The point being made was that nobody buys frequency or IPC. What people want is more performance, yet many posters on the Internet act like performance doesn't count unless it comes from this holy grail of more IPC. In fact, I would say that the opposite is desirable. When performance is derived from increased frequency (i.e. speed), all workloads improve, yet when gains in IPC are made, only those applications with the right instruction mix show improvement. It was shown back in 2015 or even earlier that Apple A-series SoCs have higher IPC than Intel's Core processors, yet they perform worse.Who cares? The kind of CPU enthusiast that inhabits these forums, that's who. All day there are discussions about the minutiae of how performance is achieved. I like most of your posts, but pretending it doesn't matter flies like a lead balloon here.
When we talk about IPC on the desktop market we usually are talking about actual performance or more specifically we are comparing CPU's on the same platform with the same workload and generalizing it. It wouldn't matter if AMD or Intel had the better theoretical IPC by far if all the benchmarks and programs we run had the other CPU doing better in all tasks. So we are talking relative IPC on similar CPU's expected to run the same software.The point being made was that nobody buys frequency or IPC. What people want is more performance, yet many posters on the Internet act like performance doesn't count unless it comes from this holy grail of more IPC. In fact, I would say that the opposite is desirable. When performance is derived from increased frequency (i.e. speed), all workloads improve, yet when gains in IPC are made, only those applications with the right instruction mix show improvement. It was shown back in 2015 or even earlier that Apple A-series SoCs have higher IPC than Intel's Core processors, yet they perform worse.
Why it matters for Skylake vs. Kaby Lake would be that really Kaby lake isn't a better overclocker at least inside its retail packaging. Without defacing and IMHO destroying the the CPU, Skylake and Kaby Lake have the same max OC. ~4.7GHz. In fact it makes SL a better overclocker since KL has much higher stock clocks. From there without any real architectural differences there is no real performance difference. Most of KL's 5% better CPU comes from Intel setting a higher clock one which they could have done with SL.
The point being made was that nobody buys frequency or IPC. What people want is more performance, yet many posters on the Internet act like performance doesn't count unless it comes from this holy grail of more IPC. In fact, I would say that the opposite is desirable. When performance is derived from increased frequency (i.e. speed), all workloads improve, yet when gains in IPC are made, only those applications with the right instruction mix show improvement. It was shown back in 2015 or even earlier that Apple A-series SoCs have higher IPC than Intel's Core processors, yet they perform worse.
So if you buy a cpu that is clocked 10% higher than its predecessor do you downclock it by 10%, since performance gains from higher clockspeeds don't seem to matter to you? Arach is right. If performance is improved, why does it matter which mechanism it is.Who cares? The kind of CPU enthusiast that inhabits these forums, that's who. All day there are discussions about the minutiae of how performance is achieved. I like most of your posts, but pretending it doesn't matter flies like a lead balloon here.
I never claimed that gains from clockspeed don't matter, there's nothing in my post that come close to saying that; your example is pulled from thin air. On the contrary, my claim is that it ALL matters. Objectively, there's no rule that states we have to ignore the total lack of IPC gain just because there was a minor speed bump. But speed bumps are terrific, go KBL, yay!So if you buy a cpu that is clocked 10% higher than its predecessor do you downclock it by 10%, since performance gains from higher clockspeeds don't seem to matter to you? Arach is right. If performance is improved, why does it matter which mechanism it is.
Yea, it is pretty obvious why some are trying to minimize clockspeed gains, especially since chips of Intel's chief competitor clock much lower. I would love to state it bluntly, but dont care to get an infraction.If clock speeds fundamentally go up, performance goes up.
Who cares if the performance is achieved with higher IPC or higher frequencies?
I've never had an infraction before, so I'll be bold: Intel has both the best IPC and clockspeeds in the business, so why the threatened posture? A little competition got Intel's panties in a bunch?Yea, it is pretty obvious why some are trying to minimize clockspeed gains, especially since chips of Intel's chief competitor clock much lower. I would love to state it bluntly, but dont care to get an infraction.
You have pretty good point. IPC and IPC gai s have been a huge focus of this Forum in general since Sandy Bridge and played a big part in people's disappointment (rightfully so of course) in Bulldozer. Clock and IPC go hand in hand, if you are losing in one you need to make it up on the other. Clock speed has the most linear effect on system performance so getting good clocks is important.I've never had an infraction before, so I'll be bold: Intel has both the best IPC and clockspeeds in the business, so why the threatened posture? A little competition got Intel's panties in a bunch?
Looks like you did just that.Yea, it is pretty obvious why some are trying to minimize clockspeed gains, especially since chips of Intel's chief competitor clock much lower. I would love to state it bluntly, but dont care to get an infraction.
Intel persistent memory will be available in 2018 as part of an Intel Xeon Processor Scalable family refresh (codename: Cascade Lake).
We have reached a point where all performance increase vectors have high diminishing returns, and it's usually alternating combinations that yield optimal increase in perf/watt/price for each generation. (frequency, core IPC, core number)What is the point of clock and core improvements discussion?
I though performance/watt/price for a given set of usecases is all that should matter?
https://newsroom.intel.com/news/int...pcoming-intel-xeon-processor-scalable-family/
Cascade Lake family next year to replace Skylake-SP & Skylake-X.
Intel makes no mention of that, though. They typically stress transistor performance improvements if they have made any.So is Cascade Lake to be based on Intel's 14nm++ process?
Edit: I think @Arachnotronic has already suggested that this is true.
Intel makes no mention of that, though. They typically stress transistor performance improvements if they have made any.
Seems like a no-brainer if Cascade Lake is just a refresh of Skylake-EP. TMFChipFool seems to think that Cascade is going 14nm++ as well.
https://www.fool.com/investing/2017/05/17/intel-corporation-announces-2018-server-chip-plan.aspx
I am going to go out on a limb and say these will be Intel's first 10nm skus, based on the fact Intel is said to be changing direction on which cpu lines are being brought to market first.
I thought it was mentioned here that while Intel was moving to a 3 step cadence with its consumer line of cpu's, it was sticking to a 2 step cadence with its Xeon line (tick-tock). I could be mistaken, but if that is true, then it could be a 10mn part.
I am going to go out on a limb and say these will be Intel's first 10nm skus, based on the fact Intel is said to be changing direction on which cpu lines are being brought to market first.
I thought it was mentioned here that while Intel was moving to a 3 step cadence with its consumer line of cpu's, it was sticking to a 2 step cadence with its Xeon line (tick-tock). I could be mistaken, but if that is true, then it could be a 10mn part.
Ashraf Eassa said:the first products to use Intel's 10-nanometer++ technology -- its third-generation 10-nanometer technology -- as well as the first products to use Intel's 7-nanometer technology will be data-center chips, rather than personal-computer chips.
Concerning "data center first": https://www.fool.com/investing/2017/02/12/intel-corporation-plans-seismic-change-to-data-cen.aspx
Refers to 10nm++ not 10nm or 10nm+. This is no option for next year or two.