The thinking is that while 100 and 200 series motherboards are not forward-compatible with 8th processors, 300 series is backward-compatible with 6th and 7th generation processors, as with the situation with 8 and 9 series motherboards, and 4th and 5th generation processors.But the 7700K is not socket-compatible with 8th-gen Core motherboards. ?
I wonder if there's any further overclocking headroom to be had by combining Kabylake with Z370 chipset mobos?The thinking is that while 100 and 200 series motherboards are not forward-compatible with 8th processors, 300 series is backward-compatible with 6th and 7th generation processors.
Many of us do a bit of mental arithmetic like this to estimate relative execution throughput, but it's not really that simple, and anyone who says that performance can be calculated accurately this way is talking out of their cloaca....Core Count X IPC X Clock speed...
See the changes from Z270 in bold:I wonder if there's any further overclocking headroom to be had by combining Kabylake with Z370 chipset mobos?
Many of us do a bit of mental arithmetic like this to estimate relative execution throughput, but it's not really that simple, and anyone who says that performance can be calculated accurately this way is talking out of their cloaca.
It's not reasonable to use this kind of estimate and then use terms like "always" and "every" so describe its accuracy, as you did in your post. I didn't quote all that, so there's still time to change it.It's a reasonable starting point. It's a hell of lot better than the "Most cores wins" argument I was debunking.
Please explain why we have 16,000 core / 32,000 thread ThreadWeaver CPUs at 3.5 MHz for mainstream then?It's a reasonable starting point. It's a hell of lot better than the "Most cores wins" argument I was debunking.
Please explain why we have 16,000 core / 32,000 thread ThreadWeaver CPUs at 3.5 MHz for mainstream then?
Oh wait, we don't.
Despite what people think about cores, we have almost exclusively gone for the billions of hertz and miniscule handful of cores optimum for a reason. On the all cores to all hertz spectrum, the needle is almost entirely on the all hertz side of the scale.
I'm giving you fodder for your argument. No CPU manufacturer has gone for the many core / slow speed route. Performance and price suck there. All CPUs go for the few cores many hertz optimum.Huh? I am not sure where you are getting the idea I am arguing more cores are the way to go? I am arguing against someone who simplistically claimed more cores are always better.
Generally fewer faster cores are usually better, and even on parallel loads they can be faster, if they have enough individual speed advantage.
It's not reasonable to use this kind of estimate and then use terms like "always" and "every" so describe its accuracy, as you did in your post. I didn't quote all that, so there's still time to change it.
Incorrect Again.
It is which ever has the highest combination of Clock Speed X Core Count, that will win on highly parallel workloads.
With the same basic cores:
4 GHz X 8 Cores Will always beat 3 GHz X 10 cores, in every workload.
Both the core count and the core performance matter in parallel workloads.
Your posts continue to indicate that whoever has the most cores win, which is NOT the case.
To get a parallel performance potential factor you need to consider:
Core Count X IPC X Clock speed.
Which is why 6 core Coffee Lake could match Ryzen 8 cores, even in highly parallel workloads. Something like this could happen:
Ryzen: 8c X 1.0ipc X 4GHz = 32 PPPF.
CofLk: 6c X 1.1ipc X 5GHz = 33 PPPF.
I used always in reference to a specific example and I stand by it.
8-4GHz cores will always beat 10-3GHz cores, in all kinds of loads, all else being equal (No shenanigans bottle-necking the 4GHz core with less memory bandwidth, or writing an OS that only allows one thread/core)
A smaller number of Faster cores can beat some greater number of lesser performing cores, even on highly parallel workloads (and are obviously much better on less parallel loads).
Some people seem to be under the mistaken impression the thread count needs to have some relation to core count. When it really doesn't.
A fast enough single core could beat an 8 core CPU on parallel loads, if it had enough of a speed advantage.
Incorrect Again.
It is which ever has the highest combination of Clock Speed X Core Count, that will win on highly parallel workloads.
With the same basic cores:
4 GHz X 8 Cores Will always beat 3 GHz X 10 cores, in every workload.
Both the core count and the core performance matter in parallel workloads.
Your posts continue to indicate that whoever has the most cores win, which is NOT the case.
To get a parallel performance potential factor you need to consider:
Core Count X IPC X Clock speed.
Which is why 6 core Coffee Lake could match Ryzen 8 cores, even in highly parallel workloads. Something like this could happen:
Ryzen: 8c X 1.0ipc X 4GHz = 32 PPPF.
CofLk: 6c X 1.1ipc X 5GHz = 33 PPPF.
Please explain why we have 16,000 core / 32,000 thread ThreadWeaver CPUs at 3.5 MHz for mainstream then?
Oh wait, we don't.
Despite what people think about cores, we have almost exclusively gone for the billions of hertz and miniscule handful of cores optimum for a reason.
A fast enough single core could beat an 8 core CPU on parallel loads, if it had enough of a speed advantage.
Regarding your 4 GHz x 8 Cores will always beat 3 GHz x 10 cores in every workload, again reeks of simplicity and lack of understanding. If the application can scale very well to ten threads or twenty with SMT, then it's definitely possible that it will overcome the clock speed deficit.
I already provided an example of this on the previous page where a 3ghz 6950x was able to trounce a 4.2ghz 7700K in Watch Dogs 2.
Not really. The laws of physics do not allow you to increase frequency beyond a certain limit. Most modern Intel high performance CPUs hit a clock wall at 5 Ghz. The current Zen hits it at 4 Ghz. I can tell you this. Neither Intel or AMD can release a desktop CPU which can run at 6 Ghz on air cooling.
Really, you are only showing again and again, that you don't get it. You keep assuming that more cores always win as long the load is parallel, which is totally absurd.
You are comparing 10 cores to 4 cores in your one sided example, you have more than double the cores, so you would have to run the 4 core at more than double the clock speed to equalize it.
Run that 6950x at 1GHz and the 7700K would beat it in any benchmark, even fully parallel ones.
Well, you could get a ~5X difference with a 7700K. You could lock it at 800mhz for all cores, and then you could unlock it and turn off HT and enable a single core.I am not arguing that you can build that CPU. It is obvious that we are close to the wall on clock speed.
But with a sufficient clock speed differential a single core could beat a significantly slower 8 core.
No laws of physics need to be challenged either. You could run the single core at 4GHz, and the 8 core at 400MHz. The single core will be faster even at full parallel loads (all else being equal).
The real overall upgrade to the 4790K/7700K has been out for months already, it's the less expensive upgradable 8 core Ryzen 1700. Sorry Intel, maybe next generation.
Intel expecting users to upgrade even though these new chips sport the exact same 1151 pins is a slap in the face to B/H/Z 170/270 owners. They probably added some stupid shit to the controller like "pay for RAID" to squeeze more out of consumers. I'll be skipping these chips.
The real overall upgrade to the 4790K/7700K has been out for months already, it's the less expensive upgradable 8 core Ryzen 1700. Sorry Intel, maybe next generation.
The more parallel a workload is, the less clock speed matters.
Edit: missed that Carfax83 changed the discussion from CPU to GPU.When was the last time you saw a GPU that had a handful of cores with extremely high clock speeds? .