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

[Destination]

They actually thought these names were intelligent? What the hell are they thinking? I do hope this is the last we see of such ridiculous naming conventions. Clearly rattled.

You mean the 108976876455000XSE?

somewhat odd, to have absolutely no clue what their conventions are now across the various levels.

9900KSSKXX and then they get very confused over what comes next.

 

[Panos]

They actually thought these names were intelligent? What the hell are they thinking? I do hope this is the last we see of such ridiculous naming conventions. Clearly rattled.

I believe Intel considers using "new" names to cut down prices a better solution, than reduce prices on previous gen and for the marketing team to justify their salaries since nothing new is released.

 

[Rroff]

You mean the 108976876455000XSE?

somewhat odd, to have absolutely no clue what their conventions are now across the various levels.

9900KSSKXX and then they get very confused over what comes next.

I think you've got a few digits missing.

 

[pmc25]

Those chips are running 100C at just 4.3Ghz and 360mm radiator to themselves. Trying to do 4.5Ghz let alone 5Ghz would be a challenge.

Also we see already that the 7920X is left behind already by the 3900X and in many cases the 9980XE also.

Here is a nice review using also 9980XE & 7920X.
https://hothardware.com/reviews/amd-ryzen-9-3900x-and-ryzen-7-3700x-zen-2-review?page=5

Now add 4 more cores to see where the 3950X would be against them. Even the 10980XE looks grim.

This is another warmed over re-spin of the same design, more or less. Tiny evolution of the other X299 products.

Best bins got over 5Ghz on those, even above 12 cores ... but at nuclear power draw. So should this.

My whole point is that even at nose bleed clocks, it's not going to beat Zen 2 or 9900K in gaming, and will get demolished by Zen 2 in heavily threaded stuff - even by lower core count Zen 2 chips (as they scale far better).

 

[Panos]

This is another warmed over re-spin of the same design, more or less. Tiny evolution of the other X299 products.

Best bins got over 5Ghz on those, even above 12 cores ... but at nuclear power draw. So should this.

My whole point is that even at nose bleed clocks, it's not going to beat Zen 2 or 9900K in gaming, and will get demolished by Zen 2 in heavily threaded stuff - even by lower core count Zen 2 chips (as they scale far better).

Yes. Level1Tech did a great video few days ago about the X299 overall and why Intel should let it die. Mainly because Intel has better stuff to offer with the 3647 platform against 2066 which might make it remotely appealing over someone getting having to pick Threadripper or an Intel HEDT CPU. Partially AM4 X570 is better platform than the X299 if you plan to use M.2 drives, putting CPU aside.

And before someone says "ooh AMD fanboy", Intel on 3647 has better products than 2066. That's a fact.

 

[pmc25]

Yes. Level1Tech did a great video few days ago about the X299 overall and why Intel should let it die. Mainly because Intel has better stuff to offer with the 3647 platform against 2066 which might make it remotely appealing over someone getting having to pick Threadripper or an Intel HEDT CPU. Partially AM4 X570 is better platform than the X299 if you plan to use M.2 drives, putting CPU aside.

And before someone says "ooh AMD fanboy", Intel on 3647 has better products than 2066. That's a fact.

Problem is that 3647 is barely available and both chips and mobos are INSANELY expensive. Plus the power draw. Plus much lower performance / scaling than any Zen 2 TR4 product will be.

3647 was also launched over a year before it actually got stock. Intel realised that even when facing TR4 1xxx, they were barely going to sell any, let alone 2xxx or soon to launch 3xxx.

I'd be very surprised if the 24 core 3xxx TR4 chip doesn't absolutely smash the 28 core 3647 chip - at 1/3rd the cost, and 1/3rd the power draw.

3647 was the kind of product they could have launched and made a killing on before Zen launched. But they were too busy gouging the hell out of people on LGA 2xxx socket.

*IF* Intel can get their 7nm out by mid 2021, and it's not a total bust wrt clocks, power and yields, maybe they can fight back then in desktop / HEDT / server etc. But the gap between AMD and them will only get larger until they have a workable 7nm process, or their all new design comes in 2022 or 2023. But even then ... AMD will transition to 7nm EUV next year, 5nm EUV in 2021, etc etc. It's been clear for at least a year that we're never going to see 10nm desktop enthusiast / HEDT Intel products ... I also have a hard time accepting that Intel's consumer graphics cards could launch next year on their own 10nm, given how crippled it is.

 

[Panos]

Problem is that 3647 is barely available and both chips and mobos are INSANELY expensive. Plus the power draw. Plus much lower performance / scaling than any Zen 2 TR4 product will be.

3647 was also launched over a year before it actually got stock. Intel realised that even when facing TR4 1xxx, they were barely going to sell any, let alone 2xxx or soon to launch 3xxx.

I'd be very surprised if the 24 core 3xxx TR4 chip doesn't absolutely smash the 28 core 3647 chip - at 1/3rd the cost, and 1/3rd the power draw.

3647 was the kind of product they could have launched and made a killing on before Zen launched. But they were too busy gouging the hell out of people on LGA 2xxx socket.

*IF* Intel can get their 7nm out by mid 2021, and it's not a total bust wrt clocks, power and yields, maybe they can fight back then in desktop / HEDT / server etc. But the gap between AMD and them will only get larger until they have a workable 7nm process, or their all new design comes in 2022 or 2023. But even then ... AMD will transition to 7nm EUV next year, 5nm EUV in 2021, etc etc. It's been clear for at least a year that we're never going to see 10nm desktop enthusiast / HEDT Intel products ... I also have a hard time accepting that Intel's consumer graphics cards could launch next year on their own 10nm, given how crippled it is.

3647 boards are expensive because there are only 2 with no competition. Intel operates with 300% profit margins on those chips, regardless how hard are to manufacture been huge and monolithic.
However if Intel just drops the ball, admit defeat, reduce prices and wait until having new architecture is ready, could understand and accept.
What we see atm, is trying to make fool of themselves and everyone around, promoting things like CascadeLake-X as the Messiah of the HEDT market. And Intel looks more stupid than AMD trying to sell $1000 the 9000 series bulldozers. Because Intel has products who can reduce their profits and not lose on all fronts so big, AMD was desperate for money without alternative product to offer. That's big difference.

Yes sure TR4 has killed the Intel HEDT market, TR 2000 hammered the coffin and the upcoming TR 3000 will be shoving the soil to bury is deep until a replacement comes around 2023. But is different thing to drop the ball than been ridiculed.

Nobody argues that the smallest of the TR 3000 (the 24 core chip) is faster even than the 28 core monstrosity. Already we see it 7402P benchmarks even at max boost 3.5Ghz and high latency ECC ram on server board, is faster than anything Intel had to offer in 2066 & 3647 sockets. And offers more as platform than the X299. Especially when comes to things like the M.2s aren't going through DMI (effectively all sharing 4 pci 3.0 lanes on X299) but directly to the CPU on PCIe 4.0. And a simple 2xMP600 on raid in 7402P look ridiculous fast also.

But come on. That's Intel we talk about that cannot go against a company which is 1/12th it's size, by simply reducing prices to try maintain some market share?
Already in mainstream is getting annihilated on anything but on the specific condition of gaming at 1080p/720p with RTX2080Ti.
Anything higher resolution or lesser graphic card there is no difference between the 3700X and the 9900K. Let alone the 3800X or 3900X.

 

[Panos]

Update this one came out
https://www.tomshardware.com/news/amd-ryzen-9-3950x-overclocking-watercooling-4.3-ghz,40555.html

That puts the 3950X is around 8% faster than the 18 core 9980XE.

 

[sideways14a]

Thats just bonkers.

 

[Rroff]

I also have a hard time accepting that Intel's consumer graphics cards could launch next year on their own 10nm, given how crippled it is.

Thing is their 10nm node is basically a hybrid - some areas important to CPU design are quite aggressive (which is one of the factors in the problems they've been having) while other areas relatively relaxed. For a GPU they can probably get away with a more tolerant design - but that will mean bringing something to market that is closer to a refined 12nm than the 7nm EUV that AMD/nVidia are ramping on.

 

[Grim5]

Gpus don't need to hit 4ghz+ or handle 1.3 to 1.5v. If anything, gpu are easier to manufacture (ignoring the R&D)

 

[Maldoror]

3647 boards are expensive because there are only 2 with no competition. Intel operates with 300% profit margins on those chips, regardless how hard are to manufacture been huge and monolithic.
However if Intel just drops the ball, admit defeat, reduce prices and wait until having new architecture is ready, could understand and accept.
What we see atm, is trying to make fool of themselves and everyone around, promoting things like CascadeLake-X as the Messiah of the HEDT market. And Intel looks more stupid than AMD trying to sell $1000 the 9000 series bulldozers. Because Intel has products who can reduce their profits and not lose on all fronts so big, AMD was desperate for money without alternative product to offer. That's big difference.

Yes sure TR4 has killed the Intel HEDT market, TR 2000 hammered the coffin and the upcoming TR 3000 will be shoving the soil to bury is deep until a replacement comes around 2023. But is different thing to drop the ball than been ridiculed.

Nobody argues that the smallest of the TR 3000 (the 24 core chip) is faster even than the 28 core monstrosity. Already we see it 7402P benchmarks even at max boost 3.5Ghz and high latency ECC ram on server board, is faster than anything Intel had to offer in 2066 & 3647 sockets. And offers more as platform than the X299. Especially when comes to things like the M.2s aren't going through DMI (effectively all sharing 4 pci 3.0 lanes on X299) but directly to the CPU on PCIe 4.0. And a simple 2xMP600 on raid in 7402P look ridiculous fast also.

But come on. That's Intel we talk about that cannot go against a company which is 1/12th it's size, by simply reducing prices to try maintain some market share?
Already in mainstream is getting annihilated on anything but on the specific condition of gaming at 1080p/720p with RTX2080Ti.
Anything higher resolution or lesser graphic card there is no difference between the 3700X and the 9900K. Let alone the 3800X or 3900X.

But Bullzdozer was never the best. It was never good at all.

Much of the HEDT market is people who want to buy the best: who need PCI-E lanes for multiple high-end GPUS, who need cores for edge cases, extreme gaming/streaming, benchmarking, etc. A 9900K/KS will still see gains over anything AMD has even at 1440p/165hz, btw - a target at which a high end 2080Ti or Titan can shift the bottleneck to the CPU.

There is a very significant chance that the 10980XE will be the go-to chip for the likes of 3DMark, gaming benchmarks and the chip that the GamersNexus/Jayz/whatever of this world will use to extreme cool and try and have their overclocking battles. It'll have more cores than the 9900KS, will reach similar clockspeeds under expensive cooling (e.g. chillers) and the mesh frequency can be overclocked considerably to catch up to ringbus. Threadripper 3 will win on Cinebench multicore and Blender but is unlikely to have the latency to compete against the 10980XE in the gaming world.

As someone whose had a 7980XE since the day it came out, it is disappointing that the 'top chip' nearly 3 years later is basically the same chip at half price. Sure I'll probably buy one just to see if it's binned better but I'm buying the same chip twice. While they've made massive strides in price/peformance in the mid-range and low-range, I still don't anticipate anything from AMD changing the pecking order at the halo end. When Kingpin turns up to overclock something in an article, he'll probably use a 10980XE and nothing from AMD will be able to compete with the result. This is overclockers uk, after all

 

[Panos]

But Bullzdozer was never the best. It was never good at all.

Much of the HEDT market is people who want to buy the best: who need PCI-E lanes for multiple high-end GPUS, who need cores for edge cases, extreme gaming/streaming, benchmarking, etc. A 9900K/KS will still see gains over anything AMD has even at 1440p/165hz, btw - a target at which a high end 2080Ti or Titan can shift the bottleneck to the CPU.

There is a very significant chance that the 10980XE will be the go-to chip for the likes of 3DMark, gaming benchmarks and the chip that the GamersNexus/Jayz/whatever of this world will use to extreme cool and try and have their overclocking battles. It'll have more cores than the 9900KS, will reach similar clockspeeds under expensive cooling (e.g. chillers) and the mesh frequency can be overclocked considerably to catch up to ringbus. Threadripper 3 will win on Cinebench multicore and Blender but is unlikely to have the latency to compete against the 10980XE in the gaming world.

As someone whose had a 7980XE since the day it came out, it is disappointing that the 'top chip' nearly 3 years later is basically the same chip at half price. Sure I'll probably buy one just to see if it's binned better but I'm buying the same chip twice. While they've made massive strides in price/peformance in the mid-range and low-range, I still don't anticipate anything from AMD changing the pecking order at the halo end. When Kingpin turns up to overclock something in an article, he'll probably use a 10980XE and nothing from AMD will be able to compete with the result. This is overclockers uk, after all

You make no sense.

 

[pmc25]

Gpus don't need to hit 4ghz+ or handle 1.3 to 1.5v. If anything, gpu are easier to manufacture (ignoring the R&D)

They don't hit the same clock speeds as CPUs, because they have a very wide pipeline.

Intel 10nm has ~1Ghz clock regression from their 14nm for CPUs for the higher clock parts. Imagine if that was the case for TSMC and Samsung from 14nm to 7nm. Clocks went up, rather than regressing.

I don't see how it's going to work for them. Even if they can hit clock targets, yields are going to be horrific. The tiny laptop chips they're making are still apparently sub 40% yield, even after being in production for a year now. Imagine what bigger GPUs will be? Yes it's highly likely they go chiplets ... but after the first gen - 2nd or 3rd.

 

[Rroff]

They don't hit the same clock speeds as CPUs, because they have a very wide pipeline.

Intel 10nm has ~1Ghz clock regression from their 14nm for CPUs for the higher clock parts. Imagine if that was the case for TSMC and Samsung from 14nm to 7nm. Clocks went up, rather than regressing.

I don't see how it's going to work for them. Even if they can hit clock targets, yields are going to be horrific. The tiny laptop chips they're making are still apparently sub 40% yield, even after being in production for a year now. Imagine what bigger GPUs will be? Yes it's highly likely they go chiplets ... but after the first gen - 2nd or 3rd.

But why are the yeilds poor? there are a number of reasons why that can happen and doesn't automatically mean that a bigger die for a different product will be a failure.

 

[4K8KW10]

REVIEW – INTEL CORE I9 10980XE – CASCADE LAKE X

Some results:

 

[DragonQ]

By "some results", you mean "the best results". I'm also not sure what the point of those comparisons really is, given it's an HEDT chip. Shouldn't it be compared against the i9-9980XE and Threadripper 2950X (in terms of product class), or i9-9900X and Threadripper 2950X/2970WX (in terms of price)?

The biggest question is how it performs against the equivalent Threadripper 3 part, but current rumours indicate there won't be a 16c/32t chip. So basically it'll have to be compared to the R9 3950X (with the caveats that it's not an HEDT chip) or a probably similarly priced R9 3960X.

 

[LePhuronn]

There is a very significant chance that the 10980XE will be the go-to chip for the likes of 3DMark, gaming benchmarks and the chip that the GamersNexus/Jayz/whatever of this world will use to extreme cool and try and have their overclocking battles.

So let them enjoy their pointless numbers and peen-swinging, the rest of us will get Threadripper and do some real work. Or 3950X and just point and laugh at Intel's HEDT getting battered by a lower-tier chip.

or a probably similarly priced R9 3960X.

3960X is Threadripper, probably 16 core that nobody thinks is happening.

AMD have said Threadripper 3 is launching with 24 cores, but the existing (and totally screwed up) nomenclature feels off to have a CPU with 8 more cores, double the PCIe lanes and 2 extra memory channels only 10 numbers higher than the 3950X. Not a big enough gap. Also I'm not entirely sure we'd see a 48 core Threadripper now simply because there is no need for AMD to do so (same with 64 cores). So I'm thinking we'll get 3970X launching with 24 cores, 3960X with 16 cores (both on November 5th) and the 3990X is 32 cores at CES.

I'm also half-expecting Threadripper to be WX too, just to further differentiate from desktop Ryzen, but that does mean we could do 24 cores as the 3960WX.

Roll on November 5th!

 

[DragonQ]

3960X is Threadripper, probably 16 core that nobody thinks is happening.

AMD have said Threadripper 3 is launching with 24 cores, but the existing (and totally screwed up) nomenclature feels off to have a CPU with 8 more cores, double the PCIe lanes and 2 extra memory channels only 10 numbers higher than the 3950X. Not a big enough gap. Also I'm not entirely sure we'd see a 48 core Threadripper now simply because there is no need for AMD to do so (same with 64 cores). So I'm thinking we'll get 3970X launching with 24 cores, 3960X with 16 cores (both on November 5th) and the 3990X is 32 cores at CES.

I'm also half-expecting Threadripper to be WX too, just to further differentiate from desktop Ryzen, but that does mean we could do 24 cores as the 3960WX.

Roll on November 5th!

I meant whichever Threadripper 3 chip will be a similar retail price, which I assume would be the 24c/48t version but we don't know yet.

 

[humbug]

But Bullzdozer was never the best. It was never good at all.

Much of the HEDT market is people who want to buy the best: who need PCI-E lanes for multiple high-end GPUS, who need cores for edge cases, extreme gaming/streaming, benchmarking, etc. A 9900K/KS will still see gains over anything AMD has even at 1440p/165hz, btw - a target at which a high end 2080Ti or Titan can shift the bottleneck to the CPU.

There is a very significant chance that the 10980XE will be the go-to chip for the likes of 3DMark, gaming benchmarks and the chip that the GamersNexus/Jayz/whatever of this world will use to extreme cool and try and have their overclocking battles. It'll have more cores than the 9900KS, will reach similar clockspeeds under expensive cooling (e.g. chillers) and the mesh frequency can be overclocked considerably to catch up to ringbus. Threadripper 3 will win on Cinebench multicore and Blender but is unlikely to have the latency to compete against the 10980XE in the gaming world.

As someone whose had a 7980XE since the day it came out, it is disappointing that the 'top chip' nearly 3 years later is basically the same chip at half price. Sure I'll probably buy one just to see if it's binned better but I'm buying the same chip twice. While they've made massive strides in price/peformance in the mid-range and low-range, I still don't anticipate anything from AMD changing the pecking order at the halo end. When Kingpin turns up to overclock something in an article, he'll probably use a 10980XE and nothing from AMD will be able to compete with the result. This is overclockers uk, after all

No it wont, Skylake X suffers from the same problem in games that Ryzen / Threadripper does, SkyLake X has a Mesh architecture, the Ring Bus in Coffeelake doesn't work beyond 10 cores.

Skylake X Mesh is actually far worse than AMD's Infinity Fabric, the gaming difference in IPC is about minus 15% Mesh vs Ring Bus, Minus 10% vs Zen + and minus 25% vs Zen 2.

Gaming by IPC, 'Not clock speed' where Coffeelake has the advantage and makes it the best Gaming CPU. By a small margin.


#1 Zen 2
#2 Coffeelake
#3 Zen +
#4 Zen
#5 Skylake X