Intel Skylake / Kaby Lake

Page 486 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
21,632
10,845
136
Shivansps said:
You can speculate all you want about how early leaks were incorrect or not based on... fill this here. But the release of the 12 to 18C on time competely kills that FUD.
.
Click to expand...

"On time" would have been August. Not that 2 months is much longer to wait, but still . . .
 
S

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Straight from the horse's mouth...

Intel's news release from the PC Gaming Show

The company also announced timing for the new Intel® Core™ X-series processor family. The 4- to 10-core processors will be available for pre-order beginning on June 19 and will start shipping to consumers the following week. Availability for the rest of the family will soon follow, with the 12-core Intel Core i9-7920X X-series processor expected to start shipping in August, and the Intel Core i9-7940X X-series processor, Intel Core i9-7960X X-series processor and 18-core Intel Core i9-7980X Extreme Edition processor, expected to start shipping in October.
Click to expand...

There's also a web site dedicated to Intel Core X-series now.



 
TheF34RChannel

TheF34RChannel

Senior member
May 18, 2017
786
309
136
I really don't want to wait another week. Reviews were supposed to make my day. Can't take another 7 days of speculation. I'll stop now lol
 
T

Timmah!

Golden Member
Jul 24, 2010
1,419
631
136
Are the base clocks of the 7920X known by now? I guess it interests me the most from the whole line-up.
 
IntelUser2000

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Ajay said:
If there is such a chip, seems like using really fast DDR4 soldered onto the mobo would help (like DDR-4000+).
Or, using GDDR 5/5x/6 would definitely do the trick - though not cheap.
Click to expand...

No, it won't. RX 470 has 224GB/s. If it has 1700SPs, its in the same ballpark. DDR-4000 dual channel is only 64GB/s. Each GDDR5 chip has 32-bit width, meaning you'd need 8 of them running at 7GT/s to do so. Maybe 6 if you go with GDDR5x. That will seriously bloat the package. HBM can do 128GB/s and HBM2 can do 256GB/s in a single package. Not only that, it uses significantly less power per GB/s.

12 cycle -> 14 cycle
Click to expand...

Is it confirmed?

Cannonlake is a low volume tablet and a ultrathin only product. I think Intel has a lot of proving to do with their execution at 10+ and future generations.
Click to expand...

It includes 15W parts, and its not that low volume. Desktop was practically nonexistent with Broadwell. Heck, they had "supply" issues. If you believe Charlie is correct so often then you should also pay attention to his 14nm article about how Broadwell was neutered to rush 14nm in. I'd think that article had some merit because the Broadwell-based Core M sucked.

They do say even 10+ has worse transistor performance compared to 14nm++ though. That means 10 is quite a bit ways from 14nm++ and that wouldn't work well for top tier performance devices.
 
SAAA

SAAA

Senior member
May 14, 2014
541
126
116
IntelUser2000 said:
They do say even 10+ has worse transistor performance compared to 14nm++ though. That means 10 is quite a bit ways from 14nm++ and that wouldn't work well for top tier performance devices.
Click to expand...

I'm not so sure of this, Intel's slides state 10nm still has better characteristics than 14nm+. The node is denser but it should have ~half the power so heat should stay constant for the area.

Early cannonlake leaks on geekbench had 1-1.2GHz speed reported (base?) for U0 stepping, found it on Sysoft too:

http://ranker.sisoftware.net/show_r...d5e3dbeedde4d5f381bc8caacfaa97a781f2cff7&l=en

So early samples are at least dual cores with 1-3.3GHz speed, caches are the same as usual, but standard IGP should double to 48 EUs. I also found this interestingly:

http://ranker.sisoftware.net/show_r...d5e3dae8dae3d0f684b989afcaaf92a284f7caf2&l=en

6C/12T under cannonlake identification, plus others hints here and there that coffe might really just be cannonlake on 14nm.
It looks weird, like a mobile i7 part with reduced L3, it also has only 24 EUs (to save die area? expected if on 14 vs 10 nm).
 
P

plopke

Senior member
Jan 26, 2010
238
74
101
Shivansps said:
Do you seriusly think they can get sku out of thin air on a blink of an eye? On a platform that is does not longer shares server skus? And the whole "12C thing" was just stupid, they whould have never only sold the 12C MCC die at X, when they can sell 14C at X+$, 16C at X+$$ and 18C at X+$$$. After they did EVERY attempt to make LGA2066 not a threat to Xeon.

You can speculate all you want about how early leaks were incorrect or not based on... fill this here. But the release of the 12 to 18C on time competely kills that FUD.

Im not going to comment futher on this issue. Closed case. /Period.
Click to expand...

on a tech forum were we all love to speculate , this above 10 core is one touchy subject it seams. I did not even mention it was an sku out of thin air. Surely companies make a entire product stack and then decide on market demand/position/profit/competition/yields/.... which one should be actually released. OFF COURSE it is not a sku out of thin air ,that was my entire point. I even emphasized the distinction in bold between the ability to make and what is shipped in the end to the consumer.

Just look at all these tech shows, they are full of stuff that might never make it to market because the business people decide what should come out. The entire X299 line-up is one big board of business managers fiddling with the stack of products. You said it even yourself with making LGA2066 move away from XEON.
 
L

lolfail9001

Golden Member
Sep 9, 2016
1,056
353
96
Shivansps said:
Well at least we can finally put to rest the BS that 18C was a response to TR, no way they could have responded this quickly on LGA2066.
Click to expand...
The unlocked version of 18C SKU is a response to TR. The 18C SKU itself is not and we knew that because of iMac Pro (Skylake-W has HCC dies and we knew that for a while) not some Intel announcement.
 
  • Like
Reactions: formulav8 and Phynaz
IntelUser2000

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
SAAA said:
I'm not so sure of this, Intel's slides state 10nm still has better characteristics than 14nm+.
Click to expand...

This was from Intel manufacturing day not too long ago: https://3s81si1s5ygj3mzby34dq6qf-wp...ntel-xeon-process-technology-enhancements.jpg

They need 10nm++ to beat 14nm++ in performance. So the first two 10nm versions will reduce power and area required but achieve worse performance.

The reduction in power isn't so noticeable anymore either. The 22nm and 14nm process both brought minor reductions in power. The 40% performance stated for 22nm only came true for Bay Trail, meaning transistor-level improvements happen only for selective devices.*

*Big part of the reason is previously the process gains came easy and fast so there wasn't much focus on the architectural level. Ever since the gains started diminishing more advanced architectural features were introduced. Now, the gains from new process isn't as clear, not just because process itself has little gains. Technologies like C9/C10 idle states greatly diminish the impact of new process, because idle power is dependent on other factors. Transistor-level power reduction is only true for load, and that's what used to matter.

Let's see others which reduce the impact of a new process.
-Power Gating: Certain sections of the chip can be powered off to reduce leakage power which can become significant in lower power modes. A new process with improved leakage characteristics benefit only in situations where Power Gating can't be applied
-Heterogenous architectures and accelerators: These are circuits that can improve its performance either being wide and slow clocked or narrow and fast clocked. If the transistors are optimized for CPU, meaning high performance and clocks, it may limit performance of the said accelerators that best operate wide and slow. The same will be true by using GPU optimized transistors, which will slow down CPU-like parts of the circuit.

The choices that are made design-level start impacting how much gain a new process could bring. Of course the reason they are there in the first place is because new process does not bring as much benefits as before.
 
Last edited:
LTC8K6

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
Why is the 7800X only DDR4-2400? Not that it matters a lot I guess, since people will run much faster memory with it.
 
  • Like
Reactions: Drazick
S

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
ASUS X299 Prime Deluxe Preview
https://www.overclock3d.net/reviews/cpu_mainboard/asus_x299_prime_deluxe_preview/1

Gigabyte X299 Aorus Gaming 7 Preview
https://www.overclock3d.net/reviews/cpu_mainboard/gigabyte_x299_aorus_gaming_7_preview/1

Bonus:

12.jpg

 
  • Like
Reactions: Phynaz
TheF34RChannel

TheF34RChannel

Senior member
May 18, 2017
786
309
136
twothreefive said:
Any thoughts on this motherboard - Biostar Racing x299:

https://ocaholic.ch/modules/news/article.php?storyid=16602

I am looking for more of a basic motherboard without all the RGB crap that will overclock a i9 7900x well. Rumor has it this board will be around $250.
Click to expand...

I'd look at the Asus X299 Prime Deluxe personally, way better quality (general Asus vs Biostar) and better BIOS (again, general). You can turn off any RGB features.
 
T

Timmah!

Golden Member
Jul 24, 2010
1,419
631
136
What are the prices of these x299 mobos? Are they all generally in 400-500 range, or is there any for 200-250? I bought my X99 last year for about 250, no way i would pay more than that.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom