Canard mentions second gen' EPYC, if true, nice!

[AndreiD]

@DarrenM343 If put to right use, but let's be honest here, most people don't have use cases that will peg your basic 4C/4T CPU.
And if we're talking media encoding, which probably would be the most common multithreaded workload, then you run into the AVX/FMA issue since H.264/H.265 have support for those instruction sets: https://www.techpowerup.com/reviews/Intel/Core_i3_8350K/6.html
1800X 8C/16T is on par with a 4C/8T 7740X overall in Premiere Pro CC: https://www.pugetsystems.com/labs/a...aby-Lake-X-Broadwell-E-Kaby-Lake-Ryzen-7-969/

I'm not saying that thew new Threadrippers/Ryzens aren't great if you can use the threads, but I honestly doubt most people do. For most consumer use I'd take a CPU with stronger per core performance almost always.

 

[DragonQ]

Something people often seem to forgot is that single core performance affects multi-core performance. Improving the former automatically improves the latter in most cases, so even as single core performance is becoming less and less important, it's always going to be a factor.

 

[gavinh87]

Something people often seem to forgot is that single core performance affects multi-core performance. Improving the former automatically improves the latter in most cases, so even as single core performance is becoming less and less important, it's always going to be a factor.

Yep!

 

[DarrenM343]

@DarrenM343 If put to right use, but let's be honest here, most people don't have use cases that will peg your basic 4C/4T CPU.
And if we're talking media encoding, which probably would be the most common multithreaded workload, then you run into the AVX/FMA issue since H.264/H.265 have support for those instruction sets: https://www.techpowerup.com/reviews/Intel/Core_i3_8350K/6.html
1800X 8C/16T is on par with a 4C/8T 7740X overall in Premiere Pro CC: https://www.pugetsystems.com/labs/a...aby-Lake-X-Broadwell-E-Kaby-Lake-Ryzen-7-969/

I'm not saying that thew new Threadrippers/Ryzens aren't great if you can use the threads, but I honestly doubt most people do. For most consumer use I'd take a CPU with stronger per core performance almost always.

I was thinking more along the lines of applications being developed to better use more cores when I said "put to right use", rather than an individual users use case. Content creation etc uses more cores but I don't see a reason why other apps cannot, other than they are not currently designed to do so. Windows 10 I'm sure will use as many cores as are available if it needs to and if you run multiple applications at the same time a multicore processors should give a smoother faster experience.

Sure, many probably don't need the cores but I think everyone can argue until the cows come home over this . I think the consensus is that the future is higher core counts with software to make better use of them. At the moment, for gaming only, of course 4-6 are probably still optimal.

I've just done some work with a company on improving performance of their complex software and was surprised to see it couldn't utilise more than four cores (soon to be changed). Actually that wasn't quite true but the performance gain just dropped off the more cores it used (>4). Software architects & developers are just not thinking enough about the hardware available. Now that we're seeing 6-18 core processors hit the mainstream hopefully things change.

 

[humbug]

Something people often seem to forgot is that single core performance affects multi-core performance. Improving the former automatically improves the latter in most cases, so even as single core performance is becoming less and less important, it's always going to be a factor.

Yes of course it does, but i think even you would agree that a 2 core CPU is no longer better than a 6 or even 4 core with a bit less per-core performance, back in the FX days yes, these days? you will find that even in all but a minority of games now the 6 and 4 core CPU with 'a bit less per-core performance' will easily out perform the 2 core with 'a bit more per-core performance'.

Thats the point being made.

 

[DragonQ]

Yes of course it does, but i think even you would agree that a 2 core CPU is no longer better than a 6 or even 4 core with a bit less per-core performance, back in the FX days yes, these days? you will find that even in all but a minority of games now the 6 and 4 core CPU with 'a bit less per-core performance' will easily out perform the 2 core with 'a bit more per-core performance'.

Thats the point being made.

Depends on your use case but sure, in general terms I'd much prefer a 6 or 8 core over a 2 or 4 CPU, even if I was sacrificing some single core performance. Multitasking is a fact of computing life these days.

 

[smilingcrow]

Something people often seem to forgot is that single core performance affects multi-core performance. Improving the former automatically improves the latter in most cases, so even as single core performance is becoming less and less important, it's always going to be a factor.

So it's generally not so much what the maximum performance is for software that only uses one core but what is the maximum performance per core when ALL or MOST cores are loaded.
This is important for software that is multi-threaded but still relies on a 'master' thread that is limited by the performance of the one core it runs on so that is the bottleneck to a degree for the whole application.
So it's about what is the maximum performance when the clock speed is as defined for an all core load or a 4 core load etc.
I think this maybe is the major reason why Ryzen lacks consistency although there are other reasons.

 

[humbug]

Ryzen's IPC is lower than CoffeeLake but higher than Haswell/Devils Canyon, this in low threaded situations where SMT is not the deciding factor, where it is Ryzen has the same IPC as CoffeeLake.

Where Ryzen "lacks" is clockspeed, a 3.7Ghz CPU is not going to look good in benchmarks next to a 4.7Ghz CPU with the same IPC.

 

[smilingcrow]

Ryzen's IPC is lower than CoffeeLake but higher than Haswell/Devils Canyon, this in low threaded situations where SMT is not the deciding factor, where it is Ryzen has the same IPC as CoffeeLake.

Where Ryzen "lacks" is clockspeed, a 3.7Ghz CPU is not going to look good in benchmarks next to a 4.7Ghz CPU with the same IPC.

Which is why performance per core still matters which is related to single core performance.
Next year we will likely see higher clock speeds with Ryzen Mk2 and more cores with Intel at 10nm.
That should be a more even match as at the moment Intel and AMD are a long way apart in terms of clock speeds and cores.
That should be an interesting comparison.

 

[Matkinson]

Getting a 10, 12 or 16 core cpu would be insane. (for mainstream prices)

I suspect this isn't that far away, seeing how intel's upcoming processors for 2018 are rumoured to be 8c16t. I imagine AMD would only want to counter this (fingers crossed)

 

[jigger]

Getting a 10, 12 or 16 core cpu would be insane. (for mainstream prices)

I suspect this isn't that far away, seeing how intel's upcoming processors for 2018 are rumoured to be 8c16t. I imagine AMD would only want to counter this (fingers crossed)

It seems unlikely will Intel will offer anything to rival a Ryzen 8 core. If Intel offer another APU then I can't see how they compete just from moving to 10nm. Unless 10nm brings a big reduction in production costs and big jump in yields then I think it will be business as usual for Intel.

 

[jigger]

So it's generally not so much what the maximum performance is for software that only uses one core but what is the maximum performance per core when ALL or MOST cores are loaded.
This is important for software that is multi-threaded but still relies on a 'master' thread that is limited by the performance of the one core it runs on so that is the bottleneck to a degree for the whole application.
So it's about what is the maximum performance when the clock speed is as defined for an all core load or a 4 core load etc.
I think this maybe is the major reason why Ryzen lacks consistency although there are other reasons.

I'm sure some software is written that way, but it's a backward design and not a very effective use of hardware. The main gain from increased performance or potential performance through increasing performance per core over a multi cores design is the all gains collectively from the physical and virtual cores.

If core 0 running thread 1 as fast as possible was a critical performance factor then all you need to see gains would be core 0 running at max clocks but we don't see that. If anything we see the opposite.

 

[smilingcrow]

I'm sure some software is written that way, but it's a backward design and not a very effective use of hardware. The main gain from increased performance or potential performance through increasing performance per core over a multi cores design is the all gains collectively from the physical and virtual cores.
If core 0 running thread 1 as fast as possible was a critical performance factor then all you need to see gains would be core 0 running at max clocks but we don't see that. If anything we see the opposite.

Software isn't written for ideal theoretical situations but for the real world and that's not a failing in the design but an acknowledgement of the complexity of life.

 

[Rroff]

Software isn't written for ideal theoretical situations but for the real world and that's not a failing in the design but an acknowledgement of the complexity of life.

Yeah it isn't just about backwards approaches but program flow tends to polarise towards stuff that can be highly parallelised or highly serial with little potential to turn highly serial logic sequences into something that can easily be processed in a parallel manner. Hence its unlikely software will ever truly not be dependant on 1-2 high tick rate threads for the ultimate performance profile even when otherwise highly threaded.

 

[jigger]

Software isn't written for ideal theoretical situations but for the real world and that's not a failing in the design but an acknowledgement of the complexity of life.

But as I said you get a lot more performance from a higher average clock speed in most situations. If you write code based around a single cores out put then it's bound by that limitation.

 

[Mercutio]

I'd say at the moment, the winner for "overall best chip" is going to be one that has a decent chunk of cores with very high boosting available to 2 or so. It's literally what's being argued above. A couple of heavy cores to carry the main program thread and a good number of lesser clocked cores to handle lots of smaller/child threads.

 

[jigger]

I'd say at the moment, the winner for "overall best chip" is going to be one that has a decent chunk of cores with very high boosting available to 2 or so. It's literally what's being argued above. A couple of heavy cores to carry the main program thread and a good number of lesser clocked cores to handle lots of smaller/child threads.

Best overall chip would be the cheapest you can put up with. In your case probably a dual core i3k

 

[smilingcrow]

But as I said you get a lot more performance from a higher average clock speed in most situations. If you write code based around a single cores out put then it's bound by that limitation.

Real world tasks are often bound by limitations which is why a 6C CL overall is still a better choice than an 8C Ryzen.
That's not something that will get addressed in software ever in some cases.

 

[Mercutio]

Best overall chip would be the cheapest you can put up with. In your case probably a dual core i3k

I wasn't meaning "<this> chip wins" I was pointing more... generally in the direction processors should probably go.
As others pointed out, Ryzen does better when more cores can be used effectively, Intel wins when something doesn't do multi-core well so needs greater performance from fewer cores...

A chip that has a couple of cores that can fly and a bunch that are "decent" would likely be an overall strong contender right now. Maybe one that could intelligently power down parts of the chip to reduce overall load and give more to a couple of higher performing chips would be ideal. It's a "one size fits all" type thought.

For me, right now, there's not a HUGE improvement to be had anywhere. Pinnacle looks promising and is going to line up with cash to buy and other personal bits. We'll see though

 

[jigger]

I wasn't meaning "<this> chip wins" I was pointing more... generally in the direction processors should probably go.
As others pointed out, Ryzen does better when more cores can be used effectively, Intel wins when something doesn't do multi-core well so needs greater performance from fewer cores...

A chip that has a couple of cores that can fly and a bunch that are "decent" would likely be an overall strong contender right now. Maybe one that could intelligently power down parts of the chip to reduce overall load and give more to a couple of higher performing chips would be ideal. It's a "one size fits all" type thought.

For me, right now, there's not a HUGE improvement to be had anywhere. Pinnacle looks promising and is going to line up with cash to buy and other personal bits. We'll see though

Well it's always been horses for courses. If I was going to brand the best overall CPU to cover 90% of situations excluding Xeons and EYPC. Then I'd say it's the 7980xe or maybe the X1950.