Intel Cannonlake, Cascade Lake, Ice Lake, Tiger Lake & Sapphire Rapid thread

[Nutella33]

I guess the 10980XE will beat the 2950x in most things so will wait to see what the new TR has.

Well, you'd hope so as its £300 more expensive, has 2C4T extra and is coming out a year later than the 2950x. And is Intels top offering against a midrange Threadripper part. The comparison for 10980XE should be vs the 2970wx since they share the same price bracket.

I wonder if AMD delayed the release partially due to this new Win10 schedular? seems advantagious to both companies to be reviewed on their new products with this fix in place.

 

[Orange Nexus]

Will it need new motherboard and chipset?

X299 will work but will only have 44 lanes and PCIe 3.OLD.

 

[RavenXXX2]

TR3 24 core will decimate this knee jerk release.

 

[Destination]

X299 will work but will only have 44 lanes and PCIe 3.OLD.

have they a new mobo range with more lanes, or is that it?

 

[Orange Nexus]

have they a new mobo range with more lanes, or is that it?

X299X or something with 48 lanes.

 

[Destination]

thx for info

 

[B1gbeard]

Well, you'd hope so as its £300 more expensive, has 2C4T extra and is coming out a year later than the 2950x. And is Intels top offering against a midrange Threadripper part. The comparison for 10980XE should be vs the 2970wx since they share the same price bracket.

I wonder if AMD delayed the release partially due to this new Win10 schedular? seems advantagious to both companies to be reviewed on their new products with this fix in place.

Both companies release products a year later that are still slower than the other companies previous CPU. It's just the market, not unusual.

There are programs, games as well, that don't play so well with the IF on the 2970 and 2990 so for some tasks it'll beat even the 2970. 2950x is also 829 on here so not £300. Silly fanboy mine beats yours is beside the point though, it's good for the market either way. Looking forward to see what the 3970wx or whatever it is called does.

 

[humbug]

Looks like AMD are controlling the HEDT space now, what a turnaround.

The thing is, while its a good try Intel but you're upagainst 32 Zen 2 cores with your 18 Skylake-X cores.

 

[MagicBoy]

Now that Intel are ramping up the cores, it's in Microsoft's interest to get their scheduler sorted out. I wonder if they'll just relent and put the server scheduler onto desktop?

What's the source that the Microsoft desktop OS scheduler is different to the related server OS?

 

[Nutella33]

Both companies release products a year later that are still slower than the other companies previous CPU. It's just the market, not unusual.

Its unusual when the top range coming a year later is only comparable to the midrange of the previous year of the competitor. Intel are in the position of AMD GPU's vs Nvidia's product stack at this point in this space. That is unusual.

2950x is also 829 on here so not £300.

Its available at £720~ from several of AMD's listed suppliers in the UK.

Silly fanboy mine beats yours is beside the point though, it's good for the market either way. Looking forward to see what the 3970wx or whatever it is called does.

Pointing out your comparisons were off isnt fanboyism with upmanship. TR3 is looking great, but when looking at this release, its worth noting though that these parts actually only really fit in against TR2 as it stands. Which is significant for intel in modern times.

 

[B1gbeard]

Its unusual when the top range coming a year later is only comparable to the midrange of the previous year of the competitor. Intel are in the position of AMD GPU's vs Nvidia's product stack at this point in this space. That is unusual.

The 1800x often lost to a 7600k and 6700k when it was released for example so no, not unusual. Core for core Intel CPU's are still quicker if reviews are to be believed so it doesn't seem anything like the GPU situation. Throw cores at it and some programs benefit.

Pointing out your comparisons were off isnt fanboyism with upmanship. TR3 is looking great, but when looking at this release, its worth noting though that these parts actually only really fit in against TR2 as it stands. Which is significant for intel in modern times.

If you had of pointed it out then that would have been great, you didn't. Whether the 2950x or 2970wx the 10980xe will beat both in many applications because single core speed is still generally most relevant and the 9980xe did it anyway. The 2970wx when it was bad was very bad as was the 2990wx, but when it was good, like rendering for example, it was very good. That's just life and not a criticism of TR. I think they are marvellous and hope TR gen 2 is faster than Intel to keep pushing things forward.

 

[humbug]

The 1800x often lost to a 7600k and 6700k when it was released for example so no, not unusual. Core for core Intel CPU's are still quicker if reviews are to be believed so it doesn't seem anything like the GPU situation. Throw cores at it and some programs benefit.



If you had of pointed it out then that would have been great, you didn't. Whether the 2950x or 2970wx the 10980xe will beat both in many applications because single core speed is still generally most relevant and the 9980xe did it anyway. The 2970wx when it was bad was very bad as was the 2990wx, but when it was good, like rendering for example, it was very good. That's just life and not a criticism of TR. I think they are marvellous and hope TR gen 2 is faster than Intel to keep pushing things forward.

No one cares about last generation Threadripper, that's not what Intel's 10 series HEDT line will have to contend with.

The IPC of Zen 2 is 10% higher than coffeelake, the only reason Coffeelake is marginally faster in some applications is much higher clock speed.

A Ryzen 3900X is faster than a 9920X, an entry level 24 Core Zen 2 will utterly humiliate the top end 10980XE, or whatever this the latest refresh of Skylake-X is called.

 

[humbug]

Just to make the point.

CB R20 ST

Score 580: Intel Core i9 9900K at 5.45Ghz, RSR
Score 535: AMD Ryzen R9 3900X at 4.6Ghz, Shac

3900X: 535 @ 4.6Ghz
9900K: 580 @ 5.45Ghz

535 vs 580 = +8%
4.6Ghz vs 5.45Ghz = +18%

Difference in ST IPC = 10% to Zen 2

https://forums.overclockers.co.uk/threads/the-official-ocuk-cinebench-r20-benchmark-thread.18849380/

A core for core 4.35Ghz Threadripper 3000 will match a 4.8Ghz Skylake-X

 

[B1gbeard]

No one cares about last generation Threadripper, that's not what Intel's 10 series HEDT line will have to contend with.

The IPC of Zen 2 is 10% higher than coffeelake, the only reason Coffeelake is marginally faster in some applications is much higher clock speed.

A Ryzen 3900X is faster than a 9920X, an entry level 24 Core Zen 2 will utterly humiliate the top end 10980XE, or whatever this the latest refresh of Skylake-X is called.

Except its not unless you decide to only choose the results where it is. Even a quick glance at IPC reviews shows it wins some and loses some and loses more in games. HWU did a recent IPC video showing this. It's good, but it is not just 10% quicker. It is purely application based. As such I imagine TR2 vs 10980xe will be as well. A reasonable assumption unless you desperately need a team to win.

It's a constant puzzle how numbers can be looked at, the ones that we don't like are almost blocked out by the brain it seems, like a red filter blanks out the numbers that are not positive, take the ones you like and proclaim "AMD is 10% faster IPC"!

 

[humbug]

Except its not unless you decide to only choose the results where it is. Even a quick glance at IPC reviews shows it wins some and loses some and loses more in games. HWU did a recent IPC video showing this. It's good, but it is not just 10% quicker. It is purely application based. As such I imagine TR2 vs 10980xe will be as well. A reasonable assumption unless you desperately need a team to win.

It's a constant puzzle how numbers can be looked at, the ones that we don't like are almost blocked out by the brain it seems, like a red filter blanks out the numbers that are not positive, take the ones you like and proclaim "AMD is 10% faster IPC"!

These are HEDT CPU's, but if you want to talk about gaming the difference Zen 2 vs Coffeelake is small, again its down to clock speed, HWU in his 30+ game benchmarks did a 3900X +PBO vs 9900K @ 5Ghz and the overall difference was 5% to the 9900K.

Skylake-X, with its "Mesh" architecture, suffers from the same problem Zen does with its IF, only to a much worse extent, the IPC in games on Skylake-X is actually so far behind the Ring bus on Coffeelake, its actually less than Zen+ never mind Zen 2.

For Productivity, which is what HEDT CPU's are for, such as Ray Traced Rendering, like Maxcon Cinema 4D, Blender.. yes Zen 2 is faster, its faster in encoding work too, even in very inefficient usually Intel leaning Adobe software Zen 2 beats Coffeelake.

I'm sure tho, as Intel will attest you can find Intel beating Zen 2 in some applications, Windows Media Player, Google Chrome, Steam, Mine Sweeper.... just what you need a HEDT CPU for.

HWU: Skylake-X in gaming vs last gen Zen+

 

[Grim5]

Will it need new motherboard and chipset?

nope

 

[Panos]

Except its not unless you decide to only choose the results where it is. Even a quick glance at IPC reviews shows it wins some and loses some and loses more in games. HWU did a recent IPC video showing this. It's good, but it is not just 10% quicker. It is purely application based. As such I imagine TR2 vs 10980xe will be as well. A reasonable assumption unless you desperately need a team to win.

It's a constant puzzle how numbers can be looked at, the ones that we don't like are almost blocked out by the brain it seems, like a red filter blanks out the numbers that are not positive, take the ones you like and proclaim "AMD is 10% faster IPC"!

I do not get what you talking about. IPC? Zen 2 (Ryzen 3000 & TR 3000) have higher IPC than Intel. Nobody disputes that except the blind.
Now by how much depends. Comparing CPUs at 3500Mhz lock, around 8%, at 4000Mhz the gap widens to 10%.
https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/
https://www.guru3d.com/articles_pages/amd_ryzen_7_3800x_review,9.html

If left to boost on their maximum settings
https://www.guru3d.com/articles_pages/amd_ryzen_7_3800x_review,10.html

So clearly Zen 2 has higher IPC than CoffeeLake based CPUs.


Now going to HEDT things are worse for Intel as @humbug said above. The new 10000XE series is just a renamed Skylake-X so flat out worst case scenario (3.5Ghz cap) the Zen 2 CPUs would have 10% higher IPC.

As for multi threading it all depends.
In HEDT platform, Intel uses MESH topology not RING since the Skylake-X came out (June 2017). Mesh is tad different than IF but the idea is the same. It allows better scaling on multi core CPUs for productivity but it's performance on games is appalling as proven numerous times and only suitable for monolithic CPUs. It doesn't scale with chiplets or multi-socket system (severs). Contrary to Infinity Fabric is which by miles superior operating within the same CCD, on various CCDs of a CPU and over multiple CPUs, all the same.

The mainstream Intel platform has managed to hold 2 1/2 years of AMD siege because of the Ring topology, but that has limitations. For sync-lock Ring & Core speed the CPU cannot be higher than 4 core. That's why 7700K & 7740K were exceptionally good CPUs for gaming (and the last ones) because of that sync-lock. Anything with more than 4 cores (6-8-10) must have Ring uncoupled to core otherwise instability and crashes occur.
Also Ring cannot go higher than 10 core CPU. After that the perf hit is so severe that ain't worth over Mesh or IF like topology.

And is one of the reasons why Intel cannot make a mainstream 10 core CPU let alone a 12 core or 16 core currently. That is why is also stuck atm as it won't have new architecture well until into 2022.
The 10nm CPUs if using Ring would be capped at 8-10 core, as it cannot go higher. And introducing Mesh on mainstream means it would be losing to AMD flat out even on gaming.
So basically Intel is sandbagging as much as possible atm behind the 5Ghz barrier and the archaic Ring topology, until they have a brand new ground up architecture, which is expected 2021-2022 the earliest.
Which by that time we are on Zen 5.

 

[sideways14a]

Zen2 does have an ipc advantage now, although its application dependant.
Older games tend to be more affected by latency and raw Mhz at the moment which is all thats keeping intels 9900k at the top of the gaming charts - by a tiny bit.
Cant see it getting any better for intel next year, zen2 made some big advances this year and all AMD needs to do is another decent latency drop and put another couple of hundred Mhz on the top end to completely wipe out intel at the top of any charts.

Intel are also going to struggle with price drops more than AMD, yes they are a big company and can .. to some extent take the hit but not for long. They simply cannot make chips as cheep as AMDs with the chiplet system and its ridiculously good use of die resources at the moment.
Yes 14nm is cheep as water now but making CPUs with chiplets means huge savings and by next year 7nm wont be that costly either even on the posh version AMD are supposedly moving to.

2020 could see the year intel looses from all metrics, its almost there now and has totally lost HEDT completely.

 

[4K8KW10]

I do not get what you talking about. IPC? Zen 2 (Ryzen 3000 & TR 3000) have higher IPC than Intel. Nobody disputes that except the blind.
Now by how much depends. Comparing CPUs at 3500Mhz lock, around 8%, at 4000Mhz the gap widens to 10%.
https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/
https://www.guru3d.com/articles_pages/amd_ryzen_7_3800x_review,9.html

If left to boost on their maximum settings
https://www.guru3d.com/articles_pages/amd_ryzen_7_3800x_review,10.html

So clearly Zen 2 has higher IPC than CoffeeLake based CPUs.


Now going to HEDT things are worse for Intel as @humbug said above. The new 10000XE series is just a renamed Skylake-X so flat out worst case scenario (3.5Ghz cap) the Zen 2 CPUs would have 10% higher IPC.

As for multi threading it all depends.
In HEDT platform, Intel uses MESH topology not RING since the Skylake-X came out (June 2017). Mesh is tad different than IF but the idea is the same. It allows better scaling on multi core CPUs for productivity but it's performance on games is appalling as proven numerous times and only suitable for monolithic CPUs. It doesn't scale with chiplets or multi-socket system (severs). Contrary to Infinity Fabric is which by miles superior operating within the same CCD, on various CCDs of a CPU and over multiple CPUs, all the same.

The mainstream Intel platform has managed to hold 2 1/2 years of AMD siege because of the Ring topology, but that has limitations. For sync-lock Ring & Core speed the CPU cannot be higher than 4 core. That's why 7700K & 7740K were exceptionally good CPUs for gaming (and the last ones) because of that sync-lock. Anything with more than 4 cores (6-8-10) must have Ring uncoupled to core otherwise instability and crashes occur.
Also Ring cannot go higher than 10 core CPU. After that the perf hit is so severe that ain't worth over Mesh or IF like topology.

And is one of the reasons why Intel cannot make a mainstream 10 core CPU let alone a 12 core or 16 core currently. That is why is also stuck atm as it won't have new architecture well until into 2022.
The 10nm CPUs if using Ring would be capped at 8-10 core, as it cannot go higher. And introducing Mesh on mainstream means it would be losing to AMD flat out even on gaming.
So basically Intel is sandbagging as much as possible atm behind the 5Ghz barrier and the archaic Ring topology, until they have a brand new ground up architecture, which is expected 2021-2022 the earliest.
Which by that time we are on Zen 5.

intel is betting everything on a potential new architecture but there is always a risk of it being a failure like the Bulldozer.
So, they must have at least a plan B just in case.

Also, they must be discussing the manufacturing arm split-off already.

 

[Rroff]

The 10nm CPUs if using Ring would be capped at 8-10 core, as it cannot go higher. And introducing Mesh on mainstream means it would be losing to AMD flat out even on gaming.
So basically Intel is sandbagging as much as possible atm behind the 5Ghz barrier and the archaic Ring topology, until they have a brand new ground up architecture, which is expected 2021-2022 the earliest.
Which by that time we are on Zen 5.

They aren't limited to 8-10 (for performance or technical reasons), some CPUs have upto 12 per ring and while it makes things a bit messy they can add additional rings with the ring router latency still around 20% below cross-CCX boundaries on existing AMD CPUs. While performance scaling does drop off as you add more cores and rings into the equation it isn't as dire as you are making out (where it becomes a significant penalty is cross-socket - but then multi-socket systems have never been a good idea for gaming).

http://www.qdpma.com/ServerSystems/MemLat2018.html

Intel and MS are working on that issue in terms of gaming performance with various techniques like "favoured cores" to overcome the needs to balance multi-thread and gaming type workloads.