spiroyster wrote:

What do you mean by 'approximation methods'? Interpolation is done between each frame/state, each frame/state is only broken down into so much detail (time/topology) so its always going to be an approximation. Approximations can be on the money though. Its not like they are always going to be wrong or slow 😀

Not what I meant. That's just convergence. You can iterate till you're happy so long as the algorithm does converge and your only limit will be your numerical representation precision. I meant simplified physical models that either have limitations in accuracy or applicability, but can work well enough for a particular purpose. Series expansions are ubiquitous examples.

spiroyster wrote:

Indeed, dependants all round, little room for parallelisation. That was my point.

Definitely doable, either by spatial or per object partitioning. Purely kinetic interactions are spatially localized. Magneto-kinetics on the other hand are quite a bit messier. This is, of course, a solution from first principles where a simplified model might be desired for computational efficiency.

gdjacobs wrote:

Definitely doable, either by spatial or per object partitioning. Purely kinetic interactions are spatially localized. Magneto-kinetics on the other hand are quite a bit messier. This is, of course, a solution from first principles where a simplified model might be desired for computational efficiency.

The devil is in the details:
In the early iterations, "everything interacts with everything", so you have lots of objects that need an iterative step in the process.
But as you continue to iterate, more and more objects 'settle', because they no longer have any other objects applying forces to them. So as you get further into the simulation, your parallelism 'dries up'. You have less objects to evaluate than you have cores to feed.
There's your Amdahl's Law again.

gdjacobs wrote:

spiroyster wrote:

What do you mean by 'approximation methods'? Interpolation is done between each frame/state, each frame/state is only broken down into so much detail (time/topology) so its always going to be an approximation. Approximations can be on the money though. Its not like they are always going to be wrong or slow 😀

Not what I meant. That's just convergence. You can iterate till you're happy so long as the algorithm does converge and your only limit will be your numerical representation precision. I meant simplified physical models that either have limitations in accuracy or applicability, but can work well enough for a particular purpose. Series expansions are ubiquitous examples.

There's always monte-carlo 😉

gdjacobs wrote:

spiroyster wrote:

Indeed, dependants all round, little room for parallelisation. That was my point.

Definitely doable, either by spatial or per object partitioning. Purely kinetic interactions are spatially localized. Magneto-kinetics on the other hand are quite a bit messier. This is, of course, a solution from first principles where a simplified model might be desired for computational efficiency.

Yes, thats what I was referring to when I mentioned 'kdtree' which is usually used when spatially partitioning something (octree is best). By grouping items this way you are assuming they are mutually exclusive to perform your calculation on. If they are not, or your tree is not constructed in a worthy fashion, objects which should be hitting each other will pass straight through. In the case of the jenga, this cannot be easily parallelised for a single jenga stack. Having multiple stacks over a wide area could be easily optimised to perform calculations on one stack (taking an assumption that the effect of a jenga stack falling over the floor is too far away to affect other stacks), and then parallelising this to concurrently calculate the other stack(s). There’s the parallelisation, but it is limited, and to be extra safe you should be building this tree on a per frame basis, which incurs the penalty of this construction each time Amda, Amda... Armdha... QED.

This is also why there should be a distinction between dynamic (can change each frame) and static (cannot change each frame) data, and accelerations should be designed accordingly.

[EDIT:] Typos and spellos.

P.S

Apologies for the massive derailment (or you could say parallel/concurrent threading 🤣... ).

I do wish AMD luck, I was all for the cause having a K6 through to Athlon, but there may be a reason they just cannot break the intel domination? idk, personally think they *might* have more luck attempting to break the gfx market, but then again it is nVidia? The processing capacity of some of their cards really is impressive, and while not great, they show potential to leave the x86 market behind if it went sour (Tegra?).

I do believe that x86 days are numbered though (I didn't think I would be saying that 10 years ago 🤣) with the increase of ARM and others. In terms of parallelisation who knows, imo the arguments go back and forth, one minute many cores is better, next minute fewer cores with each core having a higher throughput. Then it also depends on the application being run, and as discussed above o.0, how it solves its problems and can be optimised based on (a potentially) wide array of platform configurations. small low power multi-core RISC boards can be had for next to nothing these days and when (yes when!) we have platforms which favour massive amounts of cores (SMT), I think the design and architectures of programs may be entirely different. While its true that the x86 platform does appear to have stagnated with very little justification to upgrade based on real world performance and requirements, it doesn't seem to be that way with portable devices (ARM and other non CISC). We still have a way to go. At some point, when they get good enough someone in that massive server room is going to turn around and say... hmmm 1000's and 1000's of cores at hardly any power consumption and space, or hungrier x86. Just like maybe RISC wasn't the future for desktop PC's compared to CISC x86 at the time (90's)... x86 doesn't feel to me to be our future now.

I wonder how far this will, and if and when we start seeing it, it would take parallelisation to another level o.0
https://www.parallella.org/2016/10/05/epiphan … risc-processor/

</rant>

Next time maybe AMD? tbh AMD's pure existence in the market is enough to keep intel on its toes. One day the underdog may win (again), it did in the past imo!. They might have to drop to whole x86 'thing' though. 😀

spiroyster wrote:

In terms of parallelisation who knows, imo the arguments go back and forth, one minute many cores is better, next minute fewer cores with each core having a higher throughput.

After Amdahl's Law and Gustafson's Law, I'd like to introduce Scali's Law:
There are two extremes of processing:
1) Implicitly serial processing
2) Embarrassingly parallel processing

The first extreme is best served by a single core with a focus on maximizing IPC (conventional CPUs).
The second extreme is best served by a highly parallel SIMD/MIMD architecture (GPGPUs).

The more you can parallelize your algorithm, the more chance you have that the GPGPU is the best choice.
And the more you are limited by implicitly parallel components, the less it matters how many cores you have, but more how much IPC you can get.

Massive multi-core CPUs fall somewhere in between. I think as GPGPUs evolve, the usefulness of multi-core CPUs will get less and less.
Things may evolve into systems with APUs where you may have 4-8 cores focusing on high IPC, and a GPGPU for the REALLY parallel stuff.
At least, for desktop system/personal computers of course.

For server farms, things are different, as they use single machines to run multiple VMs. So how IPC vs cores/threads breaks down there is very difficult to gauge, and depends a lot on what is being virtualized, and what the workloads are.

spiroyster wrote:

P.S […]
Show full quote

P.S

Apologies for the massive derailment (or you could say parallel/concurrent threading 🤣... ).

I do wish AMD luck, I was all for the cause having a K6 through to Athlon, but there may be a reason they just cannot break the intel domination? idk, personally think they *might* have more luck attempting to break the gfx market, but then again it is nVidia? The processing capacity of some of their cards really is impressive, and while not great, they show potential to leave the x86 market behind if it went sour (Tegra?).

I do believe that x86 days are numbered though (I didn't think I would be saying that 10 years ago 🤣) with the increase of ARM and others. In terms of parallelisation who knows, imo the arguments go back and forth, one minute many cores is better, next minute fewer cores with each core having a higher throughput. Then it also depends on the application being run, and as discussed above o.0, how it solves its problems and can be optimised based on (a potentially) wide array of platform configurations. small low power multi-core RISC boards can be had for next to nothing these days and when (yes when!) we have platforms which favour massive amounts of cores (SMT), I think the design and architectures of programs may be entirely different. While its true that the x86 platform does appear to have stagnated with very little justification to upgrade based on real world performance and requirements, it doesn't seem to be that way with portable devices (ARM and other non CISC). We still have a way to go. At some point, when they get good enough someone in that massive server room is going to turn around and say... hmmm 1000's and 1000's of cores at hardly any power consumption and space, or hungrier x86. Just like maybe RISC wasn't the future for desktop PC's compared to CISC x86 at the time (90's)... x86 doesn't feel to me to be our future now.

I wonder how far this will, and if and when we start seeing it, it would take parallelisation to another level o.0
https://www.parallella.org/2016/10/05/epiphan … risc-processor/

</rant>

Next time maybe AMD? tbh AMD's pure existence in the market is enough to keep intel on its toes. One day the underdog may win (again), it did in the past imo!. They might have to drop to whole x86 'thing' though. 😀

Well if it weren't for AMD we all today might have been stuck with RAMBUS and with WAY less performance and much higher costs.

I think Intel having had the top-of-the-line playing field all by themselves has something to do with the stagnating of x86. They obviously only care about the money. Heck, they've made strides only due to competition (like massively dropping DX-4 prices as soon as competition arose, same for Pentium 1 MMX 233).

I think it will be interesting to see how parallelism will develop. And in the end it doesn't really matter if x86 is inside or not (or whether it is or is not Intel), as long as "it" does what the user wants it to do.

I'd personally prefer the continual existence of flexible desktop computers though, I can't mix and match CPUs and memory using parts I gathered from old second hand smartphones 😵

Scali wrote:

After Amdahl's Law and Gustafson's Law, I'd like to introduce Scali's Law: There are two extremes of processing: 1) Implicitly s […]
Show full quote

After Amdahl's Law and Gustafson's Law, I'd like to introduce Scali's Law:
There are two extremes of processing:
1) Implicitly serial processing
2) Embarrassingly parallel processing

I'd like to introduce Tetrium's law:

In the end consumers don't really care what's inside the box, as long as it does what the user wants it to do.

and the user will tell you long after they bought the shitty hardware they didn't do any research on before buying. heh

Tetrium wrote:

In the end consumers don't really care what's inside the box, as long as it does what the user wants it to do.

Consumers don't, it's the AMD fans arguing for "moar coars" on the internet. They (still) want to believe that there's going to be some silver bullet soon that will suddenly make all software scale like mad. Because up to now, only trained monkeys have been randomly banging on keyboards. But now we're going to get smart people to REALLY develop software!

I mean, if you believed that back in 2011 when they introduced the first 8-core Bulldozer, fine, you could claim ignorance... But it's 2017 now, AMD's "moar coars" revolution has been going on for many years. And what results has it brought us so far?

Scali wrote:

Tetrium wrote:

In the end consumers don't really care what's inside the box, as long as it does what the user wants it to do.

Consumers don't, it's the AMD fans arguing for "moar coars" on the internet. They (still) want to believe that there's going to be some silver bullet soon that will suddenly make all software scale like mad. Because up to now, only trained monkeys have been randomly banging on keyboards. But now we're going to get smart people to REALLY develop software!

I mean, if you believed that back in 2011 when they introduced the first 8-core Bulldozer, fine, you could claim ignorance... But it's 2017 now, AMD's "moar coars" revolution has been going on for many years. And what results has it brought us so far?

There's AMD fanboys, Intel Fanboys, there's red vs green. In the end they are all the exact same "wrong" anyway and are as silly as 2 religious people warring each other due to the other one not having their own religion.

Why would anyone sane want to be part of that? And why am I thinking about The X-Files now 🤣?

Tetrium wrote:

There's AMD fanboys, Intel Fanboys

Are there Intel fanboys? Hardly, I would say.
Just like Microsoft doesn't really have any 'fans' either.
These companies have an entirely different place.
It's the underdogs like AMD, linux and whatnot, where the rabid fanboys come out to 'spread the gospel'.

Because, let's face it... If fanboys would be a given percentage of the total userbase, then based on the marketshare of Intel/Microsoft, they'd have so many more fanboys that AMD/linux fanboys would have absolutely no chance in any conversation.
In reality, it's the AMD and linux advocates who are most vocal online, which is completely contrary to the position of these 'brands' in the marketplace.

Tetrium wrote:

there's red vs green.

No, that's ATi/AMD vs nVidia.

Tetrium wrote:

In the end they are all the exact same "wrong" anyway

Not really.
Firstly, not everyone who argues is a fanboy.
Secondly, even if they are fanboys, they're not necessarily wrong.

Scali wrote:

It's the underdogs like AMD, linux and whatnot, where the rabid fanboys come out to 'spread the gospel'. […]
Show full quote

Tetrium wrote:

There's AMD fanboys, Intel Fanboys

It's the underdogs like AMD, linux and whatnot, where the rabid fanboys come out to 'spread the gospel'.

Because, let's face it... If fanboys would be a given percentage of the total userbase, then based on the marketshare of Intel/Microsoft, they'd have so many more fanboys that AMD/linux fanboys would have absolutely no chance in any conversation.
In reality, it's the AMD and linux advocates who are most vocal online, which is completely contrary to the position of these 'brands' in the marketplace.

Well I guess that when looking at it from your perspective, FC Barcelona must also be an underdog. But at least they aren't very vocal about it.

Scali wrote:

No, that's ATi/AMD vs nVidia. […]
Show full quote

Tetrium wrote:

there's red vs green.

No, that's ATi/AMD vs nVidia.

Tetrium wrote:

In the end they are all the exact same "wrong" anyway

Not really.
Firstly, not everyone who argues is a fanboy.
Secondly, even if they are fanboys, they're not necessarily wrong.

Who cares what the fanboy craves? In the end they are all the same. They even share the same word for it no matter what they are fanboying about.
But fanboys do tend to be condescending about fanboys of the opposite side (like people who are Intel fanboys are condescending towards people they perceive as AMD fanboys) and this is what all fanboys tend to do and that is why I mentioned warring between different religions. They want to believe the other side is wrong and ignorant. And that in itself is what makes these fanboys stupid and ignorant.

Tetrium wrote:

Who cares what the fanboy craves?

I don't. But when I want to enter a serious technical discussion, such as this one, fanboys will pop up and inject their noise, and accuse people like me of being a fanboy. So fanboys affect me and others on this and other forums.

Scali wrote:

Tetrium wrote:

Who cares what the fanboy craves?

I don't. But when I want to enter a serious technical discussion, such as this one, fanboys will pop up and inject their noise, and accuse people like me of being a fanboy. So fanboys affect me and others on this and other forums.

Actually, you're the one who's been mentioning fanboys the most in this otherwise serious technical discussion so obviously you do care.
But let's save the debate about fanboyism for some other thread and let's get back on topic here shall we? I find the serious technical discussions here on Vogons much more interesting than the talk about fanboys anyway 😀

Tetrium wrote:

Actually, you're the one who's been mentioning fanboys the most in this otherwise serious technical discussion so obviously you do care.

As I say, I care that they affect me and the discussions I'm trying to have.
You're the one who brought 'fanboy' into this discussion, derailing it in the process.
You were the one who accused me of being an Intel fanboy. Which, given that I hate x86 and everything that goes with it, I find way off the mark, and frankly, quite insulting. It completely puts aside all the technical info I am giving, just because you perceive it to 'favour' the wrong brand.
Which is rather ironic, given that Intel has plenty of CPUs for sale with 6 cores, 8-cores and beyond. I wouldn't recommend those for gaming or general home/desktop use either. So what I say does not favour Intel any more than it does AMD. The difference is that AMD is trying to actively market their 8-core CPUs to gamers (because they currently don't have any other compelling options, not a very good reason to begin with), while Intel does not. At Intel's prices, their 6+ core CPUs are more or less automatically beyond the budget of the average gamer/home user, so we've never had to have this discussion.

Scali wrote:

Tetrium wrote:

Actually, you're the one who's been mentioning fanboys the most in this otherwise serious technical discussion so obviously you do care.

You're the one who brought 'fanboy' into this discussion, derailing it in the process.

Funny how you're accusing me of bringing up the 'fanboy' discussion whilst this is you on page 3 while I only brought it up on page 9 (without the accusations you already brought up 7 pages ago).

And besides everyone here already knows you're an Intel fanboy so no need to actually prove it.

Scali wrote:

This time it's prices, but no benchmarks or stability tests.
So far the entire internet is basing everything on a single vague test that AMD themselves performed... Gotta love how this always goes down with fanboys.

Myth busted...sorry Scali. Maybe you shouldn't accuse people of accusing people of something they didn't do and derailing this otherwise serious technical discussion in the process...again.

So I'd like to suggest (again) (and without naming anyone specifically) to stop talking about this fanboy stuff and go back to talking about stuff that most of us here actually really cares about 😀

In the next episode of Scali - The Animated Series:

"Who the hell cares about the pricetag? It's nothing more than an arbitrary number, all Intel has to do is lower it!"

Im kinda digging the 1700x my current rig is a fx8350 and that has worse heat issues then the prescott. The 1700x is rated at 65tdp vs my 8350=200 max load. Ill probably wait for the 1700x to drop in price.

Im hopping that the 1700x can support ddr3 for a cheap upgrade.

Oldskoolmaniac wrote:

Im kinda digging the 1700x my current rig is a fx8350 and that has worse heat issues then the prescott. The 1700x is rated at 65tdp vs my 8350=200 max load. Ill probably wait for the 1700x to drop in price.

Im hopping that the 1700x can support ddr3 for a cheap upgrade.

Sadly, you need a full upgrade, Ryzen and the AM4 platform is DDR4 only

R7 1700X is 95 watt TDP, R7 1700 is the 65 watt one

^ I got the two mixed.