Silicon Lottery store closing | Hacker News Reader

Silicon Lottery store closing

siliconlottery.com

101 pointsdtx162 comments

"Delidding" is the process of removing the lid--the shiny exterior metal integrated heat spreader (IHS)--and replacing the thermal interface material (TIM) between the CPU die and the spreader. This is done in hopes to reduce operating temperatures and increase overclocking potential. This is similar to the process of scraping off the stock TIM on a retail heatsink. This procedure can be difficult to perform; Silicon Lottery provided a premium service for interested customers.

Silicon Lottery also provided "bin" identification services [1] via frequency tests and sold desirable highest quality/lowest defect chips (hence the company name, I presume). Processor "binning", overly simplified, is a process of testing and sorting processors into various bins (e.g. best performance, lowest defects vs. decent performance, some defects, etc). That explanation doesn't do the process justice but should give you an idea.

[1] https://siliconlottery.com/pages/statistics

Yep and with Intel shitting the process bed all they can do is bin their existing stuff better, add some more aggressive boosting, and sell it as i9. They're essentially stuck "overclocking" their own chips.

To the point to where you often could get better performance UNDERCLOCKING what Intel sold you. Often to the amount of 1% less performance but also 15% less power draw.

> you often could get better performance UNDERCLOCKING

> often to the amount of 1% less performance

That doesn't sound like better performance to me. Better performance per watt maybe, but then if I'm buying an i9 chip I'm probably not chasing performance per watt.

You miss the whole issue with Intel's CPUs.

How much power they draw.

They hit thermal throttle quite often. Especially the high end chips where Intel is maximizing the clock to bin them as i9s.

1% overall decrease in performance to ensure a lower power draw overall to prevent reaching thermal throttling thresholds absolutely does equate to better performance.

I could be wrong, but I guess that would depend on how efficient your cooling is, no?

Intel has been misrepresenting the TDP requirements of their CPUs for some time.

https://www.extremetech.com/computing/319402-intels-desktop-...

Something they list as 65W can draw over 200 Watts at full load.

Consider that even with the most high end CPUs, they include the copper slug stock coolers than can make even a 45W rated CPU thermal throttle.

> For most of the 2010s, Intel kept its typical desktop CPU power consumption at or below the CPU’s rated TDP, even at peak power draw. Once AMD launched Ryzen and Intel had to start adding more CPU cores to its desktop parts, that changed. The Core i9-10850K draws up to 265W but claims a 125W TDP. The Core i7-10700 claims 65W, but draws up to 214W under load, at motherboard defaults.

In effect, Intel is brute forcing performance by cranking the power draw way up in order to compensate for what AMD has been able to bring to the table. They're bulldozering their way out of this mess, basically. The end result is that Intel is already pushing their chips to near maximum performance, power efficiency be damned, leaving zero room for overclocking... as the reason for this article exists.

> Consider that even with the most high end CPUs, they include the copper slug stock coolers than can make even a 45W rated CPU thermal throttle.

While isn't true for either the i9-10850K or the i7-10700K, but one is included for the i7-10700. However overclocking is mostly irrelevant on the 10700 since it requires changing your motherboards clock and most people won't bother.

Honestly I love what they're doing. With AIO coolers in abundance getting liquid cooling no longer requires researching and building a custom loop, and is nearly as easy as installing air cooling. If my PSU's 12V rail can deliver 250W, and my cooler can easily displace 250W, why not let my processor consume that much?

> The end result is that Intel is already pushing their chips to near maximum performance, power efficiency be damned, leaving zero room for overclocking

Isn't this what silicon lottery was doing before? Intel saw that market space and is now filling it directly. Overclocking has never been about power efficiency anyway (undervolting of course is), it's always been decreasing gains as you increase the voltage and clock rate. Also, as an owner of a i9-10850K, I can confirm that there is room for overclocking (tho not much or else it would be an i9-10900K), and you can find many instances of people overclocking their i9-10900Ks as well.

Intel has figured out what their high-performance non-business users want and are delivering on it perfectly. This is even one/two of the reasons that siliconlottery is listing as to why they're closing.

>If my PSU's 12V rail can deliver 250W, and my cooler can easily displace 250W, why not let my processor consume that much?

1) Because electricity costs money 2) In the summer, your AC also has to get rid of that heat from your conditioned space, using even MORE electricity.

> 1) Because electricity costs money

not to an extent that really matters to a home user. let's say you actually let your cpu draw 250W 24/7. average US consumer electricity rate is ~$0.13/kWh. that would add ~$23 to your monthly electricity bill. certainly a lot for a single component, but not likely to matter to the sort of person who would buy a high end cpu in the first place. if you could get the same performance using 100W, you would only save about $14 each month. and of course most people don't max out their cpu 24/7, so the actual savings would be even less.

> add ~$23 to your monthly electricity bill.

Household of three, my electric bill is $50-75mo. So… that seems like a huge jump as percentage.

Plus I don’t HAVE to be a hypocrite. If I say I care about climate, pollution, energy, I can chose to run a more efficient processor like a Ryzen or an M1.

> So… that seems like a huge jump as percentage.

Their estimate would require stress testing it every second of every day. Even if you game 4 hours every day (which would be a lot imo if you have a 9-5 job), you're looking at peak usage of 150-200W (most games won't utilize 20 threads), which using OPs costs numbers comes out to $3/month.

Fwiw I pay extra to get my electricity from wind and ride my bike ~25 miles round trip anytime I commute to work or the gym, which alone saves me over $3 and prevents ~20 lbs of CO2 per trip.

But it's not like what we do on an individual level really matters anyway. :/ The idea of a "carbon footprint" was BPs clever way of shifting climate responsibility from corporations to the consumers. I at least hope every little bit we do does help.

> Consider that even with the most high end CPUs, they include the copper slug stock coolers than can make even a 45W rated CPU thermal throttle.

In one of the thermal design documents of their processors (It's either Core2Quad family or i7 3rd generation, I don't remember), it clearly states that:

"The included processor cooling solution is neither guaranteed nor designed to keep CPU within acceptable thermal envelope when the CPU is 100% utilized".

So they basically say that they include a complementary HSF assembly to smoke test your CPU until your proper HSF arrives.

The date on this post is around that time.

It was considered something that would void your warranty to use a third party cooler.

https://community.intel.com/t5/Processors/Does-the-use-of-an...

> Yes, the usage of third party fan&heatsink void the Intel warranty on your processor. The Intel fan is made to keep the Intel processor working correctly. When we run the processor without the original fan means we are running the system out of the specifications and it may damage the processor.

Intel did everything they could with scare tactics to deny warranties. They basically implied that they would consider your RMA an overclocking case if you used a cooler with capabilities that exceeded the slug cooler included with your CPU.

Definitely. I think the growing TDP would be more concerning if AIO liquid coolers weren't so common these days. Setting up a water cooling loop used to be expensive and time consuming. Now it takes about $80 and 15 minutes, and Intel is catering to this.

Undervolting can lead to higher performance as well - for example, my 3080 Ti

Out of the box, it hits power limits and doesn't boost much past 1800/1850Mhz on the core, even in a water block.

Undervolting leads to a rock solid 1920Mhz for hours on end - less power, less heat, more performance.

I did a similar thing in the golden days with my AMD Athlon (Thoroughbred/B) processor.

It was 1400MHz out of the box, I clocked it to 2200MHz (200x11). None of the AMDs offerings ran at 2200MHz with 200x11 configuration, so it was blowing everything out of the water, with less heat and noise nonetheless.

Since x11 wasn't an extreme multiplier for it, I was running it slightly undervolted. That system is still running rock-solid even today, somewhere.

And AMD is also overclocking their own chips now.

The days of running a chip 20% faster are gone - now with much tweaking you'll get a few percent gain and way more power draw.

All this makes the Apple M1 chip look amazing, for single core loads it matches the best AMD and Intel have to offer.

> All this makes the Apple M1 chip look amazing, for single core loads it matches the best AMD and Intel have to offer.

M1 uses the next gen 5nm fab tech, it isn't fair to compare it to the others 7nm or 11nm. Apple bought up all next gen fab time to make M1, so right now you can only get it via Apple and that is the reason M1 is so superior. But this isn't thanks to Apple, AMD will get the same performance in their low power chips once they also get 5nm fab time.

But I assume that Apples strategy here is to always buy up the next gen fab time since they can sell their components at a significant markup in their phones. So that is the future of hardware, we will never get the latest gen hardware again without software lockin.

If only we could add a real GPU.

I wish the eGPU scene was a little more refined and mature. It was almost there with Intel thunderbolt but with the M1 we've got an entirely new thunderbolt implementation that barely supports (pretty sure it still doesn't without something like display link) more than one external display for what I can only imagine is a hardware limitation. It seems like we always almost get there and then something new comes along and everyone chases that instead.

M2 is sure to provide what you want. In the meantime, just keep buying.

So, for those of you that dont know what this is a simplified explanation is: silicon lottery does a few things. 1. You can buy processors which have lower defect rates and are therefore much better at overclocking than others of the same type. 2. You can get extra services like delidding which is normally done to improve thermal transfer making it easier to overclock and keep the system cool. The benefits of these have been decreasing over time and with chip shortages as well, the lottery is closing down.

While it's sad that they are closing, there is a positive take here: there is less need for such a service, because A) there is less variation among processors, so you know pretty much exactly what you are getting and don't need third-party binning to get top performance, and B) Intel has improved their Thermal Interface Material (TIM), so there's no need for delidding. Basically, CPU performance increased and grew more consistent, so you don't need a third-party provider to pick out the good ones for you.

> there is less variation among processors

Definitely, to the point where it feels like they're breaking up their product line to further narrow this variation, or at least that felt like the case with the i9-10850K vs the i9-10900K.

I think the demand is also just incredibly low. Overclocking is already a niche thing, but adding the idea of binning into it narrows it down much, much further. You have two options, one is you buy a bunch of CPUs yourself, find the best one, then sell the rest (likely at a loss), or two you pay a premium for a performance gain that is not noticable in every day tasks.

Combine that with how Ryzen performance scales right now, the FCLK, or memory clock. Most people definitely don't find memory overclocking to be worth the trouble (plus it is insanely hard to do beyond what the Ryzen calculator tells you to do).

Silicon Lottery was great. Perhaps oddly I never used them for any of my desktops rather a few laptops. I'd get the top binned CPU I could buy from them and tweak the clocks so that single core was OC'd to the max and multi core was relatively close to stock but undervolted. This kept fans from going nuts without having to sacrifice a great deal of performance.

I really wanted to get my 5950X through them but the shortages at the time made it hard enough to get one at all and they eventually gave up even trying to list them.

I feel like this service came several years too late. The overclocking community seemed much bigger 10-15 years ago.

AMD and Intel have gotten much better at pushing the stock CPU frequencies up to the actual process limits as compared to 10 years ago. Also the various Turbo boost features now automatically push it even further towards the limits when there is thermal headroom. Therefore, there is hardly very much headroom left on the table anymore for overclockers.

In 2009 I overclocked my i7 920 from 2.66GHz stock to 4.2GHz fully stable with custom water-cooling. More recently I have used i7 7700k and Ryzen 5950X under custom water-cooling and manual overclocking gets you less than 10% at most. Hardly even worth it. Just keep the CPU cool and let the built in boost features work.

There’s also the issue that newer cores run much hotter than ever before. The chiplets in my new Ryzen 7 5800X are so small that even my custom water loop can only do so much with the temperatures as there just isn’t enough surface area for thermal transfer. I was honestly worried that I was repeatedly messed up the mount of the CPU block, thermal paste application, or that one or both were just bad when I fired up HWInfo64 and did a burn-in test; my old Ryzen 5 1600X didn’t get near as hot on a 240MM AIO when overclocked as the 5800X does doing normal boosting on a 360+240mm custom loop.

>The overclocking community seemed much bigger 10-15 years ago.

Now you're making me nostalgic. 10-15 years ago was both my teenage years and the end of the overclocking golden era. I remember overclocking the living shit out of all kinds of Core2 stuff without anything except a stock cooler and a modest voltage bump that was still within Intel's "safe" recommendations.

One of the best gains was with an E5200 I spend around $120 on brand new. Bumped that up from 2.5GHz to 3.8GHz and the performance was closer to a then-$300 E8400 than the low end Pentium I'd purchased.

The last CPU I overclocked was an i7 2600k, which was my daily driver until last year (!!!).

On one hand, I miss that crazy era, but on the other hand I understand why it had to go away. Intel just don't have the sheer performance advantage to be able to gimp their own chips like that any more.

Edit: Just remembered some of the gimping AMD did, like the HD 6950 that could be upgraded to a 6970 with nothing but a BIOS flash, and the "3-core" Phenoms with a 4th core that could be unlocked in the BIOS.

My first Pentium I drilled a hole in and still have in my keychain.

I also remember buying a cheaper GPU from ATi, because the rumor was that it was a binned high-end GPU with very little defects that was only soft-locked. Worked like a charm.

I miss the era of the Celeron 300 mhz overclocked to 450 mhz. Was a time of rapid change.

Yeah, it was such an easy overclock, too, bumping the frontside bus from 66 MHz to 100 MHz. And suddenly your budget chip could hang with much, much more expensive chips.

How about 20 years ago? Remember the Celeron 300 MHz which could be easily overclocked +50%(!) to 450 MHz?

Heck yeah, and it supported SMP. I had two of them running on an ABit BP6 motherboard as my Linux workstation. Good times.

The delidding service wouldn't have made sense back then - the whole phenomenon arose only a few years ago when Intel changed the TIM material to something substantially worse than what you could get on the market.

It's still not clear to me why they did that.

But in any case, suddenly the thermal performance of Intel CPUs dropped significantly, which gave rise to the de-lidding phenomenon.

Now that Intel has switched back to a better TIM the era necessarily is over.

10-15 years ago folks did more overclocking but there wasn't really a need to de-lid, it was more about heat dissipation via elaborate cooling setups than messing with the chip package itself.

There is still some performance gains to be had by removing the IHS and placing the cold plate of your cooler straight on the die (and even crazy shit like lapping the die), but yeah, the major reason was intel's bad paste under the ihs.

I'm pretty sure the community has grown, simply by the increase of number of users alone. Extreme overclocking, which this service caters to, seems to have shrunken, though. Probably because of dimishing returns.

Back then, I used to overclock the $hit out of my CPU. Any chance I could get to ramp it up I took.

This one I have now is the first one I haven't overclocked in at least a decade: i9 10900K.

I went from an i7 8700k running all cores at 5GHz to the i9 10900K running stock.

The reason was that I ran HWInfo and saw that it clocked to almost 5GHz on its own with single and dual-threaded workloads anyway. And given the 20 cores, it would hammer through anything that needed multicore.

In short, this is the first time I didn't feel the need to overclock.

It probably mattered a lot more 10-15 years ago. A modern CPUs toward the higher end of consumer grade equipment is so powerful it's ridiculous.

RIP. It was a great service and really well run.

Can someone explain what this is/was?

CPUs sold under the same name can have substantial differences in their overclocking headroom. Back when variation was larger, purchasing a new CPU was thus participating in the "silicon lottery", at least if you wanted to OC.

The siliconlottery site used to buy a whole lot of CPUs, test them, determine their max stable overclocks, and then sell the good ones at markup (and the not-so-great ones at a little below market). If you wanted the very fastest cpu money could buy, and didn't want to start binning cpus yourself, they were the best option.

The market has largely died out because the manufacturers have improved their binning and reduced variability to the point that the best chips are no longer meaningfully better than the average and OC rarely makes much sense anyway. In many ways, this is good for customers, but it is an end of an era.

Looks as if they bought CPUs and did their own benchmarks/binning and resold them based of their measured performance.

Looks like you could mail them a CPU and pay them to upgrade the thermal compound under the IHS and optionally test it to determine maximum stable overclock parameters.

It looks like they also binned high-end CPUs. For those not familiar, that means they tested each CPU to determine its maximum stable overclock, then re-sold them with a markup on the fastest ones.

Overclocking, in my humble opinion, was always a sign of inefficiency from a manufacturing perspective. If the manufacturer was nearly-perfectly efficient at manufacturing and binning, almost none of their chips should have any overclocking headroom because they came out at their limits from the beginning.

The original margin for overclocking was in fact a result of being efficient.

Let’s say you build 1000 chips and bin them in two categories 900 came out near peak performance. Then you get orders for 200 of the top performance item, and 700 of the low tier. What are you supposed to do? Build another 7k chips and store the high performing ones?

The obvious solution, pick 600 top performing chips, relabel them as low tier, sell them at lower margin, and if needed you can start the next batch.

From computer accesories to top tier processors, the problem is not in profit margins, is in cashflows.

I feel the obvious solution wouldn't be to handicap your own product, but simply sell the same processors cheaper when they stop selling.

I've never heard of farmers who intentionally make their produce worse before selling it, they simply sell it cheaper if there isn't enough demand. If you make a lot of houses that don't end up popular, do you go and smash the roof and take away the plumbing, or sell the same house at a lower price?

Actually some farmers do the exact same thing.

Some farm products get sorted too, mostly because produce for human consumption needs to be eye catching (people don’t buy ugly food). Industrial clients don’t really care (who cares if the apples are uggly if I’m making juice anyway). This means good looking fruit sells at a premium.

If this year is particularly benevolent with your crops, but no one will buy your pretty apples, what are you supposed to do? Just let them rot? You ship it to the juice producer and call it a day.

This sounds like a company that is very bad at predicting demand. Presumably in 2021, Intel can predict relative demand between its product bins quite accurately?

The fact is that the demand for the high end is always much lower than successful chips and they need to have a sizeable gap to justify those chips.

Also yields and bins improve significantly over a product cycle so it's not set in stone.

Predicting demand is only really useful if you can tailor your yield function and doing so provides savings.

Intel doesn’t choose to make X number of top bin and Y number of the next bin, they try to optimize for as much of the top bin as they can manage.

A lot of 6 core processors have actually 8 cores, but two of them had flaws so they got disabled. The conclusion of this has been AMDs strategy, where everything from low tier CPUs to server grade use the same design, and products are built after binning the chiplets.

not necessarily. the "limits" exist in the context of reasonable cooling solutions and the power delivery capabilities of the cheapest supported motherboard. if intel could assume everyone had overbuilt VRMs and a 360mm radiator, they could push their top SKU further up the voltage/frequency curve.

part of why overclocking is dying out is because of the increasingly sophisticated boosting rules for modern CPUs. traditionally, you would just set all cores to run at the same speed 24/7 and go as far as you could under that regime. if you try than with zen 3, you won't be able to hit max 1T boost. I believe you are able to set per-core overclocks with zen 3, but not sure if anyone does that in practice.

Yep, that seems to be the case with AMD today. Intel, too, even though their chips still seem to have more room for overclocking (if you win the silicon lottery).

Does anyone know what they did with the "losing" chips, the ones they bought that didn't turn out to be positive outliers? Seems very tricky to get this financially viable, since 90% of chips will just be average. At least I assume so?

It seems the resold them cheaper [0]. Unfortunately, all prices are zeroed, so it's hard to tell by how much and whether it was below retail.

[0] https://siliconlottery.com/collections/cometlake/products/10... (see the description, 100% of CPUs reached this).

I still see prices at https://siliconlottery.com/collections/all

Intel Core i9 11900K @ 4.9GHz Boxed Processor $ 519.99

Intel Core i9 11900K @ 5.0GHz Boxed Processor Sale price $ 499.99 Regular price $ 579.99 Save $ 80

Intel Core i9 11900K @ 5.1GHz Boxed Processor Regular price $ 779.99

Intel Core i9 11900KF @ 4.9GHz Boxed Processor Regular price $ 499.99

Intel Core i9 11900KF @ 5.0GHz Boxed Processor Regular price $ 559.99

Intel Core i9 11900KF @ 5.1GHz Boxed Processor Sale price $ 639.99 Regular price $ 759.99

The discounted ones are the only ones not marked sold out.

I'm not knowledge on CPU prices, but looking at their prices from January [1] , it looks like you could save around 10% off retail if you used them to get the "losing" chips.

[1] https://web.archive.org/web/20210116175309/https://siliconlo...

Resell them as average?

Yeah, but will they even make a profit on those? I mean presumably people who just want a normal processor would prefer to get it from a more well known shop, so they can't even charge market price for them.

Plus their "normal" chips are actually worse than your average market chip, since they removed the good ones.

But maybe the margins are high, so they didn't lose money on those.

I had a feeling they we're going to close a month ago when I checked and they didn't have inventory.

Reply on news.ycombinator.com