flat assembler
Message board for the users of flat assembler.
 Home   FAQ   Search   Register 
 Profile   Log in to check your private messages   Log in 
flat assembler > Heap > AMD Ryzen, tide turns and probably time to get AMD?

Author
Thread Post new topic Reply to topic
sleepsleep



Joined: 05 Oct 2006
Posts: 6811
Location: ˛                              ⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣ Posts: 6699
AMD Ryzen, tide turns and probably time to get AMD?
https://www.forbes.com/sites/aarontilley/2017/02/22/amd-launches-ryzen-price-performance-win-intel/

Benchmarks show the Ryzen 7 1800X performs 9% better than an Intel Core i7 6900K. Perhaps even more crucially, however, that top-of-the-line AMD chip costs half as much as Intel's: $499 for AMD versus $1050 for Intel.

i think i only owned 1 amd desktop so far, and that was really long time ago, maybe intel, i kinda forget,

i rarely sell amd laptop or desktop, because they hardly last 2 to 3 years, well, intel desktop always last forever, at the time i type this, almost 96% of what i sold, 5 to 6 years old desktops still running without any problem / maintenance,
Post 27 Feb 2017, 12:37
View user's profile Send private message Reply with quote
revolution
When all else fails, read the source


Joined: 24 Aug 2004
Posts: 15089
Location: The Unicomplex
Yes, AMD cherry-picked benchmarks are all we need to see. Rolling Eyes

As always, test it with your workload before claiming superiority. Goes for anything, AMD, Intel, ARM, everything. Idea

Anyhow, this new shiny from AMD has not addresses the FPU performance, and apparently still won't compete with FP or DP dominated tasks. Sad

And for the average user, none of this matters even a bit. Almost no one fully uses the current batch of CPUs they already have. But, no matter, at least people will feel better if you tell them they have a faster chip, even if they have no idea how to verify that. Razz
Post 27 Feb 2017, 12:44
View user's profile Send private message Visit poster's website Reply with quote
ford



Joined: 21 Jan 2010
Posts: 102
IME, AMD has always done far better with multicore optimized loads. I have owned quite a few AMD machines since the multicore revolution as well as Intels. I have preferred the AMDs for most tasks, but audio work and gaming were always better on Intel.

That aside, I also just have a bias against Intel due to their underhanded/monopolistic business practices and price gauging.
Post 27 Feb 2017, 14:08
View user's profile Send private message Visit poster's website Reply with quote
ManOfSteel



Joined: 02 Feb 2005
Posts: 1133
This processing power war was a complete scam already a quarter century ago. And once the major players realized it was starting to become technically difficult/costly to continue at such pace, they resorted to another one: the multi-core scam.

Most people have never had to compile code, run a simulation, render complex high-poly 3D models or regularly work with audio/video processing.

The heavily-overpriced CPU and RAM gadgets are underutilized in practically every single consumer good computer on the planet.
Post 27 Feb 2017, 20:06
View user's profile Send private message Reply with quote
Furs



Joined: 04 Mar 2016
Posts: 748
Can't agree, I think multi cores are one of the best features of a CPU. In fact I find vectorized instructions (SIMD) much more of a "waste" than multi cores as it's even more ridiculous than multicores to program for properly. Why can't the CPU be smart enough to auto parallelize scalar code that has no dependencies? Instead of SIMD they could've multiplied the typical ALU/FPU units etc.

One of the nicer things about AMD compared to Intel is that they support ECC RAM in all processors, at least last time I emailed them about it. Intel only do that on Xeon chips and some i3.

Anyway multi-tasking without multicore CPU is such a pain and feels so sluggish. For "casual" applications I think a low-powered multicore CPU is still good (like Atom?). Better than a single core beast anyway.
Post 27 Feb 2017, 20:17
View user's profile Send private message Reply with quote
bitRAKE



Joined: 21 Jul 2003
Posts: 2624
Location: dank orb
If we are speaking about business applications - all local machines should be replaced by very thin clients: never have I seen a need for processing power on the business desktop (no matter how good at multi-taking the worker was). Of course, I'm excluding some areas of business: media processing, R&D, etc.

Games have usually tried to push the hardware (some even funded by processor designers) and entertainment has driven electronics sales for decades - computers are no exception.

Competition benefits us in many ways (despite the level of actual competition). There is a massive capital investment to build and advance processor fabrication. Real change takes years to reach market.

I like that AMD has shortened the pipeline and reduced branch penalties. Code that needs to be run on these processors is usually dynamic and will benefit from a processor that can adapt at many granularities. It will take some time to digest the oddities of this new design. It no doubt has it's niche.

_________________
The generation of random numbers is too important to be left to chance - Robert R Coveyou
Post 28 Feb 2017, 00:48
View user's profile Send private message Visit poster's website Reply with quote
revolution
When all else fails, read the source


Joined: 24 Aug 2004
Posts: 15089
Location: The Unicomplex
I'm still using my 14 year old single core Pentium-M Banias 1.3GHz. And I see no compelling reason to upgrade. The only things that are wearing out are the fan and the batteries. The batteries are replaceable (with some effort, but doable). But the fan, like all laptop fans it seems, is not generic, so replacing it when the need arises might be tricky.
Post 28 Feb 2017, 03:10
View user's profile Send private message Visit poster's website Reply with quote
YONG



Joined: 16 Mar 2005
Posts: 7912
Location: 22° 15' N | 114° 10' E

revolution wrote:
I'm still using my 14 year old single core Pentium-M Banias 1.3GHz. And I see no compelling reason to upgrade.

But it is a 32-bit processor. You can't do 64-bit programming with it, at least natively.

Wink
Post 28 Feb 2017, 03:47
View user's profile Send private message Visit poster's website Reply with quote
revolution
When all else fails, read the source


Joined: 24 Aug 2004
Posts: 15089
Location: The Unicomplex

YONG wrote:
But it is a 32-bit processor. You can't do 64-bit programming with it, at least natively.

I can't run 64-bit code natively. But I can write 64-bit code with fasm.

My travelling laptop has an AMD E350 1.6GHz CPU; which is 64-bit. And it is slower even though it is 8 years younger. I wouldn't even need the AMD system but my company insisted on getting new shiny ones because old stuff is obviously bad stuff according to the advertising.
Post 28 Feb 2017, 03:55
View user's profile Send private message Visit poster's website Reply with quote
Trinitek



Joined: 06 Nov 2011
Posts: 257
Even if it is cherrypicked, if the results are good enough to compare them to well performing i7's, then I think the real benchmark data should be promising.
Post 28 Feb 2017, 06:24
View user's profile Send private message Reply with quote
bitRAKE



Joined: 21 Jul 2003
Posts: 2624
Location: dank orb
It's rare for programmers to want as detailed of an understanding as they once did:
http://www.asmcommunity.net/forums/topic/?id=3547

Instead software is written to tune algorithms. Yet, without the understanding the tuning software operates with an insufficient model of the machine, and algorithm tuning takes place at a much larger granularity.

I remember when Digital went out of business, and AMD hired their CPU design team. Later I bought my first Athlon - over 15 years ago. Per cycle performance was impressive. I'd like to think AMD is heading back in that direction. Intel got sidetracked with the original P4.

_________________
The generation of random numbers is too important to be left to chance - Robert R Coveyou
Post 28 Feb 2017, 14:34
View user's profile Send private message Visit poster's website Reply with quote
revolution
When all else fails, read the source


Joined: 24 Aug 2004
Posts: 15089
Location: The Unicomplex

bitRAKE wrote:
It's rare for programmers to want as detailed of an understanding as they once did:
http://www.asmcommunity.net/forums/topic/?id=3547

Instead software is written to tune algorithms. Yet, without the understanding the tuning software operates with an insufficient model of the machine, and algorithm tuning takes place at a much larger granularity.

Things have become much more complicated now. With RAM running at its own speed, different sized caches, thermal throttling, L0/L1/L2/L3/L4 latencies, SMP (aka HT), multi-cores, multi-CPUs, multi-tasks, multi-threads, power saving modes, frequency boosts, OOE, micro-op fusion, hypervisors, VMs, 32/64 bit, etc., etc., etc. ...

So with all those things, and without a Herculean amount of effort, there is no chance we can know what is happening at any given time within the CPU. So we have little choice but to optimise the program for the system we test with. Trying to optimise some small snippet often leads to little overall value achieved because of the effects it has on other parts of the code that change their behaviour around it.

TL;DR Optimisation is hard. Sad
Post 28 Feb 2017, 15:03
View user's profile Send private message Visit poster's website Reply with quote
Furs



Joined: 04 Mar 2016
Posts: 748
Maybe optimize for the beauty of the code? (not readability, but beauty, hacks can be "beautiful" you know Razz)

After all, CPUs tend to optimize themselves for code in the first place (mostly tends to be HLLs though... bleh), so it's sort of a chicken-and-egg problem.

I find it completely idiotic that "complex" instructions are slow in a CISC CPU like x86, when the whole point of those instructions is to be specialized. Like specialized hardware. Because Intel cater to stupid garbage RISC HLL code.

Imagine you have an API, like Windows or Linux Kernel API, and they keep adding useful functionality or have APIs that perform common specialized tasks. Now they ask you to STOP USING those APIs and instead use only low-level and "simple" APIs to implement what those APIs do yourself. How retarded is that?

RISC mentality fucked logic.

Good thing though that certain "slow instructions to avoid" made a comeback (e.g. rep movsb) on newer CPUs. Also good that misalignment is not an issue anymore (except when it straddles a cache boundary). See, code using them did "the right thing" even when they were slow. That's what I mean with beauty of code.
Post 28 Feb 2017, 16:06
View user's profile Send private message Reply with quote
revolution
When all else fails, read the source


Joined: 24 Aug 2004
Posts: 15089
Location: The Unicomplex
I guess complex instructions can't be made fast. And since everyone is demanding faster clock speeds the designers have to make a choice, fast and simple or slow and specialised. And of course faster wins because who doesn't want a faster system? And as we know, no one writes in assembly anyway because the HLL folk have demonised at as some sort of archaic Satan worship (just look at AT&T syntax). So no more complex instructions for you. Twisted Evil
Post 28 Feb 2017, 22:02
View user's profile Send private message Visit poster's website Reply with quote
Furs



Joined: 04 Mar 2016
Posts: 748
Nah, complex instructions can obviously be made fast since it's just hardware. They just don't want to because HLLs don't use them. With "intrinsincs" these days they keep adding new complex instructions. What do you know, they're actually fast!

At worst, they can be made to be just as fast as the equivalent software sequence (of other instructions). In theory of course. They can be made slower by microcode and using only one execution unit because they don't find it "worth it" due to stupid HLLs.
Post 01 Mar 2017, 12:20
View user's profile Send private message Reply with quote
bitRAKE



Joined: 21 Jul 2003
Posts: 2624
Location: dank orb
Revision 1.1 of the Optimization Manual, and other Family 16h docs:

http://developer.amd.com/resources/developer-guides-manuals/

Feel free to view the instruction latency table directly in my Google Docs:

https://docs.google.com/spreadsheets/d/1GooHjClbfkRQIvshM8ESJSZEXSiZQvHjEtouTh5fNJk/edit?usp=sharing

And an interesting thread about performance character of the CPU:

https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/
Post 05 Mar 2017, 18:23
View user's profile Send private message Visit poster's website Reply with quote
Display posts from previous:
Post new topic Reply to topic

Jump to:  


< Last Thread | Next Thread >

Forum Rules:
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You can attach files in this forum
You can download files in this forum


Powered by phpBB © 2001-2005 phpBB Group.

Main index   Download   Documentation   Examples   Message board
Copyright © 2004-2016, Tomasz Grysztar.