No Mo' Moore's - Time for new Architectures and Software Languages

[HunterAP]

34 minutes ago, leadeater said:

That power usage is intentional though, it isn't so much a feat of engineering and technology that we have been able to stick to the same power usage but an actual design and technology limitation that they are unable or unwilling to exceed, mostly both. There's only so much heat, therefore power, that materials can take before they break down.

That was what I meant by the almost non-existent power increase in the chart: processors tech has advanced well enough without generating substantially more heat or consuming much more power. Even if more power and more heat is unfeasible for every day use, it's a technical feat of engineering that those metrics haven't risen so substantially in the past few years.

2 minutes ago, leadeater said:

The other problem I see with more ASICs is memory and data flows, the big advantage to a big powerful CPU with a lot of instruction sets is everything stays in one place all with access to the same memory controlled by the same memory controller. Once you start moving things off the CPU to ASICs you have to do protocol translations to get data, memory in reality, to the ASIC which has a big overhead along with hitting other limitations like bus bandwidths.

 

That's why NVLink was developed, it addresses one part of the problem which is bus bandwidth but it still doesn't address the translation issue.

 

For there to be a big uptake in ASICs there will need to be a big shift in how hardware and memory is handled within computers, Gen-Z is one of those things I have been following with a lot of interest that seeks to solve this.

 

The focus has always been processor centric, what the source is talking about and people generally have conversations about when we need to start thinking more memory centric and start trying to remove the closed off silos of resources all talking their own language (hardware/memory wise).

https://www.nextplatform.com/2017/11/21/breaking-memory-free-compute-gen-z/

https://www.nextplatform.com/2018/05/17/forging-composable-infrastructure-for-memory-centric-systems/

https://www.nextplatform.com/2018/02/15/gen-z-interconnect-ready-restore-compute-memory-balance/

I was thinking that processors would go more towards the route of having a single ASIC within a system that would be used for all commands that this system would need, but different systems would have their own ASIC rather than a CPU with lots of instructions. Think of it like an i3 and and i5 having pretty much nothing in common because they would have different instruction sets entirely, and therefore an i3 would be really good at something like browsing and doc stuff, while an i5 could do the same things and pretty much the same performance but have other instructions that the i3 would be lacking, like having better video gaming or video editing capabilities on an instruct-set level.

[leadeater]

1 minute ago, HunterAP said:

it's a technical feat of engineering that those metrics haven't risen so substantially in the past few years.

I see it more like the speed limit on the road, not a feat but a limit that we are not allowed to exceed. What they have been able to achieve within that limit is quite feat but that fact we have not gone above it is more just complying with that limit.

 

A lot like racing technology rules, they put limits on power so they increase down-force, that becomes too fast and dangerous so they put limits on aerodynamics etc. I just don't see the limit itself as being the feat.  

[GoodBytes]

Software optimization is always there.

The problem is that companies and even open source projects, wants fast and easier deliver of features/products, while having support for security and reliability. So in other words, you get to use libraries and frameworks. Why code yourself from the ground up a parsing system of text file, for say program configurations, when you can use a simple API call to fetch you the info you need from the files done for you and supported by another group of people (say Microsoft), and at the same time, use other available APIs that come with it for other things?

 

Software has gone way more complex compared to the old days, but a lot of optimization is still done, just focused on other things.

For example, very early web browser only handled HTML, pretty much. Then Java-Script came along, but it was simple, could not do much. Now, we have complex web browser with full on, highly optimized Java-Script engines powering countless websites, such as this forum for many elements you interact with. If no optimization on Java-Script is done, then website load or interact very slowly. If a lot of optimization is done, then it would far perform better, and be more responsive. In the old days, the big concern was page downloading, now we have rendering performance and accuracy (including auto-fixing web dev errors) of complex looking pages with many images and CSS for styling, and on top of that JavaScript.

 

If you want to make a web browser from scratch, well first of all, good luck, by the time finish, new web technology would have come in. Possible in the old days, not possible now. Not alone, that is, especially if you aim it to be competitive, and not half broken and requires a server farm to load a 1996 website that was professional made initially, and never updated, and now look like a a high school kid has done it while learning HTML by himself.

 

 

Anyway, what I wanted to say, is that you focus on performance and only performance. Back int he old days, we didn't care about power efficient. Now we do. Big time. It is all about how much performance we can get for the given performance. If we can get the chip faster, then great! An added bonus. Another key thing is the lack of competition on the CPU and GPU side of things. In the old days you had high competition with many small players. Now, it is nice that AMD finally rise up with a product that is making Intel panic. But AMD needs to remain highly competitive, and not fall out like they did with its AMD Athlon 64 and 64 X2 CPUs on Socket 939. As for Nvidia, well, Pascal (GeForce 1xxxx) was released in 2016. We had the same GPU until today, and today we have a price increase on a GPU that doesn't bring much on the table. AMD is not on the map, and you have no one else. Back in the old days, you have: ATI, Nvidia, 3dfx, Matrox on the gaming world side of things.

[Doobeedoo]

Well yeah we've hit GHz wall on silicon and we're nearing how much we can shrink it. Also multi-core is said to only scale to a certain degree and also adds complexity to inter communication. 

Also a ton of software is still heavily single threaded, but no doubt there needs to be better optimization on that side. Even video games. Devs be lazy. 

Who knows what future chips will be, new material, being able to shrink it further with much higher clockspeed. Software needs to follow as well of course. 

[Stefan Payne]

20 hours ago, Ryan_Vickers said:

My understanding was that transistor count was what doubled.  That appears to be on track.  I know it's often quoted as performance doubling but I'm not sure it was ever intended like that.

Yep, was only about Transistor Count, nothing else.

[asus killer]

20 hours ago, The Benjamins said:

All this proves is that single core is capped out, we have to focus on multi core performance.

you will always have a bottleneck somewhere with multi cores. It can only go so far.

[S w a t s o n]

The reality is that 2 things need to happen.

 

1. We need to rip off the band-aid that is x86 and basically allow for old programs to no longer be able to physically run on our systems without literal x86 emulation. Basically I'm saying we need to kill x86-64. It's fundamentally a tired, stretched thin architecture from the late 70's. 64bit extension from AMD was a band-aid on the band-aid.

 

2. We need to start the graphene meme, Or at least move away from traditional doped silicon. Chiplets and such are a band-aid to basically cope with the sheer horror that is making huge dies at nodes so small that we're literally bumping into quantum physical barriers.

[Coaxialgamer]

21 hours ago, Curufinwe_wins said:

Amdahl's law everyone...

Amdahl's law is kind of inappropriate for evaluating the potential speedup brought by a parallel system . Amdahl's law plots the theoretical speedup of a task given an increase in parallel resources with a portion of said task being parallel , where speedup = (1/(1 − p)) , with p being the portion of the task that is serial in nature.

But the problem is that the law calculates the speedup in latency of a given task as resources improve . This simply isn't what happens . As resources improve , we run more intensive tasks on our processors. These more complex tasks have larger datasets that have additional parallelism in them .

 

Gustafson's law is arguably more useful in this day and age , as it plots the speedup of a task of a specific execution time , ie how much more work can we throw at our computers so that they execute that task in a given time. We get larger speedups with Gustafson's law than predicted with Amdahl's law.

 

Plus , the chart above only plots single threaded performance . Multi-thread performance have kept pace in optimized workloads. And while the easy ILP is pretty much already exploited , we can keep improving our single core performance . It's just not practical . Believe it or not , most of what limits the exploitation of ILP right now is actually our branch predictors . With 100% accuracy ( which isn't possible , but for the sake of argument ) , we would only be limited by how far ahead we can look in code ( ROB and RS are the limiters here ). with 100% accuracy , we could be executing instructions speculatively 1000 instructions ahead of time .

 

 

[Coaxialgamer]

2 hours ago, S w a t s o n said:

1. We need to rip off the band-aid that is x86 and basically allow for old programs to no longer be able to physically run on our systems without literal x86 emulation. Basically I'm saying we need to kill x86-64. It's fundamentally a tired, stretched thin architecture from the late 70's. 64bit extension from AMD was a band-aid on the band-aid.

Is it though ? Yes, x86 is old and esoteric . But is that a big problem ? The power and area cost of an x86 decoder isn't really significant with modern manufacturing (back in the k8 days it was only a couple % , and that was on 130nm). You aren't going to save a lot of power by using a RISCV or ARM decoder instead. The legacy compatibility is a huge advantage however .

And yes , you do need to support the register files for the obsolete extensions such as MMX and x87 , but the actual cost of that is ridiculously small given how small these are . Apart from that , every x86 instruction  is broken down into µops at the decode stage , which are then reordered and renamed by what is essentially a RISC core . It's not like MMX support requires it's separate functional units . All the extensions of a given type (FP or int ) share the same functional units , so re-use of hardware is high.

[S w a t s o n]

2 minutes ago, Coaxialgamer said:

Is it though ? Yes, x86 is old and esoteric . But is that a big problem ? The power and area cost of an x86 decoder isn't really significant with modern manufacturing (back in the k8 days it was only a couple % , and that was on 130nm). You aren't going to save a lot of power by using a RISCV or ARM decoder instead. The legacy compatibility is a huge advantage however .

And yes , you do need to support the register files for the obsolete extensions such as MMX and x87 , but the actual cost of that is ridiculously small given how small these are . Apart from that , every x86 instruction  is broken down into µops at the decode stage , which are then reordered and renamed by what is essentially a RISC core . It's not like MMX support requires it's separate functional units . All the extensions of a given type (FP or int ) share the same functional units , so re-use of hardware is high.

I think we need a big change like photonic processors or something and at that point you're probably just going to abandon anything remotely similar to any existing architecture. We're talking the jump from vacuum tube to transitors

[HunterAP]

9 hours ago, S w a t s o n said:

The reality is that 2 things need to happen.

 

1. We need to rip off the band-aid that is x86 and basically allow for old programs to no longer be able to physically run on our systems without literal x86 emulation. Basically I'm saying we need to kill x86-64. It's fundamentally a tired, stretched thin architecture from the late 70's. 64bit extension from AMD was a band-aid on the band-aid.

 

2. We need to start the graphene meme, Or at least move away from traditional doped silicon. Chiplets and such are a band-aid to basically cope with the sheer horror that is making huge dies at nodes so small that we're literally bumping into quantum physical barriers.

I definitely agree with this, the problem is I only see articles discussing new ideas/testing for graphene or some other "super silicon" that's supposed to be much better for computing, but the testing is in the the extremely early alpha levels, so who knows how long until we'll actually see them become mainstream or even a small public testing (IE: sent to reviewers).

[williamcll]

The only way out is nano/quantum processors. There is only so much you can do with software optimizations.

[Thaldor]

There's actually a lot of stuff that should be relearned from the past and optimization is a keyword here. In my opinion we probably could get very far with the processors we already have if someone was willing to go through assembly codes and really start to do same kind of optimization that was commonly done in the 80's and really think what is the best way to performe some task. These skills were abandoned when the processor speeds started to rise and prices dropped, you didn't need to think if it was better to do a while-loop or for-loop to gain better performance, if your code didn't work, just throw money at it and get better CPU.

 

You can see the very same happening right now at this moment on the other side of computing. No one really cares about how much RAM or disk space is used. If your software requires 16GB RAM you just tell it and if customers don't have that, you tell them to throw money at the problem. You don't think about does you game require 50GB or 100GB of disk space just because you can easily get it to look good with enourmous textures and extremely complex and precise meshes with little to no work, because it's far more cheaper to tell customers to get more disk space than optimize your textures and meshes so they look almost exactly the same but take up a lot less space (for example, you can get a lot of HW demand off from Fallout 4 with some optimized textures and other optimization mods, Bethesda did very lazy job to make Fallout 4 look questionably better than Skyrim).

[bowrilla]

On 21.9.2018 at 8:45 PM, rcmaehl said:

It originally was performance doubling with the transistors but that died around late 2000s and people have been hoping we'd recover, we haven't.

Still, the Moore's law only made prediction about the complexity gain in integrated circuits. That may have meant an equal performance gain back 20 or 30 years again but it was not what Moore's law predicted.

 

We're still on track for now.

[bowrilla]

On 23.9.2018 at 4:01 PM, williamcll said:

The only way out is nano/quantum processors. There is only so much you can do with software optimizations.

I'm sorry, but quantum computers won't do much in general computing. Please read something about how quantum computers work, how they differ from binary computers and why there's nothing to gain with classic binary computing tasks. Quantum computers just to run Crisis, Fortnite or Word is absolute nonsense.

[LeSheen]

On 21-9-2018 at 9:45 PM, Deus Voltage said:

Linux presents to the casual and enthusiast pc communities a paradox that I don't see anyone discussing in great depth.

 

One of the reasons why Linux is so great is because you have far less risk of having a virus than running a windows machine. 

If you want to feel extra secure, just download clam (or clam tk for graphical interface).

 

Linux destros also tend to be far lighter, another big plus. 

 

The problem (the paradox), as I alluded in the beginning of my comment, is that if Linux becomes sufficiently popular, and enough people migrate to it, then more viruses will be made that tailor to Linux destros. 

 

So on the one hand, Linux is great, we love its open source/ light/ and almost virus free nature. But at the same time, we risk losing some Linux's benefits if it is no longer as exclusive. 

 

Maybe I'm rambling, I don't know.

 

 

I disagree the "internet" runs primarly on linux. Its estimated to run 96.5 percent on linux. There are far more devices out there running on linux.

One of the big issues in my opinion is the lack of knowledge from the end user. Linux is mostly used by professionals, while our granparrents and kids probably use windows. Yes there will be more virusses for linux if it becomes mainstream. But the security level of linux will probably stay at the same level.

[AlTech]

This is news?

[RexinOridle]

Some tasks are simply sequential. They can't be parallelized. Like you need one calculation done before starting the next one.

[Alt-F4]

Moore's Law came to and End with the release of AMD's 2990WX and EPYC Processors; 

[mr moose]

On 9/27/2018 at 1:19 AM, LeSheen said:

I disagree the "internet" runs primarly on linux. Its estimated to run 96.5 percent on linux. There are far more devices out there running on linux.

One of the big issues in my opinion is the lack of knowledge from the end user. Linux is mostly used by professionals, while our granparrents and kids probably use windows. Yes there will be more virusses for linux if it becomes mainstream. But the security level of linux will probably stay at the same level.

Given the market share of windows versus Linux at a user level, I would say that is arather erroneous statement.   Windows is used by everyone while 9% are enthusiasts or work specifically in computer science.  And yes, if Linux became mainstream it would be effected by malware/viruses to about he same  degree as any other OS for it's market share.  That is just a par for the course. 

[sof006]

On 21/09/2018 at 7:40 PM, mate_mate91 said:

Tell that to MS who release windows 10. On fresh install it has 150 services, 8.1 had about 50 and 7 had about 45 in my experience and i have reinstalled those about 50 times each! On my and friends and relatives PCs. Lately i just install linux on everyones PCs.

Also Nvidia who does not make good drivers for linux and does not make them open source.

Also many other companies. For example my thinkpad t440s had dolby digital home theater software that improved sound quality very much and it's not available for linux and they will not open source it.

 

Proprietary software is very bad! Look at linux it's open source and it became so good that almos whole internet runs on linux, 90% of smartphones run on linux, smart watches run linux, supercomputers run linux and 2.5% of desktop runs on linux too.

If game developer companies were not so lazy and create more games for linux Desktop segment would run on linux too and that would be big win for humanity. Windows 10 is full of spyware and it needs fast hardware. It's using way more resources then an OS should.

 

Look at games. Those are so unoptimized that they use 6 GB VRAM, 10 GB RAM and so on. Not so long ago i played crysis 3 on ultra setting on a 2 GB GPU +2 GB RAM + i7 3770. Now many games will crash on those specs. Hell windows 10 takes almost 1.5 GB RAM just to start. My arch linux uses 170 MB RAM on fresh start and i have more usefull services running then windows 10 has.

 

So yes open source software  = win for all humanity!

Games use a lot of VRAM usually because textures are getting bigger and bigger as well as the size of games in general. Games tend to need to store more in VRAM so the GPU has faster access to them without needing to load them into memory first.

[Linus_torvalds]

25 minutes ago, sof006 said:

Games use a lot of VRAM usually because textures are getting bigger and bigger as well as the size of games in general. Games tend to need to store more in VRAM so the GPU has faster access to them without needing to load them into memory first.

Compare any modern game with 2013 Crysis 3. Crysis 3 will have better visuals then many modern games. But i remember playing crysis 3 on ultra/high settings on my GTX 660 2 GB + 2 GB RAM. Then try to start those modern games on that specs

[sof006]

1 hour ago, mate_mate91 said:

Compare any modern game with 2013 Crysis 3. Crysis 3 will have better visuals then many modern games. But i remember playing crysis 3 on ultra/high settings on my GTX 660 2 GB + 2 GB RAM. Then try to start those modern games on that specs

I haven't played Crysis 3 in a long while but I can think of a lot of games now which probably look much better than Crysis 3. Looking at games like Red Dead 2, Horizon Zero Dawn etc. Crysis is known for just putting as much tech into a game as possible. If I remember correctly Crysis 3 went over the top with tesselation.

[GoldenLag]

10 hours ago, Alt-F4 said:

Moore's Law came to and End with the release of AMD's 2990WX and EPYC Processors; 

How did you get to that conclusion?

 

Also people seem to treat "Moore's Law" as some holy grail. Its really just an estimate made decades ago.

 

The epyc 7601 32 core at 19,2 billion transistor. Its about wyere we would expect from a doubling of transistorcount. No if next Epyc series doubles again (from what we know it seems like it) its not that far off.

 

I am assuming 17 billion transistors in 2017.