen a 5A circuit (once) in a really old place, varnished cotton clad copper wire and other interesting stuff.
Also (in NL), a fire that had started in a junction box that somehow died out without becoming more serious. That was a pretty scary find in a house that I had just bought.
> Older houses have interesting wiring. I've even seen a 5A circuit (once) in a really old place, varnished cotton clad copper wire and other interesting stuff.
I've lived in a few older homes, and have experienced some of this firsthand. My current home has some of that wire with porcelain wire nuts in the older circuits. Previously we lived in an earlier place that still had a small amount of legacy knob-and-tube wiring. (Bare copper wire kept off joists with porcelain insulators of various sorts.)
Never had to deal with fuses, or sub-10A circuits. (But the utility of a 5A household circuit seems evident in an older world where the whole-house service might be 30 or 60A.)
Quote from the end: "perhaps importantly, there is no warranty from Intel. This means that system builders will not be able to recoup costs on dead silicon, but they might give their own warranty to end users."
This would be a pretty significant liability for builders in countries that have automatic consumer protections (eg: Consumer Guarantees Act in NZ). The builders would have to eat the costs of faulty units.
True, if that’s the cost of our consumer protections I’m okay with that. It’s actually really nice to live under. That said I’ve never been on the retailers end.
By nature HFT is not highly multithreaded and avoids handoff.
So having more cores is not that different than having more discrete CPUs -- no benefits for cache locality/faster coherency.
In that regard if latency is an issue overclocked 8700K (or 8086K) should be sufficient.
Binned & overclocked i9-9980XE’s are already available though. I’m honestly at a bit of a loss to understand Intel’s latest offering. But hey, it made us stop laughing about 10nm for a few minutes.
It is, but at some point you can't do much more about the network latency; shaving off a couple dozen us, might the difference of getting the price/quote or missing it out. Especially, when you play versus similarly equipped firms.
Depends on how low the latency already is. If you're already <0.5ms away network wise then perhaps shearing off 0.1ms from a 1ms operation could make a significant difference.
I'm guessing the chance of an eventual calculation error is not worth the potential speed gains (though you could run those Pi calculators to check for CPU stability)
HFT is not so much about CPU speed but algorithmic optimization.
I've got this 20GHz bazillion core CPU for a really good price. No warranty, no guarantees, stated or otherwise implied (May or may not work). Call me for price.
I kind of like this trend of power-hungry beastly CPUs. The AMD 2990WX is also rated 250W (but for 32c/64t@3GHz). I have one with a 280mm AIO, and it sees high 60C under load at stock frequencies. I can only overclock to 3.4GHz while sustaining heavy compute 24/7 and that's probably drawing close to 400W and heats up to mid 80C (in a room with 20C ambient). It's a pleasant space heater for the winter.
Hoping to see AMD continue to push the HEDT threshold. Let the server market have their underclocked power-efficient processors. Though to be honest if things go much further I'm going to need a custom waterloop next generation.
Who needs these chips? I'm honestly curious. I've never worked on something that needed as much single-thread performance as possible and couldn't be parallelized; I'd love to know what industry will use them (and what industry would buy a small number of chips at auction prices).
Considering no games are built to support 24 threads, most don't even scale past 8 yet, any gamer would be much better off with a 9900k. I don't know who they're targeting with this, but with the 9980xe already being so expensive if someone truly wanted maximum single core performance for games and lots of cores for other things why not buy a 9900k and a 2990wx. More performance for likely a whole lot less.
My workstation has 20 cores/40 threads and a large C++ compile job parallelizes nicely, but the linking phase does not. I could use more single threaded performance to speed up linking of our many libraries and executables.
Beowulf was about the hardware and the interconnect - both being commodity. Not about the software. Most Hadoop etc runs on Beowulf I’d imagine.
Clusters used to use more custom hardware and interconnects - that’s now the exception rather than the norm. You still find that in some HPC situations but most people manage with commodity cloud.
It seems kind of scummy, but I guess otherwise your rare stable overclock chip might end up in some boring business server where it will always run at stock clocks. I do note that Intel still wants nothing to do with overclocked chips in their warranty department, even when they did the overclocking themselves.