Does AMD has some future?

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


When you say "WE" you mean AMD right?

I don´t know man, you been awfully wrong in the past... you keep talking about meeting with Rory constantly, yet you did not knew he was about to take a hike in the Company?

Well if you really work for AMD and or if you have access to the Top Execs be sure to tell them this:

They are doing everything WRONG, their "MOAR CORES" approach is idiotic, efficiency over bruteforce.
 


Linus rants/flames are well-known

http://linux.slashdot.org/story/13/07/15/2316219/kernel-dev-tells-linus-torvalds-to-stop-using-abusive-language

http://highscalability.com/blog/2014/12/31/linus-the-whole-parallel-computing-is-the-future-is-a-bunch.html

Linus just doesn't understand that both the single and multi-core era have finished and that the only way that engineers have to significantly increase performance in future is via parallelism on manycores

hsa_history_large.jpg


Not only Linus is ridiculously negating the utility of manycores, but he has repetitively avoided my question about how he expect engineers will extract performance gains in future.

He has tried to intimidate me a pair of times before, but he is so plain wrong and full of it that my reply to him went the sarcastic way

http://www.realworldtech.com/forum/?threadid=146066&curpostid=146834
 


I recall you said 20nm and other stuff now proven wrong. But in any case, I will not discuss with you. I will simply repeat AMD claim about desktops:

With regards to your specific question, we expect Carrizo will be seen in BGA form factor desktops designs from our OEM partners. The Carrizo project was focused on thermally constrained form factors, which is where you'll see the big differences in performance and other experiences that consumers value.
Click to expand...

We will probably see Carrizo in NUC-like boxes.
 
Well, algorithms scale poorly most of the time, so taking advantages of a full paradigm shift (bazillion cores) will take a lot of years.

In particular, I'm still curious as to how the ASM paradigm will be shifted for the bazillion core era. You feed the CPUs instructions in serial manner, right? Am I missing something there?

Cheers!
 



He's talking over your head. You're lucky he responds to you at all. He has more important things to do, you know, like work on the Linux kernel.
 


There are studies about how to run a serial workload over several cores to improve their performance. Even highly serial algorithms saw speed up of 8x running on 64 different cores in a manycore. But there are limits to this approach.

Algorithms with large amounts of parallelism are split into threads with each running on a different core.

If you feed the CPU with a thread, what the CPU does is to identify the hidden parallelism on the stream of instructions (the sequence of ASM) and then execute the instructions on parallel to speed up. The problem is that ordinary CPUs logic (superscalar/OOOE) spend too power and die area to identify the hidden parallelism in a sequential stream of instructions and don't scale up. Haswell can execute a maximum of eight instructions at once but sustained average is about 3 or 4 due to the logic being limited to a 192 ROB window. To achieve eight instructions sustained rate the ROB would be increased at 1000--2000 entries, but the area/power penalty is superlineal. and Haswell is already a big/complex core.
 


The point is precisely that he doesn't respond at all. He ignores the technical questions because (I know) he has no answers. He then tries to hide that by writing large rants where he calls crazy to everyone who disagrees with his narrow vision of the world. He called crazy to the people working in runahead and other alternatives to OOOE, latter he called crazy to the people working in manycores, and now he is calling crazy to the people behind the Mill machine.
 


I can see that and undestand why, but you still feed the CPUs (no matter how many there are) in a serial fashion. I want to know if that is a design restriction of CPUs or OSes. That's why you can do parallel work on serial loads, because you get the "illusion" of parallelism. Strictly speaking, a serial algorithm can be made parallel in chunks, but the atomic operation that come out of that will still be feed sequentially to the CPU(s) in assembler/binary. I think I'm missing memory management a bit on my thinking... Ugh, I'll have to go back and read memory management again (pointers and registers).

Cheers!
 

That doesn't really mean much until we see performance figures though?
 


SiS have been never a good source of information about performance. Their benchmarks are just synthetics and no information about base/turbo clocks etc.

Also I'm reading also leaks about Intel's CPUs and Skylake seem to be design full of cool, new features and IPC similar to Haswell/Broadwell. AVX512 is one of these feature, not interesting for gamers.
 


R9 290X is 290W. Whatever, they can just lower clocks a little to decrease TDP.
 



Like lol? Then they better do it if its so easy. Pretty sure Nvidia is doing that already at an accelerated rate. Really will you people think this product is so amazing if its like 30% faster then a 980 while consuming twice the power consumption. Now i pulled that 30% number out of thin air it could be twice as fast to, i don't know.

Edit as a side project i'll be selling my board and CPU and getting an unlocked pentium dual core to play around with i'm actually going to compare the benchmarks between the fx 8350. Of course i expect the pentium to lose badly in most cases. But in some games i expect to see improvements more specifically in emulators were dolphin will drop to the mid 40's in SSB with one of my cores at 90+%. However i'll be getting a 4790K soon as well as a 970, unless of course Amd comes out with something that uses the same power that is just as powerful.

What's amazing is i can upgrade my parts and not have to worry about power consumption meaning i can keep my old psu. I'll probably buy a new one just because PSUs lose efficiency each year and i don't want my antec to kill my PC.
 


I never said I met with Rory...

I did say I met with John Byrne, who is either departing, or recently departed. I still have contacts at AMD, guys in the labs, etc. No one as high up as John though.
 


This depends largely on the code.

The x86 instruction set has always been largely serial in nature, because it was designed to be, and because integer calculations (large numbers of the work done on CPUs) are mostly serial operations. Anything that can be done in parallel is more efficient once offloaded to the GPU for faster parallel processing using many small cores to do a massively parallel task.

Now, what you are going on about is only tangentially relevant, because everyone with a brain knows cache is the limiting factor for code branches being processed. It is also the most expensive portion of the die in terms of area consumption.

However, improved branch prediction logic could easily squeeze more out of the available cache we have now. In fact, if AMD could improve that quite a bit, that would be some proverbial "low hanging fruit" in terms of the fact that they already will have the space dedicated to cache on the die anyway. In terms of actual execution, well they would have to come up with a better way to decrease missed predictions and increase efficiency in the pipelines. FAR easier said than done...however, in terms of improving CPU performance within the same or smaller footprint, branch prediction is a big area where you can make up a lot of ground, and increase power efficiency with the same stroke.
 


The guy who started this thread had been beating a dead horse for the past year. Regurgitating the same tired garbage Ad nauseam. He even screwed up the the title. Does AMD have "some" future? I guess English is not his native language.
 


That is too funny. It reminds me of a few months ago at an overclocking forum a guy was actually trying to ague with the Stilt about overclocking AMD processors.
 


I don't know excact performance of that 300W (if it's actual TDP and not something else) part, but it's easy: http://upload.wikimedia.org/math/e/e/b/eeb06330735b1e13969c9b0549e6a7f4.png

They are doing it already with mobile GPUs, both AMD and Nvidia. Slightly lower clocks and much lower TDP.
 


I'm also from Europe (Juan is from Spain, I'm from Poland) and I was wondering if it's correct, thanks :)
 


It's not Linux Forum. That forum is about hardware, I'm reading it for a long time, there are people people designing actual SoCs, GPUs, CPUs (that guy who founded Mill CPU for example), also some software guys (like Linus). Really interesting place, I recommend you to read it :)
 


Half life 3 confirmed.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom