Intel i9 CPU's get the finalized specs announced

[Jito463]

2 hours ago, VagabondWraith said:

So does this mean we're in the Corean war?  

[ravenshrike]

10 minutes ago, Masada02 said:

That TIM is only there to force enthusiasts to delid thus voiding warranty so Intel doesn't have to pay for bad overclocking errors. That's really the only thing I disagree with. Once delidded it will be fine.

No, the toothpaste is there to save money during the manufacturing process and in theory some minuscule amount of RMAs. Intel really doesn't care about overclockers except as another notch in their post.

[tom_w141]

38 minutes ago, Terodius said:

So let me get this straight. Going from 6 to 8 cores is 210USD extra. Going from 10 to 12 is 200usd. going from 12 to 14 is 200usd. So you'd think that they're basically charging 100usd per extra core while AMD is charging about 50usd per extra core. Why then does it cost 400 fucking USD to upgrade from an 8-core 7820X to a 10-core 7900X? or 300 to upgrade from the 14 to the 16 core and the 16 to the 18 core? The Intel pricing is not only ridiculously uncompetitive, it also makes no goddamn sense. 

When you say it like that it sounds mad but you don't understand yields. Yield is basically chance.

 

In statistics/probability math the rules are simple. Any time you see AND it means multiply.

 

For a quad core you need core 1 AND core 2 AND core 3 AND core 4 to be functional which is WXYZ (algebraically) where W-Z are decimals as 1 = 100% chance.

 

When you multiply decimals the number always gets smaller so if WXYZ = 80% then a 6 core UVWXYZ must = <80%

 

Each time you do this the decimal or % gets smaller and smaller. By the time you have got to an 18 core its IJKLMNOPQRSTWXYZ, now I don't work at Intel so the numbers are made up but that has to be a very very low number, lets say around 30%. This means that as you go up the stack the diminishing yield gets so small that the successful ones are so uncommon that they need to be sold at a cost to justify all the "failed" dies that were either useless or had to become lower SKUs.

 

"But AMD aren't pricing anywhere near that high" I hear you cry.

 

Yes because AMD Ryzen and by extension Threadripper and Epyc are all using the same zen dies in a multi die config rather than Intel's monolithic approach.  This number may be wrong but its been floating around the internet that a functioning 8 core zen die has a yield of ~80%. Now ignoring performance and architecture differences the difference this makes is that AMD doesn't need 16 successful cores on a single die. Now we only need the event (functioning cores) to happen 8+8 concurrent times vs 16. Achieving 8 concurrent events on 2 separate attempts is far more likely than 16 concurrent events in 1 attempt.  Therefore the conclusion of a successful 16 core TR (2x8 zen dies) has a much higher % than the Intel which needs 16 functioning cores on 1 die. Thus yields are higher, they have more 16 core SKUs making the per part price lower because they have little waste. Zens design is actually so clever they have little to no waste because "failed" zen dies with only 4-7 functioning core just become R3s and R5s.

[Masada02]

1 minute ago, ravenshrike said:

No, the toothpaste is there to save money during the manufacturing process and in theory some minuscule amount of RMAs. Intel really doesn't care about overclockers except as another notch in their post.

Doesn't care about overclockers yet releases 18 core fully unlocked chip. Okay.

 

Regardless, you can't prove that theory. Intel is never going to say that it was a cost saving measure on their $2000 CPU, nor would they ever admit to my theory. Even more, switching to TIM instead of solder saves what, like $5 tops on the process. Both of those methods are incredibly easy to accomplish and all the real cost in the chip is on the silicon itself. If they really wanted to cut costs they wouldn't have used a secondary piece of substrate on the small core count X299 chips, but they still did. In reality, the choice to use the TIM was a pure marketing move, as mentioned by someone else earlier in this thread.

[grimreeper132]

2 hours ago, PCGuy_5960 said:

The clock speeds are too low, but I guess that this is the only way they can get the 165W TDP

the Ryzen 1950X might hand intel a new one when it comes to multicore performance from the looks of, it's hard to say for sure, but it looks like it quite possibly will. (AMD 32*3.4=108.8 vs intel 36*2.6=93.6, so assuming clock to clock are equivalent the AMD 16 core will be about 10% faster (very rough estimate, it's hard to say) in multi core performance)

[Cookybiscuit]

These clocks are a disaster. 1950X takes a massive dump on anything Intel can put out right now at any price point, even 1800X makes a mockery of most of these. Intel truly is bankrupt and finished, there's no coming back from this. Intel fanboys are BTFO and on suicide watch. Lisa Su's ThreadShitter just wiped out Intel overnight. 

 

Pack it up boys the show's over.

[ravenshrike]

13 minutes ago, Masada02 said:

Doesn't care about overclockers yet releases 18 core fully unlocked chip. Okay.

The 18 core was only thought up after TR leaks prior to Computex and the original literature distributed AT Computex didn't mention anything over 12 cores. So yeah, the only reason they're coming out with it is e-peen.

[Juanitology]

8 minutes ago, tom_w141 said:

When you say it like that it sounds mad but you don't understand yields. Yield is basically chance.

 

In statistics/probability math the rules are simple. Any time you see AND it means multiply.

 

For a quad core you need core 1 AND core 2 AND core 3 AND core 4 to be functional which is WXYZ (algebraically) where W-Z are decimals as 1 = 100% chance.

 

When you multiply decimals the number always gets smaller so if WXYZ = 80% then a 6 core UVWXYZ must = <80%

 

Each time you do this the decimal or % gets smaller and smaller. By the time you have got to an 18 core its IJKLMNOPQRSTWXYZ, now I don't work at Intel so the numbers are made up but that has to be a very very low number, lets say around 30%. This means that as you go up the stack the diminishing yield gets so small that the successful ones are so uncommon that they need to be sold at a cost to justify all the "failed" dies that were either useless or had to become lower SKUs.

 

"But AMD aren't pricing anywhere near that high" I hear you cry.

 

Yes because AMD Ryzen and by extension Threadripper and Epyc are all using the same zen dies in a multi die config rather than Intel's monolithic approach.  This number may be wrong but its been floating around the internet that a functioning 8 core zen die has a yield of ~80%. Now ignoring performance and architecture differences the difference this makes is that AMD doesn't need 16 successful cores on a single die. Now we only need the event (functioning cores) to happen 8+8 concurrent times vs 16. Achieving 8 concurrent events on 2 separate attempts is far more likely than 16 concurrent events in 1 attempt.  Therefore the conclusion of a successful 16 core TR (2x8 zen dies) has a much higher % than the Intel which needs 16 functioning cores on 1 die. Thus yields are higher, they have more 16 core SKUs making the per part price lower because they have little waste. Zens design is actually so clever they have little to no waste because "failed" zen dies with only 4-7 functioning core just become R3s and R5s.

Actually I am very much aware of how yield works and how die-size affects that. I am also aware of the fact that a silicon wafer costs around 2000USD.

 

So yes, they do get lots of dies that do not have 18 functional cores, but those get binned into lower SKUs, they don't just get thrown into the trash. Even if the 18-core CPU had a 30% yield(which it doesn't since their 24-core xeons on the XCC die have a 35% yield), they'd still be getting around 30 working 2000USD CPUs per wafer(around 75USD cost/CPU including packaging), on top of all the other lower-binned SKUs.

 

They're getting at the very least 100.000USD of revenue per wafer with 2.000USD of raw materials cost. They made close to 13 billion in operating income last year, I think they're doing just fine. They're making tens of thousands of dollars of profit on each wafer even after taking in every possible expense you could think of, from R&D to marketing. 

 

So yeah, you can try to justify them charging 2.000USD for a product that costs them 75USD to manufacture all you want, but the fact is they're abusing their market position and have been doing so for years.

 

They've been found guilty of anti-trust and anti-competitive violations in several occasions, so this is just business as usual for them, more of the same horse-shit they've been feeding us for years now. Except this time they got caught off-guard by AMD and had to rush this entire 12 to 18 core lineup so it'll end up having even more problems than a normal CPU generation. 

[Daniel644]

Not to give Intel a pass, because they sure don't deserve it, but gotta love the first page of people talking about Base Speeds, like BASE SPEEDS mean a damn thing in this world today or ever, when was the last time you saw your computer operate at it's base speed?

[Taf the Ghost]

1 hour ago, ravenshrike said:

The 18 core was only thought up after TR leaks prior to Computex and the original literature distributed AT Computex didn't mention anything over 12 cores. So yeah, the only reason they're coming out with it is e-peen.

It's less than just showing off. It's the reality of a 16c vs 12c part being the HEDT lineup for at least a year. Intel would be outright way behind in the space. While it's a small market, it's a very large mental market. This is why AMD sent out so many kits. They want everyone making stuff on AMD while on YouTube. That's why Intel binned and sacked CPUs that they'd sell for around $5000 USD.

[Drak3]

1 hour ago, grimreeper132 said:

the Ryzen 1950X might hand intel a new one when it comes to multicore performance from the looks of, it's hard to say for sure, but it looks like it quite possibly will. (AMD 32*3.4=108.8 vs intel 36*2.6=93.6, so assuming clock to clock are equivalent the AMD 16 core will be about 10% faster (very rough estimate, it's hard to say) in multi core performance)

1 hour ago, Cookybiscuit said:

These clocks are a disaster. 1950X takes a massive dump on anything Intel can put out right now at any price point, even 1800X makes a mockery of most of these. Intel truly is bankrupt and finished, there's no coming back from this. Intel fanboys are BTFO and on suicide watch. Lisa Su's ThreadShitter just wiped out Intel overnight. 

 

Pack it up boys the show's over.

That's assuming it's hard to get these things up to their boost speeds, which won't be hard with any adequate cooler. In which case, the 7980XE's boost clock will allow it to outpace the 1950X by a decent amount.

That's also assuming IF doesn't lose some performance when going cross die.

As it stands, ThreadRipper stands a chance for a few reasons: Price, ECC compatibility, and PCIe lane count.

There is a possibility that higher core parts on Intel's side will see memory bottlenecking in AVX instruction sets, but it's still likely that some lower core offerings will match or outpace TR for 3 reasons: Ryzen's IMC hinders AVX2 performance, TR has 1/2 the resources for AVX, and AVX512, which  can be leveraged to boost AVX2 performance, is only on SLX. AVX is far from uncommon for the target users of Intel's HEDT platform.

 

As it stands, Threadripper only threatens part of Intel's HEDT market. Intel still walks away with overclocking enthusiasts, single core performance, AVX performance, and those that want/need Thunderbolt.

28 minutes ago, Daniel644 said:

when was the last time you saw your computer operate at it's base speed?

The last time I completely disabled turbo. Or when not at load.

[Jito463]

1 hour ago, Masada02 said:

Even more, switching to TIM instead of solder saves what, like $5 tops on the process. Both of those methods are incredibly easy to accomplish and all the real cost in the chip is on the silicon itself.

As I understand it, the soldering process increases the chance that the die will fail, so soldering would reduce the yields.  Someone feel free to correct me if I'm mistaken.

[leadeater]

37 minutes ago, Daniel644 said:

Not to give Intel a pass, because they sure don't deserve it, but gotta love the first page of people talking about Base Speeds, like BASE SPEEDS mean a damn thing in this world today or ever, when was the last time you saw your computer operate at it's base speed?

Base clocks tell us a lot about the boost tables and what sort of clocks to expect when multiple cores are loaded. A CPU with a lower base clock means that for an Intel CPU with the same number of cores and the same loading on the CPU the lower base clock CPU will still be at a lower clock under that load as a CPU with a higher base clock.

 

Base clocks might seem rather irreverent since your CPU is almost never at that clock but it tells us far more information.

[leadeater]

11 minutes ago, Drak3 said:

That's assuming it's hard to get these things up to their boost speeds, which won't be hard with any adequate cooler. In which case, the 7980XE's boost clock will allow it to outpace the 1950X by a decent amount.

Board and CPUs will still impose power draw limits so even if we are going to ignore the stock boost tables and try and push all cores to the best we can it's not going to be near the boost clocks. It's one thing to be able to OC a 10 core CPU to near it's boost clocks but an 18 core is much different, math out the actual power each core can get and what it takes to get a stable clock.

 

I don't know how the 7980XE will actually stack up to the 1950X but under standard bios not in LN2 mode and power limits not disabled the 7980XE won't be getting OC'd to all core 3.8GHz or higher and that's with a really good board and cooler too. There will be people that will be able to OC higher than this but I'm talking the average person with reasonable enthusiasts parts.

 

Higher core CPUs don't run at lower clocks just because it's not stable at high clocks or extra binning is required to do it, that is a myth itself mostly. The lower clocks are due to the per core power limitations. If you have a 10 core CPU that draws 200W at load that's 20W per core, an 18 core CPU that draws 200W at load that's 11.11W per core. To push the clocks higher you have to raise the per core power and over 18 cores that's a lot of total package power increase.

 

In the above example if the cores need 20W to be stable at a clock the 18 core CPU will be at 360W, 160W higher than the 10 core CPU.

[VagabondWraith]

Found this on Anandtech. These are Skylake-X boost tables for those interested.

 

[done12many2]

34 minutes ago, VagabondWraith said:

Found this on Anandtech. These are Skylake-X boost tables for those interested.

 

 

Do you have a link? I like to read up more on it.

[VagabondWraith]

7 minutes ago, done12many2 said:

 

Do you have a link? I like to read up more on it.

It was a post but I don't know where he got it from.

 

https://forums.anandtech.com/threads/intel-skylake-kaby-lake-coffee-lake-thread-coffee-lake-s-specs-out-page-554.2428363/page-565#post-39020260

[VagabondWraith]

Here's another thing for those wondering where i9 came from:

 

This addresses an earlier debate on here: "You can trace Intel's path to its new Core i9 chip back to 2013, when the company decided to refocus its energy on enthusiasts. Up until that point, Srivatsa admits, Intel was paying more attention to new form factors, like ultraportables and convertibles, rather than innovating with desktop chips. The company noticed that enthusiasts were the one key audience that was "absolutely livid" that it decided to skip fifth-generation Core chips on desktops in 2014. Their anger was showed they cared far more about Intel's actual products than most other customers. So, it was probably a good idea to show them some love."

 

https://www.engadget.com/2017/08/07/intel-18-core-x-series/

[done12many2]

 

14 minutes ago, VagabondWraith said:

It was a post but I don't know where he got it from.

 

https://forums.anandtech.com/threads/intel-skylake-kaby-lake-coffee-lake-thread-coffee-lake-s-specs-out-page-554.2428363/page-565#post-39020260

 

Yeah, I dug through their entire site and couldn't find it from Anandtech themselves.  That 7920X table seems out of place.  

 

I'll wait for the ones from Intel.  They usually look like these.  

 

 

 

 

 

1 hour ago, leadeater said:

Base clocks tell us a lot about the boost tables and what sort of clocks to expect when multiple cores are loaded. A CPU with a lower base clock means that for an Intel CPU with the same number of cores and the same loading on the CPU the lower base clock CPU will still be at a lower clock under that load as a CPU with a higher base clock.

 

Base clocks might seem rather irreverent since your CPU is almost never at that clock but it tells us far more information.

 

It's a good thing that I never paid attention the very sad 3 GHz base clock on my 5960x.  It'll gladly run 1.7 GHz higher than that (4.7 GHz at 1.28v) with all cores loaded.  

 

 

[VagabondWraith]

delete

 

 

 

[VagabondWraith]

@done12many2

 

Found the original link. Same site that's doing the ongoing testing of the 7900X.

http://www.pcgamer.com/full-details-for-intels-core-i9-processor-lineup/

[done12many2]

23 minutes ago, VagabondWraith said:

@done12many2

 

Found the original link. Same site that's doing the ongoing testing of the 7900X.

http://www.pcgamer.com/full-details-for-intels-core-i9-processor-lineup/

 

Thanks bud.  

[leadeater]

29 minutes ago, done12many2 said:

It's a good thing that I never paid attention the very sad 3 GHz base clock on my 5960x.  It'll gladly run 1.7 GHz higher than that (4.7 GHz at 1.28v) with all cores loaded.  

It'll be interesting to see just how far people can push the top 3 SKUs and the maximum clocks for each of them.

 

Also Intel was heavily power capping the 5960X/6950X CPUs, you don't get more power efficient architectures with only a few more cores at the same TDP but go down in boost clock without power limiting being the culprit. 

 

3960X 6 cores 3.9 boost, 130W TDP

5960X 8 cores 3.5 boost, 140W TDP

6950X 10 cores 3.5 boost, 140W TDP

7900X 10 cores 4.2 boost, 140W TDP

 

The TDP figures as we know are full of shit, anyway that 4.2 boost on the 7900X is from a raised power limit and we clearly have been seeing that with CPU temps and VRM temps due to the high power draw.

[Lays]

god dam some people dumb af, whens the last time you saw an intel chip run at "base" clocks???

 

That's like looking at nvidia gpu base speeds despite GPU Boost, lol.

[done12many2]

23 minutes ago, Lays said:

god dam some people dumb af, whens the last time you saw an intel chip run at "base" clocks???

 

That's like looking at nvidia gpu base speeds despite GPU Boost, lol.

 

We know why they do it bud.