Back to Haswell sizes.
- Haswell 4+2: 177 mm²
- Skylake/Kaby Lake 4+2: 126 mm²
- Coffee Lake 6+2: 149 mm²
- Coffee Lake 8+2, all else being equal to 6+2's, and 2 cores having area ~25 mm²: 174 mm²
See also: https://en.wikichip.org/wiki/intel/microarchitectures/coffee_lake#Historical_Trend an evolution of the client processors
People who are getting worked up over more cores seem to underestimate the difficulty of building a game engine that can properly use an 8-core processor, let alone something that can effectively use >8 cores.
It might be difficult, but it's definitely still happening. Enormous progress has been made on this front over the past however many years since the current gen consoles launched. All the major 3D engines, Unreal Engine 4, Frostbite 3, CryEngine, AnvilNext, Id Tech6 etcetera, and even some from smaller devs like the 4A Engine used in the Metro series, can effectively use at least six threads. AnvilNext has been shown to be able to scale on a deca core CPU even. This development was done out of necessity, to be able to milk as much juice as they could out of the weak console CPUs, and to satiate the market's increasing desire for bigger and more realistic games. There's no stopping the progress being made on this front, as the future of gaming depends greatly on programmers' capabilities to get as much out of the hardware as is possible.
A lot of developers still haven't been able to break the four thread barrier, like CDPR's Red Engine 3. That might not have been a limitation for the Witcher 3, but it very well might be for their next game Cyberpunk 2077 which will likely be an even bigger game with more simulation. If CDPR can't get their engine to scale to more than 4 threads, then their vision for Cyberpunk 2077 will have to be compromised, much like how the Witcher 3 was severely downgraded.
Ubisoft by comparison has done more on this front than perhaps any developer, AND IT SHOWS. AC Origins will be an immense game, featuring practically an entire country with diverse environments from massive cities to small villages to deserts, to huge Pyramids and underground tombs. And all of this will be completely seamless with no loading screens. You can't have that kind of scope and detail without an engine that can effectively and properly utilize a CPU and memory. Only now is such a thing even possible, due to the advancements they've made in game technology, and proper CPU scaling is crucial to their success no doubt.
But it costs more to produce smaller process wafers, so Haswell size die at 14nm is more expensive.
Though if they manage to build 8 cores and an IGP 20mm² smaller than Ryzen that doesn't have IGP it does seem like Intel process size advantage is true.
That is not how this all works. AMD offered, the market responded and bought big time. We wanted them so bad that sales started going towards AMD instead of Intel.And again, we are getting 8 core CPU's whether we need them or not, whether games can use them or not and whether we want them or not. We are getting them because AMD straight up WENT THERE and now Intel's hand is forced.
Current 14nm process is very mature, costs should be way down compared to Skylake era (higher yields, R&D payed for etc), and even if this is not the case we have repeatedly stated in this forum that customers pay the high margins Intel demands not for the silicon size but for the performance it brings to the table. We can't have it both ways - very high margins because best performance, but lower performance because die cost is seriously affecting margins. If die size cost becomes so important as to dictate the number of cores in the mainstream products, then hold on to your hats - we'll have to explain again why Intel chooses to use such a big iGPU area for their desktop SKUs. (again, the argument is we buy performance, not silicon die area)But it costs more to produce smaller process wafers, so Haswell size die at 14nm is more expensive.
Thanks for the post! Do you have any links to AnvilNext CPU scaling? I would love to see how well it can take advantage of all those cores.
A lot of developers still haven't been able to break the four thread barrier, like CDPR's Red Engine 3. That might not have been a limitation for the Witcher 3, but it very well might be for their next game Cyberpunk 2077 which will likely be an even bigger game with more simulation. If CDPR can't get their engine to scale to more than 4 threads, then their vision for Cyberpunk 2077 will have to be compromised, much like how the Witcher 3 was severely downgraded.
What 4 thread barrier? Witcher 3 gets nice uplift moving from 4/8 to 6/12 core/threads:
https://youtu.be/uwUEVEbZxI4?t=3m48s
That is not how this all works. AMD offered, the market responded and bought big time. We wanted them so bad that sales started going towards AMD instead of Intel.
This is the market's choice, not AMD's or Intel's.
There are no IPC gains in coffee Lake and the L3 isn't any higher.What you are seeing in the above link are the effects of IPC improvements plus extra L3 cache, and not scaling past 4/8. Red Engine 3 can actually use HT though, but only in the CPU bound areas of the game like Novigrad where it can make a significant difference.
PCGH.de did a test on the CPU scaling after the game launched.
It might be difficult, but it's definitely still happening. Enormous progress has been made on this front over the past however many years since the current gen consoles launched. All the major 3D engines, Unreal Engine 4, Frostbite 3, CryEngine, AnvilNext, Id Tech6 etcetera, and even some from smaller devs like the 4A Engine used in the Metro series, can effectively use at least six threads. AnvilNext has been shown to be able to scale on a deca core CPU even. This development was done out of necessity, to be able to milk as much juice as they could out of the weak console CPUs, and to satiate the market's increasing desire for bigger and more realistic games. There's no stopping the progress being made on this front, as the future of gaming depends greatly on programmers' capabilities to get as much out of the hardware as is possible.
A lot of developers still haven't been able to break the four thread barrier, like CDPR's Red Engine 3. That might not have been a limitation for the Witcher 3, but it very well might be for their next game Cyberpunk 2077 which will likely be an even bigger game with more simulation. If CDPR can't get their engine to scale to more than 4 threads, then their vision for Cyberpunk 2077 will have to be compromised, much like how the Witcher 3 was severely downgraded.
Ubisoft by comparison has done more on this front than perhaps any developer, AND IT SHOWS. AC Origins will be an immense game, featuring practically an entire country with diverse environments from massive populated cities to small villages to deserts, to huge Pyramids and underground tombs. And all of this will be completely seamless with no loading screens. You can't have that kind of seamless scope and detail without an engine that can effectively and properly utilize a CPU and memory. Only now is such a thing even possible, due to the advancements they've made in game technology, and proper CPU scaling is crucial to their success no doubt.
It’s the market’s choice to buy, not to build. We did not have mainstream 8 core because we were in an effective monopoly, that does not mean the market would auto choose the many core AMD product.If it was really the market's choice, then why didn't we have a reasonable 8 core option before Ryzen dropped?
What you are seeing in the above link are the effects of IPC improvements plus extra L3 cache, and not scaling past 4/8. Red Engine 3 can actually use HT though, but only in the CPU bound areas of the game like Novigrad where it can make a significant difference.
PCGH.de did a test on the CPU scaling after the game launched.
I voted 8 cores will be the new standard.
Now Ive gotta sell off my 7700k/Asus Z170 combo way way sooner than i expected.
There are no IPC gains in coffee Lake and the L3 isn't any higher.
I wouldn't get too excited, it still needs to run at 30fps on the original PS4, which only has weak Jaguar CPUs. Any console ports will be held back for years to come.
@Carfax83 I love Ubisoft as much as you, maybe more than you. Rainbow 6 Siege is my favorite game. But they release the buggiest games I have ever played. They will find a way to botch this new AC unfortunately. They'll fix it eventually I'm sure. If they load all AI on core 0 again I will die.
There were no IPC improvements in CL at all. None have been shown anywhere, and nowhere is showing any significant cache improvements either
It is likely your reference created a bottleneck somewhere running the core that slow, or maybe the early build had that behavior.
But it clear that today Witcher 3 benefits VERY significantly in moving from 4 to 6 cores.
8700K has 12MB of cache, and the 8600K has 9MB of cache compared to the 8MB in the 7700K. The cache size increase will make a bigger than normal difference,
Criss crossed cables there. I was thinking per core. You obviously over all. CFL is just a 2 core increased Kabylake.OK I checked and you're right about the IPC gains. However, you're wrong about the L3 cache. 8700K has 12MB of L3 cache, compared to 8MB for the 7700K. That's going to make a big difference on a dual channel memory config.
And I would gladly take a 50% increase in cores, however I'm not willing to spend about ~$400 or more to do it.Criss crossed cables there. I was thinking per core. You obviously over all. CFL is just a 2 core increased Kabylake.