The infamous Intel Core i9 processor... finally real?

[Topweasel]

Why do you think $1700 is especially expensive? HEDT processors are, and have always been, toys for enthusiasts seeking bragging rights. Consumers buy laptops and don't care about socketed CPUs. Professionals using dual Xeon workstations (e.g. E5-2687Wv4) are paying $4000 in CPUs and likely at least another $4000 in DRAM. Large scale datacenter operators are buying top-bin E5s at $4500 each and paying far more in DRAM (12x 128 GB DIMMs is $14000+). Why do you think Facebook and Google bought out the entire first year's production of Skylake-SP if "Threadripper" or Naples are going to be so amazing?


The conversation in this thread has gone more like this:
(Us) The new Ferrari is going to be awesome.
(You) Nuh-uh, the everyday users are going to buy Lamborghinis
(Us) Working men buy trucks, not sports cars
(You) (crickets...)

Because it is a consumer CPU and not a Workstation or server chip. People spend way more then they need to on Toys. But what is going to make people want to spend 2-3 times as much on this 12c CPU for entertainment when a 8c version will clock higher, with less power, and perform better in all but workstation applications. Those guys who need them in Workstations are buying Xeon's maybe as expensive and running them at stock in warranted machines. I am also not writing off people purchasing the chip, what I am writing off is people buying the chip and immediately running this at 300w to get 4.3+ GHz Overclocks. The main fact that started all of this was me trying to set a decent estimate on baseclocks and personally didn't see a >4GHz overclock a reasonable estimation. When I said that I meant 24/7 stable as I don't someone clocking it up for a handful of benchmarks count specially as again quoted from others besides me as a "gimme". Running a 140w-160w CPU at 300w+ is asking a lot of it, the cooling, and the board and therefore not a gimme. If you think that it is nothing and completely reasonable then there is little more to talk about. That isn't even counting the realism of nobody even if they could actually doing so.

It's more like
You.) Hey a bolt on super charger could take a LaFerrari engine up to 1500HP.
Me.) a Laferrari engine is a little high strung and has pretty tight times and also is an incredibly expensive engine. Snapping a supercharger on it could cause it to blow very easily and there are probably better alternatives.
You.) But Bob got 1100HP from a supercharger on his LS7.
Me.) So? Maybe even if you could pump out 1500HP out of the V12 in the Ferrari it's still academic because nobody is going to do it.
You.) Well if people really wanted the best HP they would get something rated higher like Veryon engine and not mess with it.

I mean how am I supposed to answer that? The point becomes less and less about what Intel is actually going to be offering and more about an imaginary world where money, development, cooling, and power can be completely ignored and the Max OC projection possible is the only measuring stick for a CPU. While then turning around and saying that their are tons better options where people who actually care about performance will buy and the price of this CPU is insignificant to some of those options. It's a real circular logic that I can't grasp and I apologize for that. All I can do is continue by saying that I think people will be surprised at how low the stock clocks will be and people thinking a free 4.3.-4.5 clock are going to be disappointed and frankly no one outside the reviewers and sponsored overclockers are going to even try to get close. It's still going to be an amazing CPU for people who does purchase it, I just don't think that it will have the clock advantages that the 4c chips have.

 

[blue11]

...

Regarding pricing:

1. Enthusiasts spend lots of money on things that make no sense (like the cars you mentioned).
2. Intel doesn't care how many HEDT CPUs they sell. They exist for marketing narrative reasons.
3. Intel especially doesn't care how many of the top-bin HEDT CPU they sell. Likely less than 1000 units

Regarding other SKUs:

1. I bring up Xeon workstations and servers to eliminate the specter of people using HEDT CPUs for actual work. This doesn't happen.
2. This leaves gamers and OC enthusiasts. Gamers won't be buying $1k+ CPUs, if only because they'd rather buy $1k+ GPUs.
3. So the only people left that would buy these top-bin HEDT CPUs are those that want bragging rights.

Regarding technical details:

1. 4.5 GHz OC will only need 300+ W when running AVX workloads. You can just use a lower frequency for those.
2. For scalar workloads, the power consumption will be far lower, likely near 200 W.
3. Almost nobody will run a workload that fully loads all 12 cores (except benchmarks), because those people would be buying Xeons.

 

[whm1974]

Please correct me if I'm wrong but I'm thinking that from all the reading I've been doing is that most folks even the average power user wouldn't get that much benefit from having more then 8C/16T anyway. Hell come to think of it, the vast majority of people will do just fine with just 4C/4T.

 

[ozzy702]

Please correct me if I'm wrong but I'm thinking that from all the reading I've been doing is that most folks even the average power user wouldn't get that much benefit from having more then 8C/16T anyway. Hell come to think of it, the vast majority of people will do just fine with just 4C/4T.

The vast majority of users would do just fine with a $60 G4560. Most people just use Office and browse the web, all of which runs great on a G4560 paired with at least 8gig of RAM and an SSD.

That being said, I'm one of those guys that isn't so concerned with money as I am getting the best performance available at the time. The Skylake-X 8 and 10 core are looking promising so far, we'll know soon just how well they perform.

Slightly off topic, what's the successor to Skylake-X? I'm assuming it will use the same socket. I wonder if they will end up bumping up the L3 Cache once they have more room thanks to the die shrink.

 

[frozentundra123456]

Because it is a consumer CPU and not a Workstation or server chip. People spend way more then they need to on Toys. But what is going to make people want to spend 2-3 times as much on this 12c CPU for entertainment when a 8c version will clock higher, with less power, and perform better in all but workstation applications. Those guys who need them in Workstations are buying Xeon's maybe as expensive and running them at stock in warranted machines. I am also not writing off people purchasing the chip, what I am writing off is people buying the chip and immediately running this at 300w to get 4.3+ GHz Overclocks. The main fact that started all of this was me trying to set a decent estimate on baseclocks and personally didn't see a >4GHz overclock a reasonable estimation. When I said that I meant 24/7 stable as I don't someone clocking it up for a handful of benchmarks count specially as again quoted from others besides me as a "gimme". Running a 140w-160w CPU at 300w+ is asking a lot of it, the cooling, and the board and therefore not a gimme. If you think that it is nothing and completely reasonable then there is little more to talk about. That isn't even counting the realism of nobody even if they could actually doing so.

It's more like
You.) Hey a bolt on super charger could take a LaFerrari engine up to 1500HP.
Me.) a Laferrari engine is a little high strung and has pretty tight times and also is an incredibly expensive engine. Snapping a supercharger on it could cause it to blow very easily and there are probably better alternatives.
You.) But Bob got 1100HP from a supercharger on his LS7.
Me.) So? Maybe even if you could pump out 1500HP out of the V12 in the Ferrari it's still academic because nobody is going to do it.
You.) Well if people really wanted the best HP they would get something rated higher like Veryon engine and not mess with it.

I mean how am I supposed to answer that? The point becomes less and less about what Intel is actually going to be offering and more about an imaginary world where money, development, cooling, and power can be completely ignored and the Max OC projection possible is the only measuring stick for a CPU. While then turning around and saying that their are tons better options where people who actually care about performance will buy and the price of this CPU is insignificant to some of those options. It's a real circular logic that I can't grasp and I apologize for that. All I can do is continue by saying that I think people will be surprised at how low the stock clocks will be and people thinking a free 4.3.-4.5 clock are going to be disappointed and frankly no one outside the reviewers and sponsored overclockers are going to even try to get close. It's still going to be an amazing CPU for people who does purchase it, I just don't think that it will have the clock advantages that the 4c chips have.

8 core Haswell HEDT I believe could be expected to overclock to the mid 4ghz range, IIRC. Since Skylake/KL quads overclock at least as well (usually better) as haswell, I think mid 4ghz or a bit more is entirely reasonable for the 8 core chip. 10 and 12 cores, well that is more of a guess, but I think the 6950x could reasonably be expected to overclock to 4ghz, and SL/KL definitely overclock better than broadwell, so again I think slightly over 4ghz in an entirely reasonable expectation.

 

[Topweasel]

The vast majority of users would do just fine with a $60 G4560. Most people just use Office and browse the web, all of which runs great on a G4560 paired with at least 8gig of RAM and an SSD.

That being said, I'm one of those guys that isn't so concerned with money as I am getting the best performance available at the time. The Skylake-X 8 and 10 core are looking promising so far, we'll know soon just how well they perform.

Slightly off topic, what's the successor to Skylake-X? I'm assuming it will use the same socket. I wonder if they will end up bumping up the L3 Cache once they have more room thanks to the die shrink.

KL-X or maybe a skip directly to CL-X. But both will be 14nm so no.

 

[beginner99]

None of this has anything to do with servers. As I said in the previous post, for servers, the target AMD will have to beat is:

28 cores
2.5 GHz base
2.8 GHz (estimated) turbo
Unified cache
205 W TDP

The will beat that (except unified cache of course). Naples is 32c/64t and as was already seen with Ryzen (Ryzen strictly technical thread) the GF/Samsung process is actually highly efficient at lower clocks. In typical AMD fashion All current Ryzen chips run way, way above the best efficiency. 4 Ghz is the process + design combos limit so it's fairly possible to get say 16 Ryzen cores at 190w TDP and 4 Ghz. Simply a matter of higher TDP.

But back to efficiency. In the targeted 2.5-2.8 GHz range Ryzen easily showed about double performance/watt. If we say 32c at 4 Ghz use 380w with double efficiency at 2.5 Ghz range this will be 190w and hence completely competitive with intel. I'm gonna make a guess and say it will also cost 1/3 of intels solution and AMD then has the clear winner.

 

[blue11]

I'm gonna make a guess and say it will also cost 1/3 of intels solution and AMD then has the clear winner.

Even if Naples were free, nobody would take it if it underperforms Skylake-SP. This is because CPU costs are a fraction of DRAM costs, and anything that forces you to buy more systems and hence more DRAM is dead in the water. Everything else you said is speculation.

EDIT: This is not to say that Naples will underperform, but that trying to undercut Intel in server is a losing game, since CPUs are already cheap.

 

[Hi-Fi Man]

Regarding other SKUs:

1. I bring up Xeon workstations and servers to eliminate the specter of people using HEDT CPUs for actual work. This doesn't happen.
2. This leaves gamers and OC enthusiasts. Gamers won't be buying $1k+ CPUs, if only because they'd rather buy $1k+ GPUs.
3. So the only people left that would buy these top-bin HEDT CPUs are those that want bragging rights.

What? Speak for yourself. Half of the reason I bought my HEDT was for getting work done better than the LGA 11xx stuff. The other half was so I could still have decent single thread performance that a Xeon doesn't have. I don't give a d*** about epeen and crap. Lightroom and After Effects will happily take advantage of HEDT. In fact yesterday my render for a short 25 sec video ended up using about 29GiB of RAM and plenty of CPU time.

 

[lolfail9001]

What? Speak for yourself. Half of the reason I bought my HEDT was for getting work done better than the LGA 11xx stuff. The other half was so I could still have decent single thread performance that a Xeon doesn't have. I don't give a d*** about epeen and crap. Lightroom and After Effects will happily take advantage of HEDT. In fact yesterday my render for a short 25 sec video ended up using about 29GiB of RAM and plenty of CPU time.

If you're talking about 6800k in your sig, then i think you have to excuse him for forgetting that HEDT has $400 CPU as well as $1700 one. He was talking about the $1700 one all along.

 

[Hi-Fi Man]

If you're talking about 6800k in your sig, then i think you have to excuse him for forgetting that HEDT has $400 CPU as well as $1700 one. He was talking about the $1700 one all along.

Possibly but my use case scales up as well so long as you're willing to spend more. I'm not but others might.

 

[Denithor]

It's more like
You.) Hey a bolt on super charger could take a LaFerrari engine up to 1500HP.
Me.) a Laferrari engine is a little high strung and has pretty tight times and also is an incredibly expensive engine. Snapping a supercharger on it could cause it to blow very easily and there are probably better alternatives.
You.) But Bob got 1100HP from a supercharger on his LS7.
Me.) So? Maybe even if you could pump out 1500HP out of the V12 in the Ferrari it's still academic because nobody is going to do it.
You.) Well if people really wanted the best HP they would get something rated higher like Veryon engine and not mess with it.

Uh, yes they will. I grew up with the guy doing it.

http://www.undergroundracing.com/

 

[dullard]

Most people would assume looking at historical pricing that it would slide into the previous head honcho's spot. But I am not sure did that happen last time? I mean historically looking back just a single generation Intel slotted the new top core count HEDT Intel increased the cost by 70%. I was projecting a ~15% decrease in price.

For decades, Intel has had pricing brackets. When new chips come out, they are almost always in the same bracket as the chip that it replaces. The exact dollar value varies a bit (maybe +-$50 or so), but not by much from the old chip. That is unless one of two things happen. (1) There wasn't an equivalent chip or (2) This is a revolutionary new chip design. There do seem to be equivalent chips (roughly the same model number) and this is not a revolutionary new design. So, I expect Intel to use their old price slots. I could be wrong, but I have no reason to expect Intel to change either.

Intel's historical HEDT prices have historically been about $1000, about $600, and about $300 to $400. When they added a fourth slot with Broadwell-E, it was about $1700.

Sandy Bridge (3 chips, 3 slots):
$999 (launch): 3960X, 6 cores, 3.3 GHz, 15 MB L3 cache
* Note: A year later the 3960X was replaced by the 3970X at the same $999 price and same specs but 3.5 GHz.
$583 (launch): 3930X, 6 cores, 3.2 GHz, 12 MB L3 cache
$294 (launch): 3820, 4 cores, 3.6 GHz, 10 MB L3 cache

Ivy Bridge (3 chips, 3 slots):
$999 (launch): 4960X, 6 cores, 3.6 GHz, 15 MB L3 cache
$583 (launch): 4930K, 6 cores, 3.4 GHz, 12 MB L3 cache
$323 (launch): 4820K, 4 cores, 3.7 GHz, 10 MB L3 cache

Haswell (3 chips, 3 slots):
$999 (launch): 5960X, 8 cores, 3.0 GHz, 20 MB L3 cache
$583 (launch): 5930K, 6 cores, 3.5 GHz, 15 MB L3 cache
$389 (launch): 5820K, 6 cores, 3.3 GHz, 15 MB L3 cache

Broadwell-E (notice that there are now 4 chips instead of 3, so there was a fourth price slot added):
$1723 (launch): 6950X, 10 cores, 3.0 GHz, 25 MB L3 cache
$1089 (launch): 6900K, 8 cores, 3.2 GHz, 20 MB L3 cache
$617 (launch): 6850K, 6 cores, 3.6 GHz, 15 MB L3 cache
$434 (launch): 6800K, 6 cores, 3.4 GHz, 15 MB L3 cache

Kaby Lake / Skylake (4 chips, 3 slots):
$? (launch): 7920X, 12 cores, ? GHz, 16.5 MB L3 cache
$? (launch): 7900K, 10 cores, 3.3 GHz, 13.75 MB L3 cache
$? (launch): 7820K, 8 cores, 3.6 GHz, 11 MB L3 cache
$? (launch): 7800K, 6 cores, 3.5 GHz, 8.25 MB L3 cache

I expect the next HEDT chips to be priced about the same as Broadwell-E. Maybe a hair less now that Ryzen is actually a competitor.

The 6900K was about a 9% price increase over the 5960K that it replaced, not 70% (it had a 6% base clock increase and faster memory, so the 9% price increase wasn't that painful). The reason that you saw a 70% price increase is since Intel went from 3 HEDT chips to 4 HEDT chips and created a new price slot.

 

[Ajay]

IDK, seems like Intel would be in a better position if launch prices were more like Haswell-E for the lower three price points. If Intel prices like BWE, then I would imagine many would just go with AMD (unless AMD's rollout is horrible).

 

[Topweasel]

Uh, yes they will. I grew up with the guy doing it.

http://www.undergroundracing.com/

Well not quite because it's not a LaFerrari V12. Also it's not just the engine like I was talking about. But it's also the exception that proves the rule. Sure there are a very small handful of guys out there that buy an already expensive car and dump even more money into making it faster. Cars like the Huracan and 458 or 488 are not limited production cars. They might have limited edition versions. But you aren't destroying the value of instant collector's car. Because that's what doing this to a LaFerrari really is. Instead of a car thats value doubled the day you received it becomes a car any buyer with the money would actively avoid. The few Enzo's and even an F40 that has been tuned up like this were salvage titled cars. As for the engine my point to install in your own car. Just because you could in theory find 100k Enzo engine (don't know if you really can) and get it up to 1500hp, you wouldn't, you would get a small block or something more reasonable and then toss a bunch of parts at it.

Also when someone says no one. They mean that even within the limitations of the particular market that the actual amount of people doing is to significantly small to worry about. Not that in a world with 7+ Billion people that absolutely no one would do it.

 

[Topweasel]

For decades, Intel has had pricing brackets. When new chips come out, they are almost always in the same bracket as the chip that it replaces. The exact dollar value varies a bit (maybe +-$50 or so), but not by much from the old chip. That is unless one of two things happen. (1) There wasn't an equivalent chip or (2) This is a revolutionary new chip design. There do seem to be equivalent chips (roughly the same model number) and this is not a revolutionary new design. So, I expect Intel to use their old price slots. I could be wrong, but I have no reason to expect Intel to change either.

Intel's historical HEDT prices have historically been about $1000, about $600, and about $300 to $400. When they added a fourth slot with Broadwell-E, it was about $1700.

Sandy Bridge (3 chips, 3 slots):
$999 (launch): 3960X, 6 cores, 3.3 GHz, 15 MB L3 cache
* Note: A year later the 3960X was replaced by the 3970X at the same $999 price and same specs but 3.5 GHz.
$583 (launch): 3930X, 6 cores, 3.2 GHz, 12 MB L3 cache
$294 (launch): 3820, 4 cores, 3.6 GHz, 10 MB L3 cache

Ivy Bridge (3 chips, 3 slots):
$999 (launch): 4960X, 6 cores, 3.6 GHz, 15 MB L3 cache
$583 (launch): 4930K, 6 cores, 3.4 GHz, 12 MB L3 cache
$323 (launch): 4820K, 4 cores, 3.7 GHz, 10 MB L3 cache

Haswell (3 chips, 3 slots):
$999 (launch): 5960X, 8 cores, 3.0 GHz, 20 MB L3 cache
$583 (launch): 5930K, 6 cores, 3.5 GHz, 15 MB L3 cache
$389 (launch): 5820K, 6 cores, 3.3 GHz, 15 MB L3 cache

Broadwell-E (notice that there are now 4 chips instead of 3, so there was a fourth price slot added):
$1723 (launch): 6950X, 10 cores, 3.0 GHz, 25 MB L3 cache
$1089 (launch): 6900K, 8 cores, 3.2 GHz, 20 MB L3 cache
$617 (launch): 6850K, 6 cores, 3.6 GHz, 15 MB L3 cache
$434 (launch): 6800K, 6 cores, 3.4 GHz, 15 MB L3 cache

Kaby Lake / Skylake (4 chips, 3 slots):
$? (launch): 7920X, 12 cores, ? GHz, 16.5 MB L3 cache
$? (launch): 7900K, 10 cores, 3.3 GHz, 13.75 MB L3 cache
$? (launch): 7820K, 8 cores, 3.6 GHz, 11 MB L3 cache
$? (launch): 7800K, 6 cores, 3.5 GHz, 8.25 MB L3 cache

I expect the next HEDT chips to be priced about the same as Broadwell-E. Maybe a hair less now that Ryzen is actually a competitor.

The 6900K was about a 9% price increase over the 5960K that it replaced, not 70% (it had a 6% base clock increase and faster memory, so the 9% price increase wasn't that painful). The reason that you saw a 70% price increase is since Intel went from 3 HEDT chips to 4 HEDT chips and created a new price slot.

I know. It was a response to the statement that my price guess ($1500) was FUD. You are missing the context there. I was just messing with him because in theory the sample size is two small. They had two of the same line up core wise. Had a increase in cores and kept pricing and then had another increase in cores and drasically shifted pricing. This is probably why Intel redid the cache for SL-X because they have their pricing based on silicon size and since they are still running 14nm increasing the core count meant an increase in die size which meant them pricing it that much higher. So if they kept the cache the same it might mean it would cost 2500+. Knowing the actual die size will probably tell us what the price will be.

 

[blue11]

They had two of the same line up core wise. Had a increase in cores and kept pricing and then had another increase in cores and drasically shifted pricing. This is probably why Intel redid the cache for SL-X because they have their pricing based on silicon size and since they are still running 14nm increasing the core count meant an increase in die size which meant them pricing it that much higher. So if they kept the cache the same it might mean it would cost 2500+. Knowing the actual die size will probably tell us what the price will be.

Pricing of HEDT CPUs has almost no relation to cost. They are priced where they are ($1000-$1700) because Intel doesn't expect to actually ship any units. The "missing" L3 cache was mostly added to the L2, so the total amount of cache memory hasn't changed much, nor has the cost. The Skylake-SP LCC die in fact only goes to 10 cores, so the purported 7920X is going to be HCC-derived, which would drive up costs. On the other hand, the Gold 5122 (4C/8T) was leaked at $1200 and is XCC-derived, so the relation between price and cost is quite loose.

 

[Topweasel]

Pricing of HEDT CPUs has almost no relation to cost. They are priced where they are ($1000-$1700) because Intel doesn't expect to actually ship any units. The "missing" L3 cache was mostly added to the L2, so the total amount of cache memory hasn't changed much, nor has the cost. The Skylake-SP LCC die in fact only goes to 10 cores, so the purported 7920X is going to be HCC-derived, which would drive up costs. On the other hand, the Gold 5122 (4C/8T) was leaked at $1200 and is XCC-derived, so the relation between price and cost is quite loose.

You are right I was thinking back to my question a couple days ago about unique cache. Either which way I still doubt that my $1500 estimate is FUD.