Mid-range gaming PC graphics cards, CPU and Mobo advice

Axe_meister

There are a couple if issues with current CPU technology that means moores law is coming to an end.
At 7nm we are getting to the point that we just can't make anything smaller, we are in the realms of making a CPU out of individual atoms.
We have to find a new way of doing computing which will (if we stick to turing computing) will mean higher parallelization of workloads.
And yes the speed of light is becoming a bit of a problem.
I work with some bloody big computer systems (16TB of Memory in a single OS image) and once you get past two 19inch rack units the speed of light in fibre/cable becomes a big problem and you introduce massive latency into the equation.
Intel has been
a) Making a heck of a lot of money
b) Have made a lot of false starts (Mobile Processors)
c) GPUs
d) In this case I think they have been caught with their pants down. AMD have argubly the best
processor designer on their team (the very same person who came up with the Athlon in the first place).
e) Itanium distaster
AMD: 1st for onboard Memory controller
          1st for SSE extentions (3DNow)
          1st for 64bit extensions
          1st for Multicore
          1st for onboard GPU

It almost as if AMD is the skunkworks for intel. They do all the innovation and intel just
copies them.

Also intel may have a massive R&D budget but alot of that actually goes on fabrication rather than chip design. Not an issue AMD has as they sold off their fabrication assetts

Myranda

Axe_meister said:

There are a couple if issues with current CPU technology that means moores law is coming to an end.
At 7nm we are getting to the point that we just can't make anything smaller, we are in the realms of making a CPU out of individual atoms.
We have to find a new way of doing computing which will (if we stick to turing computing) will mean higher parallelization of workloads.
And yes the speed of light is becoming a bit of a problem.
I work with some bloody big computer systems (16TB of Memory in a single OS image) and once you get past two 19inch rack units the speed of light in fibre/cable becomes a big problem and you introduce massive latency into the equation.
Intel has been
a) Making a heck of a lot of money
b) Have made a lot of false starts (Mobile Processors)
c) GPUs
d) In this case I think they have been caught with their pants down. AMD have argubly the best
processor designer on their team (the very same person who came up with the Athlon in the first place).
e) Itanium distaster
AMD: 1st for onboard Memory controller
          1st for SSE extentions (3DNow)
          1st for 64bit extensions
          1st for Multicore
          1st for onboard GPU

It almost as if AMD is the skunkworks for intel. They do all the innovation and intel just
copies them.

Also intel may have a massive R&D budget but alot of that actually goes on fabrication rather than chip design. Not an issue AMD has as they sold off their fabrication assetts

While you're spot on about some stuff, the theoretical limit of CPU clock speed is WAAAAAAAAAAAY higher than we're at now. One of the things that came up in the first term was Fundamental Limits of computer hardware now rather that root out my coursework I have googled it and found a number that was slightly higher than expected at 10^17 Hz. (seem to remember that our lecturer said it was lower than that)

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

digitalscream

@Axe_meister - gawd, I forgot about Itanium. That was arguably a bigger foul-up than AMD's Bulldozer; the only thing that saved them there was that Intel are so big they could swallow the cost with profits from other business units.

I disagree about false starts on mobile CPUs, though. It's arguable that the mobile division saved their asses when the P4 was shown to be the Emperor's New Clothes - they noticed that enthusiasts were taking their Pentium M CPUs and overclocking them to 2GHz and thrashing the pants off their 3.5GHz P4s, so they brought out the Tualatin range (Pentium III with lots more cache) and that's what became the Core Duo CPUs.

I suspect we'd all have much higher-performing CPUs right now if Intel hadn't tried to beat AMD with marketing by lengthening the pipeline on the P4 to hit higher clock speeds. Oh, and let's not forget the RAMBUS debacle...

Axe_meister

I'm talking ultra mobile, i.e. ARM competitor.

Axe_meister

Myranda said:

Axe_meister said:

There are a couple if issues with current CPU technology that means moores law is coming to an end.
At 7nm we are getting to the point that we just can't make anything smaller, we are in the realms of making a CPU out of individual atoms.
We have to find a new way of doing computing which will (if we stick to turing computing) will mean higher parallelization of workloads.
And yes the speed of light is becoming a bit of a problem.
I work with some bloody big computer systems (16TB of Memory in a single OS image) and once you get past two 19inch rack units the speed of light in fibre/cable becomes a big problem and you introduce massive latency into the equation.
Intel has been
a) Making a heck of a lot of money
b) Have made a lot of false starts (Mobile Processors)
c) GPUs
d) In this case I think they have been caught with their pants down. AMD have argubly the best
processor designer on their team (the very same person who came up with the Athlon in the first place).
e) Itanium distaster
AMD: 1st for onboard Memory controller
          1st for SSE extentions (3DNow)
          1st for 64bit extensions
          1st for Multicore
          1st for onboard GPU

It almost as if AMD is the skunkworks for intel. They do all the innovation and intel just
copies them.

Also intel may have a massive R&D budget but alot of that actually goes on fabrication rather than chip design. Not an issue AMD has as they sold off their fabrication assetts

While you're spot on about some stuff, the theoretical limit of CPU clock speed is WAAAAAAAAAAAY higher than we're at now. One of the things that came up in the first term was Fundamental Limits of computer hardware now rather that root out my coursework I have googled it and found a number that was slightly higher than expected at 10^17 Hz. (seem to remember that our lecturer said it was lower than that)

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

Where did I mention clock speed ;o) We just can't get that much smaller using silicon and the current doping materials. A single silicon atom is .21nm, copper .14nm.
I think the limit is a 5nm node size but we start to experience quantum tunneling at 7nm.
 
The trouble with increasing clock speed is the heat, same with increasing cores counts.
One of our current biggest sellers is a dual socket E5-2699v4 based server that has 44Cores in total but clocked down at 2.3Ghz.

digitalscream

Axe_meister said:

I'm talking ultra mobile, i.e. ARM competitor.

Ah, yes. That's fair...the Atom has no redeeming features as far as I can tell, and there's no way it can compete with ARM at this point.

Axe_meister said:

Myranda said:

While you're spot on about some stuff, the theoretical limit of CPU clock speed is WAAAAAAAAAAAY higher than we're at now. One of the things that came up in the first term was Fundamental Limits of computer hardware now rather that root out my coursework I have googled it and found a number that was slightly higher than expected at 10^17 Hz. (seem to remember that our lecturer said it was lower than that)

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

Where did I mention clock speed ;o) We just can't get that much smaller using silicon and the current doping materials. A single silicon atom is .21nm, copper .14nm.
I think the limit is a 5nm node size but we start to experience quantum tunneling at 7nm.
 
The trouble with increasing clock speed is the heat, same with increasing cores counts.
One of our current biggest sellers is a dual socket E5-2699v4 based server that has 44Cores in total but clocked down at 2.3Ghz.

I suspect we'll see the focus come back to instructions-per-clock again - AMD are already ahead of the curve on this, whereas Intel have been doing their traditional thing of increasing the pipeline length (up to 16 stages with Skylake again) and compensating by optimising the cache, just like they did with the P4 (although that had 31 stages in the Prescott core!).

It does very much seem to me that Intel tend to chuck brute force at successive CPU generations to increase performance, whereas AMD try to be much more elegant about it. That alone makes me root for the little guy

mrkb

Thanks all, I followed the initial discussions!?!? Will check those recommendations out.

digitalscream

mrkb said:

Thanks all, I followed the initial discussions!?!? Will check those recommendations out.

Yeah, sorry...we nerded out a bit there. Par for the course, sadly...we obviously don't get enough chances to talk about this stuff

mrkb

It was interesting bit of Anorak-fest ;-)

Axe_meister

Yeah sorry. You would think working for one of the worlds largest IT companies I'd be sick of geek stuff.

Myranda

Axe_meister said:

Yeah sorry. You would think working for one of the worlds largest IT companies I'd be sick of geek stuff.

Geeks will geek. 

TwoSpellWizard

Going back to the original question (lol), I'd recommend having a look at http://www.logicalincrements.com/

It lists the best build you can put together at any particular price point and is regularly updated when any new hardware comes out. For what you outlined in the OP, something in the fair-good bracket would be fine.

If you're building yourself rather than buying a prebuilt then I've found both https://uk.pcpartpicker.com/ (for compatibilty and price comparison) and https://www.reddit.com/r/buildapc/ (for general advice, build help and troubleshooting) to be extremely helpful resources

ThePrettyDamned

digitalscream said:

Axe_meister said:

Wait till after next week. The new AMD Ryzen processor is being released and so far from leaked benchmarks it looks bloody good and will cause intel to drop their prices.

I'd completely forgotten about that. Looking through the motherboard leak...even the mid-range ones have two x16 slots and M.2 PCIe slots. This could be very interesting.

Ahh man, dual m.2 at Pci-e speed... In raid... 

Ultimate overkill? Be interesting though. 

digitalscream

ThePrettyDamned said:

digitalscream said:

Axe_meister said:

Wait till after next week. The new AMD Ryzen processor is being released and so far from leaked benchmarks it looks bloody good and will cause intel to drop their prices.

I'd completely forgotten about that. Looking through the motherboard leak...even the mid-range ones have two x16 slots and M.2 PCIe slots. This could be very interesting.

Ahh man, dual m.2 at Pci-e speed... In raid... 

Ultimate overkill? Be interesting though. 

At those speeds, RAID striping might actually hurt performance. SSDs are already effectively internally RAIDed (hence higher capacity = more chips in parallel = more speed), so there's probably not a lot of point.

Sporky

Myranda said:

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

Dunno how it varies for PCBs, but for cables signals do about half the speed of light, for optical fibres it's about a third.

mrkb

TwoSpellWizard said:

Going back to the original question (lol), I'd recommend having a look at http://www.logicalincrements.com/

It lists the best build you can put together at any particular price point and is regularly updated when any new hardware comes out. For what you outlined in the OP, something in the fair-good bracket would be fine.

If you're building yourself rather than buying a prebuilt then I've found both https://uk.pcpartpicker.com/ (for compatibilty and price comparison) and https://www.reddit.com/r/buildapc/ (for general advice, build help and troubleshooting) to be extremely helpful resources

Thanks!

Myranda

Sporky said:

Myranda said:

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

Dunno how it varies for PCBs, but for cables signals do about half the speed of light, for optical fibres it's about a third.

Last term we measured it in a wire and got 20% and weren't told we were wrong... 

randella

Good bit of summary info re. Ryzen here...

http://www.eurogamer.net/articles/digitalfoundry-2017-how-amds-ryzen-will-disrupt-the-cpu-market

Sporky

Myranda said:

Sporky said:

Myranda said:

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

Dunno how it varies for PCBs, but for cables signals do about half the speed of light, for optical fibres it's about a third.

Last term we measured it in a wire and got 20% and weren't told we were wrong... 

20% of the speed of light, or 20% slower?

It does depend on the cable structure and any dielectric - I think the 0.5 figure is for twisted pair.

Myranda

Sporky said:

Myranda said:

Sporky said:

Myranda said:

There are more realistic limits - as you say the speed of light ... and in copper you get the 20%ish speed limitation as electrical signals don't travel at the full speed of light, so long copper tracks on motherboards are an issue, hence people talking about light instead.

Dunno how it varies for PCBs, but for cables signals do about half the speed of light, for optical fibres it's about a third.

Last term we measured it in a wire and got 20% and weren't told we were wrong... 

20% of the speed of light, or 20% slower?

It does depend on the cable structure and any dielectric - I think the 0.5 figure is for twisted pair.

20% slower - was a single core, copper-of-some-sort wire with nothing on the other end, so measuring the bounce-back of the initial signal pulse