AMD CPU speculation... and expert conjecture

Page 230 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.


Hajifur is however underestimating the Kaveri CPU. From "expert analysis", I can conclude it will be at about 2550K performance.

Hajifur is currently: this

We need this this to happen to him, so just start ignoring.
 
Ah... Gentoo... Bitter sweet memories...

Compiling code to "fit" your CPU is half the story, but still half is a lot, right? Anyway, to answer to you, noob, when you compile code, it's usually for yourself and to fit your needs; obviously. When benchmarking, a single binary for all uarchs is kind of dumb (if you can re-compile, that is) since you're not using the "best fit" for each individual CPU/uArch; it's even unfair IMO. There's a trade off as well in there: either use common denominator (SSE2/3/AVX) or a flag set that covers as much CPUs as you can (compile for PD and Sandy leave the rest as is, for example) with bloated binaries. Compiling, as the joke goes, takes a LOT of time for complex programs. I remember GNOME taking around 2-3 hours and the kernel 40mins (my old Athlon64). Imagine that for 4+ CPU uArchs and 12+ programs. That's a LOT of time.

The ideal world is very unfair and toxic, but it is there.

I think that's why openbenchmarking site exists for.

Cheers!
 

szatkus

Honorable
Jul 9, 2013
382
0
10,780


I wrote that 2 modules won't have performance like 4 cores with similar single thread performance. If you want something competetive with i5 (price, performance, average use) then wait for Steamroller FX.

 


Thanks for your useless input, here are some images.
04232455518l.png

Phenom_II_Chip_2.jpg

amd-fx-8350-piledriver-4.jpg

AMD_FX_8350_01.jpg

KL_AMD_Athlon_64_X2_Brisbane.jpg

th

 

8350rocks

Distinguished


I SAID KNOCK IT OFF ABOUT INTEL ALREADY!!!!

Now you get the wrath...I warned you:

Intel has constantly been trying to use under handed, unethical practices to cut AMD out of any and all business possible, see here:

http://www.dailytech.com/European+Commission+Fines+Inte...

http://www.edn.com/electronics-news/4313130/Intel-faces...

AMD has won 2 separate lawsuits against Intel and was a successful co-litigator in a 3rd:

In 1992, a court awarded AMD $10 million because Intel used business practices which violated industry ethical standards and were anti-competitive.

In 2009, a court awarded AMD 1.25 billion because Intel was back to the same old shenanigans and the same laws were broken, yet again. Additionally, this case required Intel to remove it's "Genuine Intel" CPUID check from the ICC Compiler...which has still not been done to this day over 4 years later.

http://en.wikipedia.org/wiki/AMD_v._Intel

http://are.berkeley.edu/~sberto/AMDIntel.pdf

Intel has a history of providing biased benchmarks as part of their "propaganda" as it's come to be colloquially called among industry insiders...one such example is the recently unveiled AnTuTu benchmark scandal...that is but one of many.

Intel's process is Tri-Gate, which is a complicated process...but they use bulk wafers to produce it, instead of SOI. This means they must take a few extra steps in the production process (extra mask layers) in order to make the chips. This reduces yields, and results in an inferior quality of product with a greater variety of results seen in the capability of the CPU to achieve clockspeeds. This is why all of their Tri-Gate CPUs are a crap shoot for overclockers.

Intel has undertaken every possible endeavor to cheat their way through keeping their competition out of the race, and have been caught and called out numerous times the fact that there is anyone even willing to buy their products boggles my mind, as the company truly is the evil empire. They paid Dell computers $6 billion for them to not offer an AMD product in their PCs when the Athlon was a superior product to Intel in every possible way. There are several other examples of this with other manufacturers.

No amount of performance per watt discussion will change anyone in this thread's mind about what kind of company Intel is, and so you can kindly shut-up about Intel from this point forward. Keep your Intel rants to yourself, all you are doing here is racking up post count and driving up the number of pages of this thread because, frankly, you're really starting to irk the people who read this thread for productive information about AMD.

Now please...go do something else!!!
 




Simply copy and paste his response to everything hafijur says, problem solved. 8350rocks is not a troll, he is in defense.
 


Lets have some fun, shall we?
Phenom_II_x6_die_clear_48078C.jpg

system-block-diagram.gif

9a0d86d7_4-6ghz-load.jpeg


 

jdwii

Splendid


I disagree with you 110% everything he said is true and its over with now and just an excuse now its been some time now and Amd still is losing market share and their still not as good as Intel CPU's. And yes i'm a Amd fan but even i know this.
 

jdwii

Splendid
I fine it actually sad that people do support company's like that and that is annoying but Amd isn't perfect either they lie ALL the time and always over estimate their products which is a bad thing.
 

cowboy44mag

Guest
Jan 24, 2013
315
0
10,810


Meant GDDR5 - got to typing too fast ;)
 

cowboy44mag

Guest
Jan 24, 2013
315
0
10,810
Once and for all, can we keep the posts on topic? If you truly believe that Intel processors are 30 years ahead of AMD, fine that's your opinion. Go find an Intel fan page and brag about power usage and your vast supremacy for being better than everyone else and buying such a great Intel processor. You guys want a pat on the back for having Intel Inside? Well here it is:

Congratulations your processor is the greatest thing ever invented in the history of the universe and we should all start a religion to worship its grand and glorious design, and you must be a god yourself for having one..... Really I can't be anymore sarcastic than that, but I hope you get the point. Please go find an "Intel humpers" thread, this isn't the place to be posting about Intel's grand supremacy.

Now FOR THE LOVE OF ALL THAT IS HOLY does anyone have any Steamroller news? Any AMD news of any kind?
 


Some AMD news:

http://www.tomshardware.com/news/Rory-Reed-2Q-2013-Xbox-PLayStation-Temash,23597.html

http://www.tomshardware.com/news/AMD-Gaming-Evolved-Games-Logo-Partners-Titles,23560.html


 

cowboy44mag

Guest
Jan 24, 2013
315
0
10,810


Thank you!! Some actual AMD news on an AMD thread:bounce:
 

GOM3RPLY3R

Honorable
Mar 16, 2013
658
0
11,010
I kind of skipped around, so I thought that for now i'll just expand on the 3930k vs the 9590.

Completely agree that the 9590 is overpriced. That being, even at 4.7 Ghz may get close to or even on par with a 3930k stock, but the OC abilities on the 9590 will definitely be more limited than the 3930k. Consider per core performance, then multiply how much extra OCing there will be. I'd say with a stock water cooling system that is in push, but using cool air from the outside, could get a 3930k up to 5.0 Ghz @ ~1.27 - 1.28 V-core, maybe even 5.2 Ghz but max. The 9590, considering that AMD is known for heating up greatly under loads, even if it has more than enough cores, if you are encoding video, there's a 100% chance that it'll be over 50 % usage total, which will prompt great heat. Thus being said, the 9590 on that cooling system may make it to 6.2Ghz, probably 6.5 Ghz the max before a superior cooling solution will have to be used. Even if it will be a 1.5-1.8 Ghz improvement, the 3930k will also have similar improvements, at a ~ lower heat signature.

On another note, bringing into perspective the power draw, which many people bringing up, is being misused. That point is certainly valid, but only when talking about the performance.

CPU X gets 10 points of work done in Time A with power @ 5
However CPU Y gets 6-7 points of work done in Time A with power @ 7-8

It's really showing that with what amounts of power (according to standard clocks and V-cores), the system can use. In terms of the product being used for total power usage, it will come in concern with laptops. HOWEVER, if it is used in the sense that a processor is better at it's job than the other, then that's the valid term.

It's a very complex way of thinking about it, but essentially I can understand what these people are saying. They just are showing that, as an example: Person A gets full and satisfied by eating 1 meal, but person B needs 2 meals to be satisfied.

Thus I must say, people, stop saying this. The intention that is given by the conspicuous, can only be understood by the unique, otherwise, it is seen as ignorant.
 


The issue AMD has to sort out with Steamroller is really getting on with single-threaded performance in order to compete with Intel in that segment. I admit, because you seem to be a laptop user, you care alot about wattage. Why Intel got rid of soldering the IHS to the CPU, we will never know. Why the IHS should even be on overclocker-oriented products in the first place is the real question ;) I like that you seem to have a more constructive approach to stating the issues of AMD's current products.
 

cowboy44mag

Guest
Jan 24, 2013
315
0
10,810


You know, that is the pure bitch of it. Intel needs AMD and AMD needs Intel for consumers to have a cost effective good product. If AMD would have never become a rival processor producer Intel chips would be much higher priced and we would probably be talking about that powerful new dual core as we wouldn't have quad cores. If Intel were to go out of business tomorrow AMD might keep its prices down for awhile, but with no competition they would jack their prices as well, there wouldn't be any incentive at R&D so advancements would stagnate. Both companies need each other to give us the best products at the best prices.
 

cowboy44mag

Guest
Jan 24, 2013
315
0
10,810


+1
 

Cazalan

Distinguished
Sep 4, 2011
2,672
0
20,810
This is a bit old but good to know AMD continues advancing their SDKs to take advantage of their APUs/GPUs.

AMD CodeXL, Bolt and AMD APP SDK
http://developer.amd.com/community/blog/latest-versions-of-amd-codexl-bolt-and-amd-app-sdk-now-available/

OpenCV-CL: Computer vision with OpenCL acceleration
http://developer.amd.com/community/blog/opencv-cl-computer-vision-with-opencl-acceleration/


 

mlscrow

Distinguished
Oct 15, 2010
71
0
18,640


I have two theories on why Intel got rid of the soldering process:

1) Some non-enthusiast analyst employed by Intel saw an opportunity to make him or herself look better to their higher-ups by saving the company x millions of dollars per x number of chips produced by abandoning their patented soldering process and going with a less expensive TIM application instead. He/she saw that the chip would still work under normal conditions and perhaps he/she would not be overlooked for a promotion that year as a result.

Why they kept the crappy TIM on their overclock oriented K versions is just plain dumb, but still, less expensive.

2) Pretty much the same as the first theory, except instead of a non-enthusiast analyst, it's a super enthusiast analyst who realized that any like-enthusiast will just de-lid the damn thing and apply phase-change/liquid metal alloy to achieve better thermal conduction to the IHS anyway (e.g., Liquid Metal Pro). It won't be as efficient as the solder, but it'll be the best you can get without it, at least while continuing to use the IHS.
 


You are talking about perfect case scenarios which differ because no chips are the same. I have seen very few Sandy E's do 5ghz without hitting 1.4v ranges and spiking heat displacement most average out at 4.5ghz before they reach instability. I did a youtube recording once with three identical i5 2500K's in the exact same builds running simultaneously and the most was 4.8ghz the one barely ran 4.5ghz stable. I have also shown that a month of running a 2500K at its limits is enough to kill the chip and everyone by now or those that do sub zero cooling that Intel is notorious for its cold bug, at very low temperatures a Intel CPU can instantly die. Since no intel records are really set on maximum cores yet AMD have legitimate 8ghz results on 4 cores and even a certified 8ghz on a FX8350 one can conclude that the extra TDP on the 9590 was designed to break records.

The long and the short is that the FX9590 is not a benching part, it is not a gaming part it is a pure enthusiast overclockers part, one that knows and is experienced in keeping CPU's cool.

If you take stock of everything yes Intel have lowered TDP but to say their thermals are under control is a gross exaggeration, a 4770K on stock TIM roasts at 4ghz and since a Intel chip is more susceptable to heat degradation and failure that is quite severe and despite running down the process and lowering TDP's the heat effect is actually magnifying rather than discipating. Noob2222 had posted links of 14nm then Broadwell DT prototypes having severe thermal degredation and now we know that Broadwell DT is about as much of a mystery as to whether it will ever happen. Its just that Intel hide it and reviewers don't really exploit the inner workings just punch the numbers.

On the AMD front the FX9590 can probably do 5.3ghz on a good cooler endorsed to cool it, it is designed to be withing the threshold and the difference between a 4.7 and 5.3 FX9590 is not linear, some benches are showing 15% or higher improvements.

 

Ranth

Honorable
May 3, 2012
144
0
10,680

(I'm adding one)
3) Intel extreme series need to be more compelling to the enthusiasts so by putting bad TIM on the normal parts 3770k/4770k might make the overclocking enthusiast upgrade and get a 3930k/4930k instead as they will overclock better due to the solder. (AFAIK Ivy-e will be soldered.)
 
OK here is my latest AMD speculation and non-expert conjecture. If you look at the past the Phenom II x4 is more competitive now against the newest i3s than it was against 1st gen i3s. I would expect this to be true in 2 years with the FX 83xx series vs i5. I think its undeniable Intel have an advantage in performance per core and per core per clock, AMD makes up for this with more cores at higher clocks and lower prices but there is still many applications that prefer the more powerful cores despite being less of them. Intel annoyingly bring in a new chipset with nearly every new CPU line despite slow performance advances since sandy bridge which I don't see being justified but at least there chips do at least move forward every time (arguably with Haswell) unlike Bulldozer.
As for steamroller I would not like to even guess on its performance since Bulldozer was so far behind what I expected, but it must be a step forward whatever. I am worried about the lack of confirmation there will be a 8 core FX chip at all, but we all need to realise the enthusiast market is only a tiny part of the overall market that both Intel and AMD could live without, even the home desktop market is not huge compared to laptops, tablets and corporate desktop markets. This I think is the main reason Intels chips are getting smaller performance gains and they are designing then for Ultrabooks and to scale to lower power not for best performance in desktops.
If AMD could increase CPU performance like they do GPUs from 1 gen to the next then they would be years ahead of Intel, 3 years ago to get a balanced gaming PC you had to spend 2-3 times more on the GPU than the CPU now its like 1-1.5 times.
 
This I think is the main reason Intels chips are getting smaller performance gains and they are designing then for Ultrabooks and to scale to lower power not for best performance in desktops.

Ding ding ding.

Intel is trying, desperately, to better position themselves in the mobile market. Hence the lower power signature at lower clock speeds compared to IB, the heavily upgraded GPU, the Atom re-design, and the like.

In those regards, Haswell has done EXACTLY what Intel has wanted it to do. Understand the Desktop is not a prime developmental focus anymore, and Intel's decisions make a lot more sense.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


GCC is not the best in an absolute sense. Other compilers can give better code. Check this example where CLANG produces much much better code

http://openbenchmarking.org/embed.php?i=1305170-UT-LLVMCLANG75&sha=ff347ed&p=2

But you ignore the point (as usual in your anti-AMD rant). It is not who prepared what. The point was that tests cited are not still showing the full potential of the FX-8350 because:

with the current GCC release there isn't much improvement out of the "bdver2" optimizations for this processor that should expose the CPU's BMI, TBM, F16C, and FMA3 capabilities over the original AMD Bulldozer processors.

 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790


I agree completely with your conclusion. Above, I showed a x264 benchmark of A10 trinity vs several i3, i5, and i7. Extrapolating, we obtain Kaveri CPU will be between the i5-2500K and the i5-3470. This fits very well in your 2550k prediction!

I am preparing a collection of graphs with estimated Kaveri CPU performance in a series of tests. I will share when ready.

I think we would move away from CPU to try to estimate which will be GPU performance of Kaveri.
 
Status
Not open for further replies.