Silicon Lottery Announces Ryzen 3000 Bins

[leadeater]

8 hours ago, S w a t s o n said:

Tick Tock was always a marketing scheme. I refuse to believe intel's foundry wasnt making constant tweaks to each node during it's lifetime rather than waiting for the next node.

They could have made manufacturing tweaks to improve yields and the like but there were no changes to anything with the actual transistor specifications of that node, Intel didn't make any changes to 22nm until 2017, well after Devil's Canyon.

 

[porina]

19 hours ago, S w a t s o n said:

zen 2 doesn't need or really have much of an avx2 offset compared to Intel who does.

It is still early days with Zen 2 but it is clear that AVX2 is still a more demanding workload than others. There isn't an AVX offset ratio as the limiting mechanism in Zen 2 is far more advanced than Intel's implementation. For normal users this is great as they get good performance under various workloads. This actually causes problems for overclockers who are using fixed clocks/voltages. We return to the old problem of AVX2 being unstable earlier while lesser loads can be pushed further. So in that respect I'd like to still have an AVX offset on Zen 2, not that I expect to get one being incredibly niche in application. Haven't figured out if it is possible to do yet, but overclockers might have to switch to power based limiting and offsets elsewhere to achieve better all round performance.

 

Note I only overclock for fun. My 24/7 running is stock.

[MMKing]

Am i missing something, or is the Silicon Lottery 4.05GHZ 3700x under MSRP? AMD lists it for 329, while you can pick it up for 299 on SL. Their most premium 3700x is only a 10USD markup. So it doesn't seem too bad.

 

There seems to be some misunderstanding what exactly Silicon Lottery is selling. AMD and it's retail partners is selling their 3900x (lets use that on for this example), with the guarantee that it will run at 3.8GHZ all core, and that it will boost to 4.6GHZ. Provided the customer dissipates the required 105W TDP, and that an AM4 motherboard is used. Note that ANY supported motherboard, guarantees 3.8GHZ all core as long as sufficient heat is dissipated away from the CPU. How much it will boost, will wary depending on available power, motherboard and may require additional heat dissipation. Remember the boost frequency figure is a maximum.

 

What Silicon Lottery sells you, when they sell you the 3900x 4.2GHZ. Is a guarantee that the product will reach 4.2GHZ all core. Provided.

 

1. You input the settings in your BIOS as is written on the product page.

 

2. You use one of these motherboards (newegg price)

 

ASUS X570 Crosshair VIII Formula (700 USD)
ASUS X570 Crosshair VIII Hero (360 USD)
ASUS X570 Prime X570-Pro (250 USD)
ASUS X570 Pro WS X570-ACE (380 USD)
ASUS X570 STRIX X570-E (330 USD)
ASUS X570 STRIX X570-F (300 USD)
Gigabyte X570 Aorus Xtreme (700 USD)
Gigabyte X570 Aorus Master (360 USD)
Gigabyte X570 Aorus Pro (270 USD)
Gigabyte X570 Aorus Ultra (300 USD)

 

Note that the minimum price is 250USD, the lack of MSI or Asrock options and the fact that there are only X570 motherboards listed.

 

Failing to use one of these motherboards, will not guarantee the 4.2ghz all core overclock. Even if you were to argue that your motherboard is on par, or superior to any of these motherboard. If your motherboard is not on this list, there is no guarantee.

 

3. The overlock has only been confirmed to work with the following memory settings and setups.

 

Dual Channel DDR4-3200 (Up to 4 dimms single or dual rank)
Dual Channel DDR4-3333 (Up to 4 dimms single rank or 2 dimms dual rank)
Dual Channel DDR4-3466 (Up to 2 dimms single rank)
Dual Channel DDR4-3600 (Up to 2 dimms single rank)

 

If you want faster memory. You are not within the guarantee. If you have 2 dimms clocked at 4000mhz and the CPU is not stable at 4.2GHZ or if you have 4 dimms clocked at 3466mhz. You are outside the guarantee.

 

4. You have to use one of the following thermal compounds

 

ARCTIC MX4
Gelid Solutions GC-Extreme
Noctua NT-H1
Thermal Grizzly Hydronaut
Thermal Grizzly Kryonaut

 

5. Last but not least. You are required to use a liquid cooler with at least a 240mm radiator. AIO closed loop or custom built does not matter.

 

 

If your business is a 3950x, or even a 3900x. There may be some point to buying a guarantee from Silicon Lottery, that the CPU you buy. Will reach a certain frequency. If you are already in that price range, there are few other options to go further and chances are you already planned to buy a high end motherboard, a case with excellent air flow and a liquid cooler. If the items on this requirements list were already in your shopping cart, it's not a bad idea.

 

However. If the items on this expensive requirements list was not in your shopping cart. You are paying an extortionate amount of money for very little performance gain. I would rather run a 3900x on a mid range X470 board at stock settings. Rather than buying an expensive X570 motherboard and cooling to run a 3700x at a slight overclock.

 

Lastly. The 610USD 3800x is the stupidest shit i have ever seen. It's almost twice the MSRP of a 3700x, which competes very well with a stock 3800x. At that point, just buy a 3900x or save up for the 3950x.

[S w a t s o n]

4 hours ago, leadeater said:

They could have made manufacturing tweaks to improve yields and the like but there were no changes to anything with the actual transistor specifications of that node, Intel didn't make any changes to 22nm until 2017, well after Devil's Canyon.

 

Intel also has to give that information out. Back when they followed tick-tock they were only advertising the new node, not the node itself over the lifetime of said node. Because there was no need, there was always a new node to look forward too. The death of tick tock is directly due to the delays of 10nm.

 

This is besides the fact that purely yield driven improvements can lead to minor performance increases (max clock) as well.

[Neftex]

On 7/22/2019 at 3:25 PM, cj09beira said:

its really interesting that they have their max vcore so low, 1.25 is their top bin so clocks could increase quite a bit just by allowing more voltage, i wonder where they got that voltage, i think someone from amd talked about 1.3 + something, maybe they went from better safe than sorry

sent them a message asking if there is a particular reason as to why its 1.25v

i found out on my 3600 1.25V was the sweetspot for performance(not losing too much on single thread score + high multithread score). going higher raised temps resulting in worse performance. then again this was on stock cooler so it shouldnt apply to SL

[cj09beira]

weirdly enough i cant find my first post with data, well i found stilts matise Strictly Technical though this time he only used the 3700x so its not the best silicon but here it is

(S means data from stilt)

 

sadly i didn't see the same data for intel cpus from stilt will try to look for it though as i have no clue how much variance things like coolers used could have 

[dalekphalm]

On 7/24/2019 at 10:03 AM, S w a t s o n said:

Intel also has to give that information out. Back when they followed tick-tock they were only advertising the new node, not the node itself over the lifetime of said node. Because there was no need, there was always a new node to look forward too. The death of tick tock is directly due to the delays of 10nm.

 

This is besides the fact that purely yield driven improvements can lead to minor performance increases (max clock) as well.

The fact remains that you haven't provided any evidence to suggest that Intel is improving the performance of the node throughout said node's lifetime. We expect yield improvements, yes.

 

Also, please provide a source that yield improvements can affect max clock or any other performance metric.

[W-L]

Keep discussion strictly to the topic at hand. 

 

-Thread Cleaned- 

[dalekphalm]

1 hour ago, S w a t s o n said:

Source that is strictly on topic about the SPEED of the chips:

  Reveal hidden contents

 

 

As I suspected. An unsubstantiated claim. 

 

Anyway back on topic, I don’t know why people are confused about SL’s business model. It’s pretty straight forward. You takes your chances or you pay for guarantees. 

[S w a t s o n]

7 hours ago, dalekphalm said:

As I suspected. An unsubstantiated claim. 

 

Anyway back on topic, I don’t know why people are confused about SL’s business model. It’s pretty straight forward. You takes your chances or you pay for guarantees. 

I thought it was very substantiated? Are you telling me that you only felt that way once I made that post, so i guess before you thought it was substantiated even though you want a source for how basic foundry improvements work? lmfao

[79wjd]

1 hour ago, S w a t s o n said:

I thought it was very substantiated? Are you telling me that you only felt that way once I made that post, so i guess before you thought it was substantiated even though you want a source for how basic foundry improvements work? lmfao

Vishera comes to mind as a counter example. The FX8370 clocked no better than an FX8350 from two years earlier.

[dalekphalm]

1 hour ago, S w a t s o n said:

I thought it was very substantiated? Are you telling me that you only felt that way once I made that post, so i guess before you thought it was substantiated even though you want a source for how basic foundry improvements work? lmfao

You've provided no information to substantiate how you think basic foundry improvements work.

 

Perhaps you're right. But you made a claim - one you seem to think is common knowledge, and it's not. So either the claim remains unsubstantiated or you provide evidence to back it up.

 

I'm not even saying what you've claimed is wrong. I don't know if it's wrong or not. But I've definitely seen no information to indicate it is correct either. So, provide a source, and I'll happy agree with you.

[S w a t s o n]

5 hours ago, dalekphalm said:

You've provided no information to substantiate how you think basic foundry improvements work.

 

Perhaps you're right. But you made a claim - one you seem to think is common knowledge, and it's not. So either the claim remains unsubstantiated or you provide evidence to back it up.

 

I'm not even saying what you've claimed is wrong. I don't know if it's wrong or not. But I've definitely seen no information to indicate it is correct either. So, provide a source, and I'll happy agree with you.

Like I said, I don't really care if you agree with me or not

[S w a t s o n]

5 hours ago, 79wjd said:

Vishera comes to mind as a counter example. The FX8370 clocked no better than an FX8350 from two years earlier.

Except that the 8370 was literally a new bin created with a higher clock and lower minimum voltage. Almost exactly like they could now on average bin the chips better due to improvments. Check the AMD FX processor page on Wikipedia

[dalekphalm]

4 minutes ago, S w a t s o n said:

Like I said, I don't really care if you agree with me or not

That's cool - again, unsubstantiated claims. Honestly, I don't care whether you provide the proof or not. If you don't, your claims will continue to be dismissed, and that's okay with me too.

[S w a t s o n]

Just now, dalekphalm said:

That's cool - again, unsubstantiated claims. Honestly, I don't care whether you provide the proof or not. If you don't, your claims will continue to be dismissed, and that's okay with me too.

Again, you literally commented on a conversation between me and someone else who agreed with me on various points. You can dismiss, literally no one cares

[Drak3]

5 minutes ago, S w a t s o n said:

literally no one cares

You do.

[S w a t s o n]

Just now, Drak3 said:

You do.

You sure got me

[dalekphalm]

8 minutes ago, S w a t s o n said:

Again, you literally commented on a conversation between me and someone else who agreed with me on various points. You can dismiss, literally no one cares

You're posting in a public forum on a public thread. You're not having a private conversation via Direct Message. Anything you post here is privy to me (or anyone else) responding. You also say you don't care, yet you repeatedly re-engage in the conversation.

 

Maybe you don't care - but you've got an odd way of showing it.

 

Also, yes, the other person may have agreed with you on various points. That's irrelevant. You're claiming something that isn't common knowledge, and acting like it is. That's about the end of it. Perhaps your claims are true, perhaps they aren't. But them being true isn't the default position. So. Here we are.

 

I digress however, it's obvious this is going nowhere.

[S w a t s o n]

Just now, dalekphalm said:

You're posting in a public forum on a public thread. You're not having a private conversation via Direct Message. Anything you post here is privy to me (or anyone else) responding. You also say you don't care, yet you repeatedly re-engage in the conversation.

 

Maybe you don't care - but you've got an odd way of showing it.

 

Also, yes, the other person may have agreed with you on various points. That's irrelevant. You're claiming something that isn't common knowledge, and acting like it is. That's about the end of it. Perhaps your claims are true, perhaps they aren't. But them being true isn't the default position. So. Here we are.

 

I digress however, it's obvious this is going nowhere.

The fact is if you understood what you were talking about you'd know that there's no fucking advertisement from Intel to use a source. Go learn about how the process actually happens and learn about semiconductor technology in general. There's no magic source from Intel for intra node tweaks because that's proprietary industry secrets. You get what they decide to give you and that was the point of the conversation. Go find me a source for Ford making minor changes to a model year of a car, and I'm talking about all the fucking minutia, not the "we added a new sound system". You don't advertise literally everything you do.

[79wjd]

5 minutes ago, S w a t s o n said:

Except that the 8370 was literally a new bin created with a higher clock and lower minimum voltage. Almost exactly like they could now on average bin the chips better due to improvments. Check the AMD FX processor page on Wikipedia

Stock clocks were the same, the boost clock was 100Mhz higher meanwhile the overclock potential and power draw was virtually identical.

 

https://pcper.com/2014/09/amd-2014-fx-refresh-fx-9590-fx-8370-and-fx-8370e-review/9/

[S w a t s o n]

2 minutes ago, 79wjd said:

Stock clocks were the same, the boost clock was 100Mhz higher meanwhile the overclock potential and power draw was virtually identical.

 

https://pcper.com/2014/09/amd-2014-fx-refresh-fx-9590-fx-8370-and-fx-8370e-review/9/

Yea it's a bin not a refresh, I didn't say they OC to the moon because of process improvements, it's going to vary node to node, chip.to chip, but I would definitely say that the 8370 is going to have an easier time the reaching 9590 level overclocks. Which the 8350 could already do as well.

[dalekphalm]

Just now, S w a t s o n said:

The fact is if you understood what you were talking about you'd know that there's no fucking advertisement from Intel to use a source.

So pretend I don't know anything about Processor manufacturing, then. You'd still need to provide a source to back up your claims.

Just now, S w a t s o n said:

Go learn about how the process actually happens and learn about semiconductor technology in general.

I don't need a source from Intel. Any source (confirmed and reliable of course) will do. Such as: a source that reliably and repeatedly tests CPU's of the same model manufactured at the beginning of the fab process consistently and constantly performing worse than the exact same model produced at the end of the fab process.

Just now, S w a t s o n said:

There's no magic source from Intel for intra node tweaks because that's proprietary industry secrets. You get what they decide to give you and that was the point of the conversation.

You don't need information from the manufacturer, if it's something that can be reliably measured. You have yet to prove this.

Just now, S w a t s o n said:

Go find me a source for Ford making minor changes to a model year of a car, and I'm talking about all the fucking minutia, not the "we added a new sound system". You don't advertise literally everything you do.

Actually, you can find sources on minor changes between model years - maybe not every single minor change, but reviewers and mechanics and enthusiasts do indeed look at these little details. Just not from Ford itself.

 

2 minutes ago, 79wjd said:

Stock clocks were the same, the boost clock was 100Mhz higher meanwhile the overclock potential and power draw was virtually identical.

 

https://pcper.com/2014/09/amd-2014-fx-refresh-fx-9590-fx-8370-and-fx-8370e-review/9/

Looks like at best, a better binned version of the exact same silicon. At worst, it's just literally the same part with a higher factory clock. That doesn't prove that the node improved over time allowing better performance.

[dalekphalm]

Just now, S w a t s o n said:

I would definitely say that the 8370 is going to have an easier the reaching 9590 level overclocks.

What is this comment based on? Where's your source to verify this claim?

 

This is entirely my point. You say something, but there's no evidence or proof that what you say is correct.

 

Perhaps the 8370 can OC better. But until I see evidence of that (For example, consist reviews that show higher OC's across the board compared to the older chip, or statistically significantly higher overall OC's on an OC tracking site, etc), it remains unsubstantiated.

[S w a t s o n]

1 minute ago, dalekphalm said:

What is this comment based on? Where's your source to verify this claim?

Literally the fucking fact that it's a better binned chip? As for your long ass post. No, go read a fucking book, or in this case a white paper.

 

I asked for all the minor details, the literal way it's manufactured. Which you just admit is not freely available.