Intel's Future Chips: News, Rumours & Reviews

[adamsleath]

...well, the 8 core intel desktop, priced at 550usd? more? less? performance per dollar. ryzen is 20-30ish% points behind in single threaded now. this gap could be closed to around 10% by pinnacle ridge?

i think amd will compete. and 7nm process will close the gap further. what will intel do?

the top desktop cpu is relatively too expensive on a perf/$ basis; historically they always have been.

 

[juanrga]

MERGED QUESTION
Question from AngeNoir0324 : "Discussion on Leaked Rumor of 9th gen Intel CPUs"

Gamermeld just release a video covering this topic in part, I’ll summarize it, but you should go watch his video.
i7 will be 8 core/ 16 thread behemoths
i5 will be 6 core, some with hyper threading
i3 will be 4 core cpus some with hyper threading

My main questions today are this:

-How do you guys think this reflects on how Coffee Lake was thrown together?

-How can AMD’s Ryzen chips compare, or will the expensive z370 and z390 mobos (the only way to use The full 8 cores of the 9700k) still make Ryzen a compelling option?

-What kind of preformance must Ryzen gen 2 have to stay relevant and powerful if there usb stock of 9th gen intel CPUs?


Thanks for this little discussion.

I would prefer for Icelake

i7 = 8C / 16T
i5 = 6C / 12T
i3 = 4C / 8T


- Kabylake and CoffeeLake are stopgaps created due to delays with 10nm node.

- Intel is the preferred choice for gamers. For everything else it depends.

- 6C CoffeLake beats 8C Ryzen even in multithreaded scenarios as Blender, because each Intel core is about 40% faster than Zen core. Icelake is expected to bring two more cores and a frequency update. Next year Pinnacle Ridge is expected to be Zen with slightly higher clocks, maybe 10%. So the performance gap will increase.

 

[aldaia]

Right now 1800X is selling for $300 aprox while 8700K is selling for about $400 25% more. Not to mention that motherboards are also more expensive.
Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call)

Pinacle ridge is expected on Q1 2018, while rumors point out that 9th generation won't be available until late 2018.
Ryzen 2 is expected early 2019. If those rumors are true, 9th generation will have an advantage for a litle while, but will compete with ryzen 2 later on. Some rumors point out Ryzen 2 may offer up to 16C/32T on a single die.

 

[adamsleath]

https://www.gamersnexus.net/hwreviews/3076-intel-i7-8700k-review-vs-ryzen-streaming-gaming-overclocking/page-6
note the blender benchmarks.

https://www.pcworld.com/article/3230369/components-processors/core-i7-8700k-review-prices-specs-benchmarks.html

 

[aldaia]

https://www.pcworld.com/article/3230369/components-processors/core-i7-8700k-review-prices-specs-benchmarks.html

This one is interesting, shows several multi-threaded benchmarks besides Blender: Cinebench, POV-Ray, Handbrake, Premiere CC
In all of them the $280 1700x has clearly better performance than the $414 8700k except blender where the $414 part ties with a $280 part and still loses to the $319 1800X

 

[gamerk316]

https://www.pcworld.com/article/3230369/components-processors/core-i7-8700k-review-prices-specs-benchmarks.html

This one is interesting, shows several multi-threaded benchmarks besides Blender: Cinebench, POV-Ray, Handbrake, Premiere CC
In all of them the $280 1700x has clearly better performance than the $414 8700k except blender where the $414 part ties with a $280 part and still loses to the $319 1800X

In applications that scale, having more cores tends to trump per-core performance. Mathematically speaking, if you have twice the cores on one CPU, the competing CPU would need double the per-core performance (IPC * Clock) to keep up.

AVX2 helps Intel in some cases. It's possible wider AVX2 adoption will shift things back Intels way, though I doubt the current situation changes much until Intels next refresh.

It's worth noting that in benchmarks that don't scale, Intel still crushes AMD.

 

[adamsleath]

it's around 30% +/- single core advantage. slightly reduced by Pinnacle Ridge.
dont forget that the new 10nm process wont match the performance of 14nm++ obviously 8 core intel will trump (lol) all desktop ryzens...and it may even be on 14nm??

 

[juanrga]

Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call).

BMW: i7-8700k is 12% faster than R7-1800X. The R7 has 33% moar cores. So

1.12 * 8/6 = 1.49.

GooseBerry: i7-8700k is 8% faster than R7-1800X. So

1.08 * 8/6 = 1.44.

So each core in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

 

[juanrga]

As I posted some pages ago 10nm is ready and works fine, but Intel want better yields

http://fudzilla.com/news/processors/45091-intel-10nm-is-better-than-everyone-thinks

 

[-Fran-]

Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call).

BMW: i7-8700k is 12% faster than R7-1800X. The R7 has 33% moar cores. So

1.12 * 8/6 = 1.49.

GooseBerry: i7-8700k is 8% faster than R7-1800X. So

1.08 * 8/6 = 1.44.

So each core in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Uhm... So motorcycles that make the 1/4 mile 2 second faster are actually 45% faster than cars because they have 2 wheels instead of 4? 😵

What kind of logic is that? XD

You just do the % against the scores if you want an overall measurement. What you did there is just an insult to everyone's intelligence:

345.3/303.02 = 1.1395... ~ 14% *overall* slower in blender in that particular test.

 

[adamsleath]

....geez. look at intel's own presentations regarding their 10nm process. no need to bs. it is what it is.

just look at the 7820X that'll give you a ballpark. (+15ish% multicore perf. vs ryzen, both at OC clocks, & a wide range of benchmarks)
---------
https://www.cpubenchmark.net/singleThread.html
+40% passmark single threaded , stock clocks with turbo enabled. 8700k vs 1600x

based on a wider range of single threaded tests it is around 30+% (between 20-40% Single thread? depending on the test)

(https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+7+1800X&id=2966) 5% difference across the board...

for multicore performance ryzen 8 core is ahead (taking a wide range of tests) of 8700k, both at OC clocks.

1800X is still cheaper to buy now than the 8700k....with the recent discounting....or the 1700x.
-----
the single core advantage will decrease next year.

 

[aldaia]

 

[YoAndy]

Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call).

BMW: i7-8700k is 12% faster than R7-1800X. The R7 has 33% moar cores. So

1.12 * 8/6 = 1.49.

GooseBerry: i7-8700k is 8% faster than R7-1800X. So

1.08 * 8/6 = 1.44.

So each core in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Uhm... So motorcycles that make the 1/4 mile 2 second faster are actually 45% faster than cars because they have 2 wheels instead of 4? 😵

What kind of logic is that? XD

You just do the % against the scores if you want an overall measurement. What you did there is just an insult to everyone's intelligence:

345.3/303.02 = 1.1395... ~ 14% *overall* slower in blender in that particular test.

Don't you understand what he is trying to explain?..Intel 6 core CPU is 12% faster than AMD 8 core CPU that has 33% more cores (2 more cores 4 more threads)...

Meaning that EACH CORE in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Overall performance a six core CoffeLake chip still 12% percent faster than an 8 core AMD chip...


Let me explain it better for you:
6 horses from Intel have 12% more horsepower than 8 horses from AMD. Meaning that a single horse from Intel has 44-49% more horsepower than a single horse from AMD..So what's AMD response?.. Since horse per horse they can't beat intel, what are they doing? They keep adding more weaker horses. Got it? 😉

 

[juanrga]

Spoiler

Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call).

BMW: i7-8700k is 12% faster than R7-1800X. The R7 has 33% moar cores. So

1.12 * 8/6 = 1.49.

GooseBerry: i7-8700k is 8% faster than R7-1800X. So

1.08 * 8/6 = 1.44.

So each core in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Uhm... So motorcycles that make the 1/4 mile 2 second faster are actually 45% faster than cars because they have 2 wheels instead of 4? 😵

What kind of logic is that? XD

You just do the % against the scores if you want an overall measurement. What you did there is just an insult to everyone's intelligence:

345.3/303.02 = 1.1395... ~ 14% *overall* slower in blender in that particular test.

Don't you understand what he is trying to explain?..Intel 6 core CPU is 12% faster than AMD 8 core CPU that has 33% more cores (2 more cores 4 more threads)...

Meaning that EACH CORE in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Overall performance a six core CoffeLake chip still 12% percent faster than an 8 core AMD chip...


Let me explain it better for you:
6 horses from Intel have 12% more horsepower than 8 horses from AMD. Meaning that a single horse from Intel has 44-49% more horsepower than a single horse from AMD..So what's AMD response?.. Since horse per horse they can't beat intel, what are they doing? They keep adding more weaker horses. Got it? 😉

Exactly.

 

[juanrga]

Anandtech, PCWorld, TechReport, and ExtremeTech are using beta BIOS and/or disabling performance profiles.

PcPer, Arstechnica, OC3D, Techspot and others aren't.

Extremetech gets the above scores for Goosebery whereas TecSpot and PcPEr get something different.

From Arstechnica:

Performance

As you'd expect, running six cores at 4.7GHz results in some stellar benchmark results (more so if you can brave 5.0GHz). Even though it has two fewer cores than the Ryzen 1800X (a CPU that costs a hefty £437), the 8700K comes in faster in many production workloads. It's four seconds quicker in Blender at stock, and 11 seconds quicker when overclocked. It's faster at Handbrake video encoding too, and miles ahead in 7-Zip's synthetic benchmark, which tends to favour clock speed even in multithreaded mode.

It's only in PovRay and Cinebench that 1800X comes out on top—and only then by a small amount.

So with BETA bios and performance profiles disabled the CFL core is 'only' 30% faster than Zen core in most multithreaded scenarios. With final BIOS and performance profiles enabled the CFL core is 45% faster than Zen core in most multithreaded scenarios.

And throughput-workloads such as Blender, CineBench, POV-Ray... are best cases for the throughput-optimized Zen microarchitecture. The gap between CFL and Zen cores is higher on latency-workloads such as games, where 6C CFL can be up 60% faster than 8C Zen.

 

[juanrga]

1800X is still cheaper to buy now than the 8700k....with the recent discounting....or the 1700x.

At time of writing this I can purchase the i7-8700k by 409€ whereas the R7-1800X is 417€.

https://www.pccomponentes.com/amd-ryzen-7-1800x-40ghz-sin-disipador
https://www.pccomponentes.com/intel-core-i7-8700k-37ghz-box

So the R7-1800X is both slower and more expensive.

the single core advantage will decrease next year.

Maybe but Intel will continue being about 30% ahead in rendering applications and about 50% on games.

 

[juanrga]

Kabylake-R mobile chips are much more efficient than Zen mobile.

https://techreport.com/blog/32904/how-much-does-screen-size-matter-in-comparing-ryzen-mobile-and-kaby-lake-r-battery-life

Reviewer eliminated the laptop screen size factor to get a direct comparison of the efficiency of each chip.

Anxiously expecting the 10nm Canonlake chips...

 

[goldstone77]

Kabylake-R mobile chips are much more efficient than Zen mobile.

https://techreport.com/blog/32904/how-much-does-screen-size-matter-in-comparing-ryzen-mobile-and-kaby-lake-r-battery-life

Reviewer eliminated the laptop screen size factor to get a direct comparison of the efficiency of each chip.

Anxiously expecting the 10nm Canonlake chips...

Instead of doing real world testing?..... I can see the difference in battery life if you were going to use a monitor, and leave your laptop unplugged sitting next to an outlet... But I get what the article is pointing out even if I don't find this information useful for the average everyday user that would simply plug their laptop into the wall outlet it's sitting next too.

While I wanted to get real-world battery-life testing out of the way first, I can certainly appreciate the elegance of leveling the playing field that way. Now we have.

Edit: It would be nice to know the size of the batteries in each laptop being tested. 6, 8, 10 cells? That makes a difference, and the lack of information on the actually batteries being tested doesn't help these metrics. They could be testing a 10 cell battery in the Intel Laptop compared to a 6 cell on the AMD.

 

[-Fran-]

Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call).

BMW: i7-8700k is 12% faster than R7-1800X. The R7 has 33% moar cores. So

1.12 * 8/6 = 1.49.

GooseBerry: i7-8700k is 8% faster than R7-1800X. So

1.08 * 8/6 = 1.44.

So each core in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Uhm... So motorcycles that make the 1/4 mile 2 second faster are actually 45% faster than cars because they have 2 wheels instead of 4? 😵

What kind of logic is that? XD

You just do the % against the scores if you want an overall measurement. What you did there is just an insult to everyone's intelligence:

345.3/303.02 = 1.1395... ~ 14% *overall* slower in blender in that particular test.

Don't you understand what he is trying to explain?..Intel 6 core CPU is 12% faster than AMD 8 core CPU that has 33% more cores (2 more cores 4 more threads)...

Meaning that EACH CORE in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Overall performance a six core CoffeLake chip still 12% percent faster than an 8 core AMD chip...


Let me explain it better for you:
6 horses from Intel have 12% more horsepower than 8 horses from AMD. Meaning that a single horse from Intel has 44-49% more horsepower than a single horse from AMD..So what's AMD response?.. Since horse per horse they can't beat intel, what are they doing? They keep adding more weaker horses. Got it? 😉

So disregarding scaling to cores and intrinsics of the software is perfectly fine? That number has a big margin of error and he's quoting it as a perfectly safe number where there are other reviews showing the exact opposite picture.

If you want to get "single core" performance, get a test for it. Do not extrapolate from a MT test. Like I said, it's an insult to everyone; I got perfectly well what he tried to paint, hence my response.

 

[goldstone77]

1800X is still cheaper to buy now than the 8700k....with the recent discounting....or the 1700x.

At time of writing this I can purchase the i7-8700k by 409€ whereas the R7-1800X is 417€.

https://www.pccomponentes.com/amd-ryzen-7-1800x-40ghz-sin-disipador
https://www.pccomponentes.com/intel-core-i7-8700k-37ghz-box

So the R7-1800X is both slower and more expensive.

the single core advantage will decrease next year.

Maybe but Intel will continue being about 30% ahead in rendering applications and about 50% on games.

Price will vary in different countries, and performance varies depending on the application. I find the 8700 lock CPU to be a much better buy over the 8700K when you calculate added cost for cooling $100+. Same goes for the 1700 vs 1800X. Similar performance doesn't justify added cost if cost is a variable.

 

[adamsleath]

yup, 8700 non k is the deal

"Maybe but Intel will continue being about 30% ahead in rendering applications and about 50% on games."

50% faster on games....lol - nowhere near it...maybe on some cherry picked examples

cherry picked. i'll stick with my broad range of tests for a realistic picture.

across the broad range of applications these figures are quite simply not true.
----------------------------------------------------------------------
30% ahead in single threaded apps (averaged)
10% behind on multithreaded apps. (averaged)

as distinct from cherry picked.

intel 8 core (7820X) 15% more multicore performance for 2x the price
---------
cherry picking tests has been the confuser the whole time.

 

[juanrga]

Instead of doing real world testing?..... I can see the difference in battery life if you were going to use a monitor, and leave your laptop unplugged sitting next to an outlet... But I get what the article is pointing out even if I don't find this information useful for the average everyday user that would simply plug their laptop into the wall outlet it's sitting next too.

While I wanted to get real-world battery-life testing out of the way first, I can certainly appreciate the elegance of leveling the playing field that way. Now we have.

Edit: It would be nice to know the size of the batteries in each laptop being tested. 6, 8, 10 cells? That makes a difference, and the lack of information on the actually batteries being tested doesn't help these metrics. They could be testing a 10 cell battery in the Intel Laptop compared to a 6 cell on the AMD.

He is doing real-world testing. But he is not testing the laptops, but the chips inside. That is the reason why he had to isolate the laptop screen size effect by using the same external monitor for both laptops.

He gives both direct measurement numbers and normalized battery numbers. The number of cells is irrelevant.

Spoiler

1800X is still cheaper to buy now than the 8700k....with the recent discounting....or the 1700x.

At time of writing this I can purchase the i7-8700k by 409€ whereas the R7-1800X is 417€.

https://www.pccomponentes.com/amd-ryzen-7-1800x-40ghz-sin-disipador
https://www.pccomponentes.com/intel-core-i7-8700k-37ghz-box

So the R7-1800X is both slower and more expensive.

the single core advantage will decrease next year.

Maybe but Intel will continue being about 30% ahead in rendering applications and about 50% on games.

Price will vary in different countries, and performance varies depending on the application. I find the 8700 lock CPU to be a much better buy over the 8700K when you calculate added cost for cooling $100+. Same goes for the 1700 vs 1800X. Similar performance doesn't justify added cost if cost is a variable.

As mentioned in the CoffeeLake thread, Newegg is milking users. I can purchase the 8700k by less money than the 1800X.

I know that performance varies with the application. That is why I gave two different percentages for two different user cases.

 

[juanrga]

Spoiler

Single core performance gap is actually in the range 20%-30% but 1800x offers 30% more cores.
On multithreaded scenarios 1800X still performs better than 8700k (including blender, though this ons is a close call).

BMW: i7-8700k is 12% faster than R7-1800X. The R7 has 33% moar cores. So

1.12 * 8/6 = 1.49.

GooseBerry: i7-8700k is 8% faster than R7-1800X. So

1.08 * 8/6 = 1.44.

So each core in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Uhm... So motorcycles that make the 1/4 mile 2 second faster are actually 45% faster than cars because they have 2 wheels instead of 4? 😵

What kind of logic is that? XD

You just do the % against the scores if you want an overall measurement. What you did there is just an insult to everyone's intelligence:

345.3/303.02 = 1.1395... ~ 14% *overall* slower in blender in that particular test.

Don't you understand what he is trying to explain?..Intel 6 core CPU is 12% faster than AMD 8 core CPU that has 33% more cores (2 more cores 4 more threads)...

Meaning that EACH CORE in the CoffeLake chip is 44--49% faster than each Zen core in Blender.

Overall performance a six core CoffeLake chip still 12% percent faster than an 8 core AMD chip...


Let me explain it better for you:
6 horses from Intel have 12% more horsepower than 8 horses from AMD. Meaning that a single horse from Intel has 44-49% more horsepower than a single horse from AMD..So what's AMD response?.. Since horse per horse they can't beat intel, what are they doing? They keep adding more weaker horses. Got it? 😉

So disregarding scaling to cores and intrinsics of the software is perfectly fine? That number has a big margin of error and he's quoting it as a perfectly safe number where there are other reviews showing the exact opposite picture.

If you want to get "single core" performance, get a test for it. Do not extrapolate from a MT test. Like I said, it's an insult to everyone; I got perfectly well what he tried to paint, hence my response.

Wrong. During a single-thread measurement the core is working at single-core turbo, so the performance of the core during a single-thread measurement is not the performance of the core during a multithread measurement, when single-core turbo does not apply.

The performance equation for multithread is well-known: it is the product of the performance of each core and the total number of cores used. Blender scales almost linearly well above eight cores, so the number of cores used by Blender is the total number of cores in the chip: 6C for 8700k and 8C for 1800X.

8C Zen is 14% slower than 6C CFL in Blender workloads because each Zen core is ~50% slower than one CFL core.

8/6 * 1.14 = 1.52

 

[juanrga]

50% faster on games....lol - nowhere near it...maybe on some cherry picked examples

The 8700k is 60% faster in Hitman and 55% faster in Ashes. Not many time ago Ashes was considered the prototype of future games by many. It is a game sponsored by AMD and it was one of the first on receiving RyZen optimizations

Is it a cherry picked game now?

across the broad range of applications these figures are quite simply not true.
----------------------------------------------------------------------
30% ahead in single threaded apps (averaged)
10% behind on multithreaded apps. (averaged)

as distinct from cherry picked.

10% behind on multithreaded apps. (averaged)?

From Arstechnica review of CoffeeLake:

Even though it has two fewer cores than the Ryzen 1800X (a CPU that costs a hefty £437), the 8700K comes in faster in many production workloads. It's four seconds quicker in Blender at stock, and 11 seconds quicker when overclocked. It's faster at Handbrake video encoding too, and miles ahead in 7-Zip's synthetic benchmark, which tends to favour clock speed even in multithreaded mode.

It's only in PovRay and Cinebench that 1800X comes out on top—and only then by a small amount.

So if CoffeeLake only looses (by a small amount) in Povray and CB, but wins by a large margin in "many production workloads", it is evident that the average multithread performance of CoffeeLake cannot be 10% behind...

 

[adamsleath]

"The 8700k is 60% faster in Hitman and 55% faster in Ashes"
you are kidding yourself with these figures. there are so many tests which do not agree with this bs.

the same goes for rendering. you're only fooling yourself there.
----

not only cherry picked. but also bogus.

in the same class as the 'glued together cores' spin

your stated logic is based on non representational and questionable results. results which are atypical when compared to the majority of results i have viewed.