The pace of obsolescence.

[IKnight]

With Moores law coming to a halt pretty soon, do you guys think that technology will keep becoming as obsolete at the same pace as it did from (let say) 1990-2010? 

And also, are we reaching a limit as to how small things can get/how much we can cram in (for example HDDs)?

[Moress]

This pace has already slowed considerably. 4+ year old CPUs (Looking at you Haswell) are still within a small margin (10 - 15% IPC) of the most modern CPUs on the market. 

[BloodyWaters]

3 minutes ago, Moress said:

This pace has already slowed considerably. 4+ year old CPUs (Looking at you Haswell) are still within a small margin (10 - 15% IPC) of the most modern CPUs on the market. 

I have a 6 year old CPU and it still keeps up with the latest games. 

[IKnight]

5 minutes ago, Moress said:

This pace has already slowed considerably. 4+ year old CPUs (Looking at you Haswell) are still within a small margin (10 - 15% IPC) of the most modern CPUs on the market. 

 

1 minute ago, BloodyWaters said:

I have a 6 year old CPU and it still keeps up with the latest games. 

That's a good point, considering my lga1366 Xeon still keeps up well. 

[Arika]

depends on the hardware mainly.

[IKnight]

Just now, Sierra Fox said:

depends on the hardware mainly.

I was mostly thinking about computer hardware like GPUs,CPUs.... 

[Arika]

1 minute ago, ImadKnight said:

I was mostly thinking about computer hardware like GPUs,CPUs.... 

again, depends on the hardware. if you compare a 5 year old CPU to a new one, there's going to be a small improvement. whereas a 5 year old GPU to one today, gets absolutely destroyed

 

 

[Coaxialgamer]

27 minutes ago, Moress said:

This pace has already slowed considerably. 4+ year old CPUs (Looking at you Haswell) are still within a small margin (10 - 15% IPC) of the most modern CPUs on the market. 

That's on intel and their complacency  . Look at progress on the gpu front as an example of performance increases . 

POWER has had quite a few improvements as well . 

 

[Coaxialgamer]

Things are not getting smaller at the rate they used to , but i doubt performance increases will stall . The industry has quite a few tricks up it's sleeve.

 

And besides, companies WILL fight against Rock's Law . 

[IKnight]

9 minutes ago, Coaxialgamer said:

That's on intel and their complacency  . Look at progress on the gpu front as an example of performance increases . 

POWER has had quite a few improvements as well . 

 

You mean power as in terms of efficiency or how "good" GPUs are getting? 

5 minutes ago, Coaxialgamer said:

Things are not getting smaller at the rate they used to , but i doubt performance increases will stall . The industry has quite a few tricks up it's sleeve.

 

And besides, companies WILL fight against Rock's Law . 

5

I think we're going to be pushing for more cores/clock speeds, though even that has its limits. 

[Moress]

10 minutes ago, ImadKnight said:

You mean power as in terms of efficiency or how "good" GPUs are getting? 

Both. Pascal/Polaris were big jumps in both raw processing power and lowering of power usage ( Ignoring Vega <.<  )

[IKnight]

12 minutes ago, Moress said:

Both. Pascal/Polaris were big jumps in both raw processing power and lowering of power usage ( Ignoring Vega <.<  )

True, the 1050ti can run on just 75w, whilst 1060s take 120w if I'm not mistaken.

And AMD has traditionally had a higher TDP/Voltage. 

[Coaxialgamer]

5 minutes ago, ImadKnight said:

You mean power as in terms of efficiency or how "good" GPUs are getting? 

I think we're going to be pushing for more cores/clock speeds, though even that has its limits. 

I meant IBM's POWER architecture , not power consumption.

POWER 7 had single core perf similar to haswell back in 2010 , and was considerably faster in MT applications .

POWER 8 made significant advancements as well , and had 8 way SMT . 

IBM POWER 9 should provide even more performance. 

They are monster chips , and IBM uses those extra transistors very well.

 

Gpus are examples of chips that can absorb process node shrinks very well . 

Between 2010 and 2017 we went through 3 node shrinks .

We went from the gtx 480 to the titan Xp . That's an improvements of over 5x in terms of performance . Power consumption has even dropped .

 

We are going to see more cores for sure , i wouldn't doubt that. But that has it's limits . sooner or latter , chip designers will have to be smarter in how they use each transistor .New , more efficient architectures can help improve performance , especially when aided by AI . 

Accelerators can also provide HUGE performance advantages when coded for . Look at GPGPU as an example : decoding video is orders of magnitude faster when done in hardware vs software. I expect more in more processors to drop general purpose units in favor of specialized , very efficient number crunchers .

We might see existing computation modules such as Von Neumann and Harvard dropped entirely , in favor of massively parallel architectures , such as DARPA's HIVE. The issue isn't in building them , it's in coding for them .

 

I don't see NAND or RAM stopping in growth . They can easily be stacked , and already are . MRAM is also gaining traction . Of course , there is a practical limit to chip height , but i don't think it'll be an issue for a long time.

 

Of course , this is just covering existing technologies . New tech ( some of it possibly already being researched ) could solve many issues , but it's just speculation at this point . Graphene hasn't really made it out of the lab , and we aren't capable of producing quantum computers that don't require a near 0 Kelvin Operating temperature , so don't count on that in the short term. Producing Consumer-grade quantum computers would require a breakthrough as revolutionary as the transistor , and we haven't seen it yet ...

 

 

 

[Coaxialgamer]

29 minutes ago, Moress said:

Both. Pascal/Polaris were big jumps in both raw processing power and lowering of power usage ( Ignoring Vega <.<  )

Vega is a number crunching power house . It is more efficient than polaris in that regard , with support for FP16 ( with 2 FP16 per clock ) , and Variable width SIMD . Problem is it doesn't translate well into gaming.

[Mira Yurizaki]

Yay, wall of text coming!

 

I'm now struggling to understand where people think every generation needs to have a significant jump in performance after taking a look at various reviews over the years. I went back to look at reviews of:

• Athlon XP vs. Athlon 64
• Pentium 4 vs. Conroe
• Phenom vs. Phenom II
• Penryn vs. Conroe
• Lynnfield vs. Penryn

This covers a good portion of the 2000s. The only time I saw a performance jump that people seem to be lusting for was at Pentium 4 vs. Conroe. However, I would argue most of that was because the Pentium 4 was such an inefficient architecture to begin with. Similarly with the performance jump from Bulldozer/Piledriver/Excavator to Zen. Everything else was within < 15% of an increase in performance on average when you compare it clock for clock.

 

Though the one thing I did notice that possibly helped contribute to performance jumps was the fact on average for most of the 2000s, the average clock speed of a processor was under 3.0GHz. During the last year of that decade, we were just breaking that speed for a midrange processor.

 

However, I won't deny that back in the 80s and 90s, there were huge jumps in performance every few years. But I would also argue those were due to major advances in hardware features in the CPU. Things like:

• Moving to microcodes that acts like RISC instead of using pure CISC
• Integrating L2 cache
• Superscalar pipelining
• Branch prediction
• Out-of-order execution
• Integrating SIMD processing

The last major improvement in the 2000s? The only two things I can think of are integrating the memory controller and multiple cores together. Neither really do much for IPC performance.

 

So why can GPUs continue to receive what amounts to 50%+ performance boosts per generation? Because two things:

• They use very simple "processor cores", making it easy to pack in more execution units together.
• They work on embarrassingly parallel problems that will effectively scale up to "impracticality"

Most of the applications most people tend to throw on a CPU tend to be I/O bound. For example, the web browser. You can make it as fast all you want, but if your internet connection sucks, it's going to perform just as bad.

[IKnight]

39 minutes ago, M.Yurizaki said:

Yay, wall of text coming!

 

I'm now struggling to understand where people think every generation needs to have a significant jump in performance after taking a look at various reviews over the years. I went back to look at reviews of:

• Athlon XP vs. Athlon 64
• Pentium 4 vs. Conroe
• Phenom vs. Phenom II
• Penryn vs. Conroe
• Lynnfield vs. Penryn

This covers a good portion of the 2000s. The only time I saw a performance jump that people seem to be lusting for was at Pentium 4 vs. Conroe. However, I would argue most of that was because the Pentium 4 was such an inefficient architecture to begin with. Similarly with the performance jump from Bulldozer/Piledriver/Excavator to Zen. Everything else was within < 15% of an increase in performance on average when you compare it clock for clock.

 

Though the one thing I did notice that possibly helped contribute to performance jumps was the fact on average for most of the 2000s, the average clock speed of a processor was under 3.0GHz. During the last year of that decade, we were just breaking that speed for a midrange processor.

 

However, I won't deny that back in the 80s and 90s, there were huge jumps in performance every few years. But I would also argue those were due to major advances in hardware features in the CPU. Things like:

• Moving to microcodes that acts like RISC instead of using pure CISC
• Integrating L2 cache
• Superscalar pipelining
• Branch prediction
• Out-of-order execution
• Integrating SIMD processing

The last major improvement in the 2000s? The only two things I can think of are integrating the memory controller and multiple cores together. Neither really do much for IPC performance.

 

So why can GPUs continue to receive what amounts to 50%+ performance boosts per generation? Because two things:

• They use very simple "processor cores", making it easy to pack in more execution units together.
• They work on embarrassingly parallel problems that will effectively scale up to "impracticality"

Most of the applications most people tend to throw on a CPU tend to be I/O bound. For example, the web browser. You can make it as fast all you want, but if your internet connection sucks, it's going to perform just as bad.

3

I mean, I understand that every generation only brings an incremental upgrade. Over time, the performance difference becomes substantial. And it's that lack of innovation in the last 10-20 years that you have pointed out that, that makes me question if we are 'moving' at the same pace (which from what I gather is a no).

 

(I should explain, I'm mostly asking this because, in 10-15 years, I want to build a 'current dream computer' using hardware like Titan XPs and Intel X edition CPUs (which will hopefully by then become obsolete)).

[Terryv]

5 hours ago, ImadKnight said:

With Moores law coming to a halt pretty soon, do you guys think that technology will keep becoming as obsolete at the same pace as it did from (let say) 1990-2010? 

It's only slowing because we're approaching the limits of silicon. Once we transition into something else (likely carbon based semiconductors), we will see larger leaps between generations and Moore's law may fall into place again.

[Zando_]

4 hours ago, ImadKnight said:

That's a good point, considering my lga1366 Xeon still keeps up well. 

 Off topic, but: @WhisperingKnickers, Xeon CPU Master Race FTW!

[IKnight]

16 minutes ago, Zando Bob said:

 Off topic, but: @WhisperingKnickers, Xeon CPU Master Race FTW!

Care to explain? 

 

18 minutes ago, Terryv said:

It's only slowing because we're approaching the limits of silicon. Once we transition into something else (likely carbon based semiconductors), we will see larger leaps between generations and Moore's law may fall into place again.

Aren't we still fairly far from transitioning from silicon?... 

[Zando_]

4 minutes ago, ImadKnight said:

Care to explain? 

 

Aren't we still fairly far from transitioning from silicon?... 

WhisperingKnickers is building up a massive LGA1366 Xeon collection, he keeps getting good deals on them. I really want them now, since a solid 3+ GHZ 6-core CPU for $50 is pretty attractive. I'm getting a 4c/8t Xeon and mobo from him as an upgrade from my 4c/4t i5 2400. 

[IKnight]

1 minute ago, Zando Bob said:

WhisperingKnickers is building up a massive LGA1366 Xeon collection, he keeps getting good deals on them. I really want them now, since a solid 3+ GHZ 6-core CPU for $50 is pretty attractive. I'm getting a 4c/8t Xeon and mobo from him as an upgrade from my 4c/4t i5 2400. 

Oh ok. I got mine for around €20.  Granted though, mine has been giving me problems with overclocking. Anything higher than 3.6ghz won't work, even with added voltage. And now that I've got new ram, anything about 3.2ish GHZ and it will refuse to post+only 2/4 ram sticks become detected. sigh.... 

[Zando_]

1 minute ago, ImadKnight said:

Oh ok. I got mine for around €20.  Granted though, mine has been giving me problems with overclocking. Anything higher than 3.6ghz won't work, even with added voltage. And now that I've got new ram, anything about 3.2ish GHZ and it will refuse to post+only 2/4 ram sticks become detected. sigh.... 

Huh. I know WK has a dead RAM stick, but I don't think his board supports overclocking, since it's a dual socket server board. The only dual socket ones that do are $500-$600. 

[NoxiousOdor]

7 minutes ago, ImadKnight said:

Oh ok. I got mine for around €20.  Granted though, mine has been giving me problems with overclocking. Anything higher than 3.6ghz won't work, even with added voltage. And now that I've got new ram, anything about 3.2ish GHZ and it will refuse to post+only 2/4 ram sticks become detected. sigh.... 

I havent overclocked due to most dual socket boards not allowing it though I might look into a modded bios in increase my base clock. Even without overclocking, I once saw my x5690s hit 3.9ghz while I was folding so that is exciting  

 

That is weird with the ram, the only issue I've had is that it won't detect one of my sticks but that could very easily be that one of my dimms is bad or I have a bad stick of ram

[Terryv]

28 minutes ago, ImadKnight said:

 

 

Aren't we still fairly far from transitioning from silicon?... 

Silicon has a theoretical minimum size f about 5-6 nanometers. Intel, nvida and AMD will have 7nm in the next 12-24 months. We're very close to the limits. Only way to cheat that is to increase core counts, but that only goes so far and CPU's get larger in the process.

[IKnight]

10 minutes ago, WhisperingKnickers said:

I havent overclocked due to most dual socket boards not allowing it though I might look into a modded bios in increase my base clock. Even without overclocking, I once saw my x5690s hit 3.9ghz while I was folding so that is exciting  

 

That is weird with the ram, the only issue I've had is that it won't detect one of my sticks but that could very easily be that one of my dimms is bad or I have a bad stick of ram

The (single CPU) board I have is really nice cause its got crossfire/sli support, USB3.0, overclocking support... I think that the overlocking problem for might just be bad luck with the silicon lottery. And it also looks like 2 of my RAM slots arent working (possibly due to bent pins on my MOBO?...).