AMD CPU speculation... and expert conjecture

[noob2222]

I really hate to burst your bubble; however, in your earlier post...you equated AMD to 66% performance...

Its really annoying when people like you and noob, read a person’s statement in half, and then start jumping around without even understanding what it means!!

The amd being only 66% of an intel core value (ie, intel being 50% faster) was calculated for AAC encoding. To the best of my knowledge, for AAC encoding, this is a true value, and I STILL stand by it.
I challenge you to prove to me that the IPG ratio of Intel SB(or IB) , verses AMD PD, for AAC encoding, is less than 1.50 by a significant margin, atleast around 1.33, as per your claims. Can you??

again: Relying on one lopsided benchmark will give you your lopsided point of view.

Ive already shown that povray is only a 23% advantage to SB.
Lame mp3 with ? compiler on toms test is only 40% to IB, 32% to sb.

personally, your constant bringing up Itunes sounds like this:
you love Itunes AAC encoding ... it gives Intel its biggest advantage over AMD, therefore all other tests are invalid or ignored.

AAC encoding != all benchmarks. this is why its not worth arguing, it keeps going to "AAC AAC AAC". btw, Itunes isn't the only AAC encoder, just happens to be commonly benched with, so that arguement isn't even AAC, its Itunes, Itunes, Itunes.

Itunes and Cinebench are only 2 benchmarks around that show 50% or more lead for Intel. Why ignore the rest?

 

[gamerk316]

Which again breaks down to efficiency. You're comparing a refined architecture that's been improved many times, to 2nd gen architecture. The efficiency is going to be drastically better for intel. That's what it is...fewer miscalculations through the front end of the architecture allow intel to run more efficiently at the CPU level. Though I still do not accept your argument entirely on the grounds that systems were not optimized for the individual architectures.

I never said that; I said specifically, "assuming" the code paths are more or less the same. Only way to know for sure would be to decompile the program and go looking for processor macros. Its entirely possible the app is using an Intel codepath that is significantly more efficient, thus giving it better numbers. Likewise, I also pointed out (numerous times) that if you don't get 100% core scaling in multithreaded benchmarks, the numbers get skewed (though you can at least correct by using Task Manger in this case).

If Intel was 51% better than AMD...then how on earth does the FX8350 beat Intel's i5-3570k and i7-3770k at so many things? Is it just 175% better at certain things? How do you close the loop for your logic?

Clockspeed. IPC calculations remove clockspeed from consideration, but that extra 600+MHz per core? Yeah, that makes up for a LOT in terms of performance.

Ok, let's compare a wide range of benchmarks, shall we:

http://www.xbitlabs.com/articles/cpu/display/fx-8350-83...

WinRAR 4.2:
FX8350: 68 seconds
i7-3770k: 57 seconds (+19%)
i5-3570k: 83 seconds

TrueCrypt 7.1a
FX 8350: 259 MB/s (+13%)
i7-3770k: 229 MB/s
i5-3570k: 171 MB/s

iTunes:
FX8350: 185.1 seconds
i7-3770k: 126.8 sec (+49%)
i5-3570k: 130.2 sec

Adobe Lightroom 4.2:
FX8350: 294.5 sec.
i7-3770k: 259.1 sec (+13%)
i5-3570k: 272.8 sec

Adobe Premiere ProCS6:
FX8350: 602
i7-3770k: 624 (+3.6%)
i5-3570k: 520

x264 encoding (1st pass):
FX8350: 62.07 fps
i7-3770k: 74.16 fps (+19.4%)
i5-3570k: 55.23 fps

x264 encoding (2nd pass):
FX8350: 15.16 fps (+4.1%)
i7-3770k: 14.55 fps
i5-3570k: 12.1 fps

3ds max 2011(CPU):
FX8350: 8.28
i7-3770k: 9.76 (+17.8%)
i5-3570k: 8.78

3ds max 2011 (render):
FX8350: 6.34
i7-3770k: 6.85 (+8.0%)
i5-3570k: 6.01

http://www.anandtech.com/bench/Product/697?vs=551

POV_RAY 3.7RC6 Multithreaded:
FX8350: 1504.4 (+10.3%)
i7-3770k: 1363.6

POV_RAY 3.7RC6 Singlethreaded:
FX8350: 252.1
i7-3770k: 312.4 (+23.9%)

Cinebench R10 Singlethread:
FX8350: 1.1
i7-3770k: 1.66 (+50.9%)

Cinebench 11.5 multithread:
FX8350: 6.89
i7-3770k: 7.61 (+10.1%)

POV-RAY 3.7 Beta 23 SMP:
FX8350: 5008
i7-3770k: 5408 (+7.9%)

Microsoft Excel 2007 SP1 Monte Carlo:
FX8350: 12.6
i7-3770k: 10.0 (+26%)

7-zip:
FX8350: 23407 (+2.6%)
i7-3770k: 22810

Doesn't look so drastically offset anymore does it? Hmm...that's interesting...so the FX8350 keeps within 10-20% of the i7-3770k in nearly everything but extremely singlethreaded applications with very specific uses? It also destroys the i5-3570k at nearly everything...how odd...but I thought...the i5-3570k was supposed to be the second coming or something...funny, huh?

Trying to figure out your conclusion here. I see one significant (>10%) win for FX (TrueCrypt 7.1a) and, even throwing out the signlethreaded benchmarks, 5 >10% wins for Intel (WinRAR 4.2, iTunes, Adobe Lightroom 4.2, x264 encoding (1st pass), 3ds max 2011(CPU), POV_RAY 3.7RC6 Multithreaded). Throw in the single threaded benchmarks, and you end up with a rout.

And that's the problem, the same exact one I pointed out back when BD was in its early design stages: The majority of work is going to be done on a limited number of threads, so IPC + Clockspeed is a greater performance factor then number of cores. Yes, FX "hangs around" due to a clock speed advantage, but how long is that going to last do you think? What happens when Intel starts releasing chips at 3.8GHz stock and throws a +10% IPC gain? Woops, Intel takes a giant performance lead again, and AMD needs to re-design their arch to be more IPC friendly. FX is basically banking on the fact Intel won't be able to close in clockspeed, and given Intels inherent process edge, I really doubt that will be the case.

Simple test: Clock an i5 at 3.8GHz and retest. FX suddenly won't look nearly as attractive. Thats the danger AMD is walking into: If Intel closes in clockspeed, AMD is done.

(Also, does anyone else REALLY hate the descending quotes in the new forums? Can we at least get an option to disable those when quoting someone?)

 

[de5_Roy]

Spoiler

Which again breaks down to efficiency. You're comparing a refined architecture that's been improved many times, to 2nd gen architecture. The efficiency is going to be drastically better for intel. That's what it is...fewer miscalculations through the front end of the architecture allow intel to run more efficiently at the CPU level. Though I still do not accept your argument entirely on the grounds that systems were not optimized for the individual architectures.

I never said that; I said specifically, "assuming" the code paths are more or less the same. Only way to know for sure would be to decompile the program and go looking for processor macros. Its entirely possible the app is using an Intel codepath that is significantly more efficient, thus giving it better numbers. Likewise, I also pointed out (numerous times) that if you don't get 100% core scaling in multithreaded benchmarks, the numbers get skewed (though you can at least correct by using Task Manger in this case).

If Intel was 51% better than AMD...then how on earth does the FX8350 beat Intel's i5-3570k and i7-3770k at so many things? Is it just 175% better at certain things? How do you close the loop for your logic?

Clockspeed. IPC calculations remove clockspeed from consideration, but that extra 600+MHz per core? Yeah, that makes up for a LOT in terms of performance.

Ok, let's compare a wide range of benchmarks, shall we:

http://www.xbitlabs.com/articles/cpu/display/fx-8350-83...

WinRAR 4.2:
FX8350: 68 seconds
i7-3770k: 57 seconds (+19%)
i5-3570k: 83 seconds

TrueCrypt 7.1a
FX 8350: 259 MB/s (+13%)
i7-3770k: 229 MB/s
i5-3570k: 171 MB/s

iTunes:
FX8350: 185.1 seconds
i7-3770k: 126.8 sec (+49%)
i5-3570k: 130.2 sec

Adobe Lightroom 4.2:
FX8350: 294.5 sec.
i7-3770k: 259.1 sec (+13%)
i5-3570k: 272.8 sec

Adobe Premiere ProCS6:
FX8350: 602
i7-3770k: 624 (+3.6%)
i5-3570k: 520

x264 encoding (1st pass):
FX8350: 62.07 fps
i7-3770k: 74.16 fps (+19.4%)
i5-3570k: 55.23 fps

x264 encoding (2nd pass):
FX8350: 15.16 fps (+4.1%)
i7-3770k: 14.55 fps
i5-3570k: 12.1 fps

3ds max 2011(CPU):
FX8350: 8.28
i7-3770k: 9.76 (+17.8%)
i5-3570k: 8.78

3ds max 2011 (render):
FX8350: 6.34
i7-3770k: 6.85 (+8.0%)
i5-3570k: 6.01

http://www.anandtech.com/bench/Product/697?vs=551

POV_RAY 3.7RC6 Multithreaded:
FX8350: 1504.4 (+10.3%)
i7-3770k: 1363.6

POV_RAY 3.7RC6 Singlethreaded:
FX8350: 252.1
i7-3770k: 312.4 (+23.9%)

Cinebench R10 Singlethread:
FX8350: 1.1
i7-3770k: 1.66 (+50.9%)

Cinebench 11.5 multithread:
FX8350: 6.89
i7-3770k: 7.61 (+10.1%)

POV-RAY 3.7 Beta 23 SMP:
FX8350: 5008
i7-3770k: 5408 (+7.9%)

Microsoft Excel 2007 SP1 Monte Carlo:
FX8350: 12.6
i7-3770k: 10.0 (+26%)

7-zip:
FX8350: 23407 (+2.6%)
i7-3770k: 22810

Doesn't look so drastically offset anymore does it? Hmm...that's interesting...so the FX8350 keeps within 10-20% of the i7-3770k in nearly everything but extremely singlethreaded applications with very specific uses? It also destroys the i5-3570k at nearly everything...how odd...but I thought...the i5-3570k was supposed to be the second coming or something...funny, huh?

Trying to figure out your conclusion here. I see one significant (>10%) win for FX (TrueCrypt 7.1a) and, even throwing out the signlethreaded benchmarks, 5 >10% wins for Intel (WinRAR 4.2, iTunes, Adobe Lightroom 4.2, x264 encoding (1st pass), 3ds max 2011(CPU), POV_RAY 3.7RC6 Multithreaded). Throw in the single threaded benchmarks, and you end up with a rout.

And that's the problem, the same exact one I pointed out back when BD was in its early design stages: The majority of work is going to be done on a limited number of threads, so IPC + Clockspeed is a greater performance factor then number of cores. Yes, FX "hangs around" due to a clock speed advantage, but how long is that going to last do you think? What happens when Intel starts releasing chips at 3.8GHz stock and throws a +10% IPC gain? Woops, Intel takes a giant performance lead again, and AMD needs to re-design their arch to be more IPC friendly. FX is basically banking on the fact Intel won't be able to close in clockspeed, and given Intels inherent process edge, I really doubt that will be the case.

Simple test: Clock an i5 at 3.8GHz and retest. FX suddenly won't look nearly as attractive. Thats the danger AMD is walking into: If Intel closes in clockspeed, AMD is done.

(Also, does anyone else REALLY hate the descending quotes in the new forums? Can we at least get an option to disable those when quoting someone?)

i (ab)use the spoiler tag to hide walls of text. it still doesn't make replying easier though, as your reply has noob's id but you're clearly replying to someone else. new quoting system has higher chance of mis-edits, lol.

 

[sarinaide]

I don't like the new system at all, if it ain't broke don't fix it. B)

 

[$hawn]

Totally agree!!! It's very confusing searching through walls of cascaded quoted text
In fact, I actually like the whole site the old way

 

[$hawn]

again: Relying on one lopsided benchmark will give you your lopsided point of view.

Ive already shown that povray is only a 23% advantage to SB.
Lame mp3 with ? compiler on toms test is only 40% to IB, 32% to sb.

personally, your constant bringing up Itunes sounds like this:
you love Itunes AAC encoding ... it gives Intel its biggest advantage over AMD, therefore all other tests are invalid or ignored.

AAC encoding != all benchmarks. this is why its not worth arguing, it keeps going to "AAC AAC AAC". btw, Itunes isn't the only AAC encoder, just happens to be commonly benched with, so that arguement isn't even AAC, its Itunes, Itunes, Itunes.

Itunes and Cinebench are only 2 benchmarks around that show 50% or more lead for Intel. Why ignore the rest?

In case you haven't read my previous post addressed to you, I have already apologized for using a single single-threaded test. But again, as proved already, on an average, Intel is 30 to 50% faster in the IPG ratio game.

While doing my initial Centurion vs SB-E guestimate, I've already stated that audio encoding is my favourite single core test. Here was my quote,

All these numbers I've put up are a result of reading numerous comparison articles, some common sense and a little guess work. Do NOT treat this as scientific data!!

If you want to know how i arrived at the 50% faster value, have a look at this AAC encoding chart (audio encoding is my favourite single core performance measuring yardstick).

This was never meant to prove that AMD s**ks vs Intel or anything. It was people like you who just read 50% and then went crazy!!
In fact, I actually wanted to prove that even in the worst benchmark for AMD, a 5GHz centurion chip would theoretically stand neck to neck with a SB-E. If I took a benchmark that favoured AMD, I would get called an AMD fanboy once again. That was what I was trying to avoid.

Here's a few quotes that were appearing before my post, on Page 51,

griptwister said:
Anyways, I'm curios to see how a 5Ghz chip compares to a 4 Ghz chip when people say they don't notice a performance improvement past 4.5Ghz on the 8350 chip. Also, how will this chip compete against Intel? Especially at that price point. I just don't see this chip beating a hexicore i7, even at 5 ghz. Unless they somehow changed the singlethread performance...

truegenius said:
imo fx8350 needs ~6GHz to match i7-3970 on stock in multithreaded tasks
that is why it is priced below i7-3960/70

 

[gamerk316]

Anyways, I'm curios to see how a 5Ghz chip compares to a 4 Ghz chip when people say they don't notice a performance improvement past 4.5Ghz on the 8350 chip. Also, how will this chip compete against Intel? Especially at that price point. I just don't see this chip beating a hexicore i7, even at 5 ghz. Unless they somehow changed the singlethread performance...

This one is simple to explain: You eventually hit a point where the CPU is able to get all its work done without any significant periods where the thread(s) for the application are not able to be run due to CPU bottleneck issues [EG, the work gets done before the OS preempts the thread]. At that point, RAM, HDD, and other parts of the system become a much larger performance bottleneck [EG: Your thread is waiting for data to get to the CPU, and can't run until it is done].

In short, the higher you clock the CPU, the more other components will bottleneck you. Hence why you don't get linear performance increases as you improve IPC/Clockspeed.

 

[8350rocks]

I never said that; I said specifically, "assuming" the code paths are more or less the same. Only way to know for sure would be to decompile the program and go looking for processor macros. Its entirely possible the app is using an Intel codepath that is significantly more efficient, thus giving it better numbers. Likewise, I also pointed out (numerous times) that if you don't get 100% core scaling in multithreaded benchmarks, the numbers get skewed (though you can at least correct by using Task Manger in this case).

Benchmarks like Cinebench that show favorable leads by intel are compiled with ICC, this is common knowledge. I really wish AMD would release a compiler...not sure how itunes is compiled, but I would tend to think it's ICC as that would explain the drastic outlier. POV RAY in single threaded is a 40% difference and multithreaded is a significantly smaller gap...closer to 23%

Clockspeed. IPC calculations remove clockspeed from consideration, but that extra 600+MHz per core? Yeah, that makes up for a LOT in terms of performance.

Intel's process will not be able to close the gap in clockspeed, their bulk wafer use will continue to plague them when it comes to factory clocks. Also, the Intel chips out typically overperform their stated clockspeeds by a 100-200 MHz...so your argument isn't really all that strong.

Trying to figure out your conclusion here. I see one significant (>10%) win for FX (TrueCrypt 7.1a) and, even throwing out the signlethreaded benchmarks, 5 >10% wins for Intel (WinRAR 4.2, iTunes, Adobe Lightroom 4.2, x264 encoding (1st pass), 3ds max 2011(CPU), POV_RAY 3.7RC6 Multithreaded). Throw in the single threaded benchmarks, and you end up with a rout.

Against the venerable i7-3770k it "hangs around" Against the i5-3570k, which most intel people claim destroys the 8350, it's a clear rout in favor of AMD...all day...every day. So, while it merely "hangs around" with the CPU that's 175% as expensive, it simply outperforms the CPU that's "only" 135% as expensive. I would say the value for the dollar is exceptionally high. It's like getting a $275 CPU for $180...that sounds really good to me.

And that's the problem, the same exact one I pointed out back when BD was in its early design stages: The majority of work is going to be done on a limited number of threads, so IPC + Clockspeed is a greater performance factor then number of cores. Yes, FX "hangs around" due to a clock speed advantage, but how long is that going to last do you think? What happens when Intel starts releasing chips at 3.8GHz stock and throws a +10% IPC gain? Woops, Intel takes a giant performance lead again, and AMD needs to re-design their arch to be more IPC friendly. FX is basically banking on the fact Intel won't be able to close in clockspeed, and given Intels inherent process edge, I really doubt that will be the case.

Again, back to efficiency. Steamroller is going to be a 30% gain in efficiency for AMD...let me just repeat that 30% gain in efficiency How large was that gap again? Roughly 40%? How much improvement was Haswell looking to be? 6-7%?

To me, that looks like AMD will be within 15-20% on the most lopsided benchmark...which means, they will likely meet, or overtake, intel on everything else. That's a company, 1/17th the size of intel, handing it's best 4 core CPUs a recipe for crow. Plus, without going to SOI or some other higher grade wafer, intel won't close the gap for clockspeed. Besides, even if they somehow managed to do that, with wafer/process improvements, SR is a 38% improvement in hardware IPC capability over PD. Going from 32 IPC to 44 IPC is going to be an enormous difference in singlethreaded performance.

Add in that GloFo just announced 3D stacking capabilities on it's 20nm process, which is the planned jump for excavator...and this is really starting to look bad. Can you imagine a chip 1/2 the size of what we have now with twice the transistors? Intel doesn't have that technology...they have been trying to get it right for the last 10 years, and they don't have the capability to stack yet. That's what happens when you get complacent and rest on your laurels and underestimate the little guy.

Simple test: Clock an i5 at 3.8GHz and retest. FX suddenly won't look nearly as attractive. Thats the danger AMD is walking into: If Intel closes in clockspeed, AMD is done.

You'd only essentially be gaining 200 MHz on the i5...most stock i5-3570k's run about 3.5-3.6 95% of the time, due to turbo. So, if you disabled turbo and OC'ed to 3.8 you wouldn't really gain much of anything. That's another intel trick, and this has been shown numerous times in benchmarks. They first observed this with SB back a couple years ago when they first hit.

(Also, does anyone else REALLY hate the descending quotes in the new forums? Can we at least get an option to disable those when quoting someone?)

Yes, the system for this is annoying, I didn't see how it was before...but it is a bit of a cluster *%$&.

 

[mayankleoboy1]

This one is simple to explain: You eventually hit a point where the CPU is able to get all its work done without any significant periods where the thread(s) for the application are not able to be run due to CPU bottleneck issues [EG, the work gets done before the OS preempts the thread]. At that point, RAM, HDD, and other parts of the system become a much larger performance bottleneck [EG: Your thread is waiting for data to get to the CPU, and can't run until it is done].

In short, the higher you clock the CPU, the more other components will bottleneck you. Hence why you don't get linear performance increases as you improve IPC/Clockspeed.

Assuming you have a SSD, what other bottlenecks would be here ? Most probably RAM ?

 

[de5_Roy]

i'll just chip in a tiny li'l bit. correct me where i'm wrong.

Benchmarks like Cinebench that show favorable leads by intel are compiled with ICC, this is common knowledge. I really wish AMD would release a compiler...not sure how itunes is compiled, but I would tend to think it's ICC as that would explain the drastic outlier. POV RAY in single threaded is a 40% difference and multithreaded is a significantly smaller gap...closer to 23%

there's no getting around the fact that intel's own compiler favors intel (obviously). being the giant evil corporation they are, they grab every advantage they can. i won't justify intel's practices. but. until amd makes their own compiler and pushes for optimizations for their own cpus, crying foul every time some software rumored to be compiled by icc is kinda useless. there's also the (rumored) point that icc produces faster performance for amd cpus than other compilers. why not get the devs compile on non-intel compilers and then bench? shouldn't be that hard for amd or hsa foundation or cpa. intel maintains hardware and (unfair) compiler advantages, that's that.

Intel's process will not be able to close the gap in clockspeed, their bulk wafer use will continue to plague them when it comes to factory clocks. Also, the Intel chips out typically overperform their stated clockspeeds by a 100-200 MHz...so your argument isn't really all that strong.

what that says is that intel has better turbo and amd doesn't. both fx and sb/ib cpus have auto-overclocking turbo features.

Against the venerable i7-3770k it "hangs around" Against the i5-3570k, which most intel people claim destroys the 8350, it's a clear rout in favor of AMD...all day...every day. So, while it merely "hangs around" with the CPU that's 175% as expensive, it simply outperforms the CPU that's "only" 135% as expensive. I would say the value for the dollar is exceptionally high. It's like getting a $275 CPU for $180...that sounds really good to me.

you really shouldn't throw around prices like that when prices may change everyday. checked newegg, fx8350 is going for $200 and 3570k is for $221.
i am not gonna go into what destroys what. benchmarks can be made to favor either way. i can find gaming benches where fx8350 beats a 3570k across the board as well as multithreaded benches (if that's what one calls 'destroying').
although, if you're gonna pit 3570k/3770k vs 8350, solely cpu vs cpu, the fx loses instantly because it has no igpu. no amount of cheapness can fix that. :lol: well.. may be you could run the fx box using remote desktop... i think most of the am3 mobos with built in gfx don't support fx... there was something about old am3 mobos not supporting fx or something... i don't remember well.. may be remote desktop can help.
assuming [strike]both intel and amd[/strike] the amd pc (see it's not solely about cpus anymore) has discreet gfx (only for the sake of comparing the cpu performance), fx can gain it's usual upperhand and such. still, the cost of a discreet card significantly lowers fx's perf/dollar.
sensationalizing goes both ways.

Again, back to efficiency. Steamroller is going to be a 30% gain in efficiency for AMD...let me just repeat that 30% gain in efficiency How large was that gap again? Roughly 40%? How much improvement was Haswell looking to be? 6-7%?

that 30% number is sorta vague. if we take 30%, assuming amd achieves perfect scaling, sr should be 30% more efficient(perf/watt) than bulldozer (not pd) according to their own cpu roadmap. then again, amd hasn't uttered the word 'steamroller' in months. they call their sr-based apu kaveri and never mentions what kind of cpu core will be in it.

To me, that looks like AMD will be within 15-20% on the most lopsided benchmark...which means, they will likely meet, or overtake, intel on everything else. That's a company, 1/17th the size of intel, handing it's best 4 core CPUs a recipe for crow. Plus, without going to SOI or some other higher grade wafer, intel won't close the gap for clockspeed. Besides, even if they somehow managed to do that, with wafer/process improvements, SR is a 38% improvement in hardware IPC capability over PD. Going from 32 IPC to 44 IPC is going to be an enormous difference in singlethreaded performance.

Add in that GloFo just announced 3D stacking capabilities on it's 20nm process, which is the planned jump for excavator...and this is really starting to look bad. Can you imagine a chip 1/2 the size of what we have now with twice the transistors? Intel doesn't have that technology...they have been trying to get it right for the last 10 years, and they don't have the capability to stack yet. That's what happens when you get complacent and rest on your laurels and underestimate the little guy.

if i remember glofo's recent process roadmap, their focus is on ultramobile socs (their future 14nm-xm node's - xm stands for extreme mobility, not 'we'll pwn skymont with this awesomeness'), not high performance high end cpus(amd fx). and there's no mention of 3d(...?) for 20nm node. i could be wrong in reading the roadmap, perhaps someone with better knowledge can explain it better.
http://semiaccurate.com/2013/02/08/global-foundries-lays-out-their-finfet-plans/
http://www.xbitlabs.com/news/other/display/20120920113524_Globalfoundries_Accelerates_Roadmap_Expects_First_14nm_Tape_Outs_in_2013.html
good thing is, this will help amd fiercely compete in mobile sector (jaguar's successors).
tbh, i haven't seen anyone underestimate amd. they came from behind and build llano and brazos, made gcn and are a major proponent of hsa. if any one says bad things, it's (ironically) the amd fanboys.

 

[JAYDEEJOHN]

OK, first, for all you newby boys out there, this isnt new, this "new" quoting system is exactly like the older quoting system, and when it changed people hated the "new" quoting system sheeeeeesh

Secondly, the new AMD chips a moving forwards in improvements at a good clip, I imagine many will be pleased and or surprised, but wont be passing Intel overall anytime soon, unless its for certain uses, and thats all I have to say.
Someone once told me, hope for the best, expect the worse when it comes to HW, and I hope everyone sees how good this approach can be

 

[gamerk316]

Benchmarks like Cinebench that show favorable leads by intel are compiled with ICC, this is common knowledge.

And...can you prove that Intel specific code paths are in use for the given application? What compiler switches are in use?

Intel's process will not be able to close the gap in clockspeed, their bulk wafer use will continue to plague them when it comes to factory clocks. Also, the Intel chips out typically overperform their stated clockspeeds by a 100-200 MHz...so your argument isn't really all that strong.

I'm not convinced, given how well SB/IB OC (Intel lockdowns aside, obviously). A +1GHz OC on a 2500k isn't exactly uncommon...

Against the venerable i7-3770k it "hangs around" Against the i5-3570k, which most intel people claim destroys the 8350, it's a clear rout in favor of AMD...all day...every day. So, while it merely "hangs around" with the CPU that's 175% as expensive, it simply outperforms the CPU that's "only" 135% as expensive. I would say the value for the dollar is exceptionally high. It's like getting a $275 CPU for $180...that sounds really good to me.

Ok, great. Now OC both CPUs to their max and bench again. The k series is an unlocked part after all, and theres no reason to get one and NOT OC. Do that, and you end up with a i5-3570k @ ~4.7GHz and a FX-8350 ~@4.8GHz. Guess what? Not a landslide anymore.

Again, back to efficiency. Steamroller is going to be a 30% gain in efficiency for AMD...let me just repeat that 30% gain in efficiency How large was that gap again? Roughly 40%? How much improvement was Haswell looking to be? 6-7%?

Despite the fact neither Intel or AMD have EVER hit their stated efficiency gains each generation? Do people always fall for the same things over and over again?

That 30% is going to be a best-case value. I figure 10-15% is more likely in real-world apps.

 

[sarinaide]

Intel/AMD/Nvidia often stipulate target projections, whether they hit them or not comes down to the ingenuity involved in achieving it. While yes it is naive to believe it outright, its almost as naive to undermine or under play targets because I think a little more often than not they do get within fair gains to said targets. Example: AMD's richland desktop touted 20-40% official from AMD (link is above) the leaks out have shown generally 15-40% depending on the workload, so 15% lower limit meaning they came within 5% of the target low limit, most often its around 25% gains, this is a major kudos to AMD.

 

[gamerk316]

Intel/AMD/Nvidia often stipulate target projections, whether they hit them or not comes down to the ingenuity involved in achieving it. While yes it is naive to believe it outright, its almost as naive to undermine or under play targets because I think a little more often than not they do get within fair gains to said targets. Example: AMD's richland desktop touted 20-40% official from AMD (link is above) the leaks out have shown generally 15-40% depending on the workload, so 15% lower limit meaning they came within 5% of the target low limit, most often its around 25% gains, this is a major kudos to AMD.

And the other end of the spectrum: Bulldozer.

 

[sarinaide]

What about it?

 

[Cazalan]

Simple test: Clock an i5 at 3.8GHz and retest. FX suddenly won't look nearly as attractive. Thats the danger AMD is walking into: If Intel closes in clockspeed, AMD is done.

With Intel focusing on low power that's not a concern this year at least. Haswell SKUs are known and they're the same exact speeds as Ivy.

 

[JAYDEEJOHN]

Simple test: Clock an i5 at 3.8GHz and retest. FX suddenly won't look nearly as attractive. Thats the danger AMD is walking into: If Intel closes in clockspeed, AMD is done.

With Intel focusing on low power that's not a concern this year at least. Haswell SKUs are known and they're the same exact speeds as Ivy.

OK, I have to add to this. Increasing their portfolio wouldnt be an easy thing to do here, as their larger clients are locked into what they expect at a given price point.
Reducing the number loses monies as well, and would then make the new higher rated/clocked cores more expensive etc.
Such a drastic move would be seen as a true monopolistic move, and the consequences wouldnt be pretty

 

[Cazalan]

there's no getting around the fact that intel's own compiler favors intel (obviously). being the giant evil corporation they are, they grab every advantage they can. i won't justify intel's practices. but. until amd makes their own compiler and pushes for optimizations for their own cpus, crying foul every time some software rumored to be compiled by icc is kinda useless. there's also the (rumored) point that icc produces faster performance for amd cpus than other compilers. why not get the devs compile on non-intel compilers and then bench? shouldn't be that hard for amd or hsa foundation or cpa. intel maintains hardware and (unfair) compiler advantages, that's that.

Could always go to Linux benchmarks that are compiled with gcc to remove the bias. You can find many here: http://www.phoronix.com

AMD should just work closer with Microsoft and add optimizations. AMD really doesn't have the resources to make a compiler from scratch. It's a massive undertaking.

 

[Cazalan]

Again, back to efficiency. Steamroller is going to be a 30% gain in efficiency for AMD...let me just repeat that 30% gain in efficiency How large was that gap again? Roughly 40%? How much improvement was Haswell looking to be? 6-7%?

The front end improvements to the instruction decoder will certainly help a lot but you have to be careful here with the numbers. In situations where the front end wasn't the bottleneck there will be little improvement. The execution units aren't changing much. That's why AMD can't say there will be an across the board X% increase.

If you correlate it to a gaming benchmark they'll be able to bump their min fps but the max fps won't change much.

 

[JAYDEEJOHN]

http://browser.primatelabs.com/geekbench2/1591657
http://browser.primatelabs.com/geekbench2/1852876

 

[de5_Roy]

Could always go to Linux benchmarks that are compiled with gcc to remove the bias. You can find many here: http://www.phoronix.com

exactly. compile with gcc, remove the bias and then bench and see what happens. there won't be any room for excuses.
it makes me wonder though, what if amd gave up and decided to live with the current crop of compilers?

AMD should just work closer with Microsoft and add optimizations. AMD really doesn't have the resources to make a compiler from scratch. It's a massive undertaking.

really? i didn't know that. if amd can do all these thousand other things, they should be able to make a compiler... or pick an open source one and work on that. amd's software support needs a lot of improvement.

 

[8350rocks]

And...can you prove that Intel specific code paths are in use for the given application? What compiler switches are in use?

ICC still has the built in cripple_AMD function...don't believe me? Look at agner...he is a coder and a big proponent of AMD, the dispatcher/switches are not the issue, it's the internal flag that offers a less optimal code path. Yes it compiles faster code for AMD than MSVC, but it's also much, much faster for intel...so the advantage gained by AMD is well overcome by the huge advantage intel has. You're talking a 20% difference in coding efficiency....20%!!!

I'm not convinced, given how well SB/IB OC (Intel lockdowns aside, obviously). A +1GHz OC on a 2500k isn't exactly uncommon...

So all the 1.0+ GHz OC's on FX8350's are what then? Cause there's a ton of them as well...

Ok, great. Now OC both CPUs to their max and bench again. The k series is an unlocked part after all, and theres no reason to get one and NOT OC. Do that, and you end up with a i5-3570k @ ~4.7GHz and a FX-8350 ~@4.8GHz. Guess what? Not a landslide anymore.

That would be an FX8350 @ 8.17 GHz running all 8 cores vs. i7-3770k @ 7.16 GHz running 1 core...do you really want to run that benchmark?

If you want to talk practical...the highest OC FX8350 running stable benches I have seen is 5.7 GHz and change...I haven't seen an intel break 5.0 yet with stability...so we're still 1 GHz advantage running OC'ed benchmarks...which actually...in all fairness...is a disadvantage to intel, as opposed to running at stock clocks where the gap is a mere 400-500 MHz.

Despite the fact neither Intel or AMD have EVER hit their stated efficiency gains each generation? Do people always fall for the same things over and over again?

They have nearly every time...in fact the only time they didn't was with 1st gen BD as far as I can recall...do you ever read anything about AMD?

That 30% is going to be a best-case value. I figure 10-15% is more likely in real-world apps.

Just wanted to point out here the stated gains were over piledriver architecture...not bulldozer. In real world apps it will literally make the biggest impact, because all the efficiency gains will be geared toward mostly improving single threaded performance...since it's blatantly obvious intel still hasn't caught up in terms of multithreaded apps or multitasking. When you have a 150% advantage in performance at one arena, you can afford to augment other arenas that are performing less optimally...

Wait and see...steamroller will be amazing.

Furthermore, AMD has stated the architecture for Kaveri...google Kaveri architecture diagram. Amazingly it pops up about 5 times on the first page showing the steamroller improvements.

I suppose you'd have to research to find that out though...

 

[Cazalan]

http://browser.primatelabs.com/geekbench2/1591657
http://browser.primatelabs.com/geekbench2/1852876

Not exactly sure what we're looking at here.

Richland vs Kaveri?

A6-5350M is Richland
Bantry is the code name for the Kaveri platform.

But then the Processor ID was identical:
AuthenticAMD Family 21 Model 19 Stepping 1

 

[8350rocks]

i'll just chip in a tiny li'l bit. correct me where i'm wrong.

there's no getting around the fact that intel's own compiler favors intel (obviously). being the giant evil corporation they are, they grab every advantage they can. i won't justify intel's practices. but. until amd makes their own compiler and pushes for optimizations for their own cpus, crying foul every time some software rumored to be compiled by icc is kinda useless. there's also the (rumored) point that icc produces faster performance for amd cpus than other compilers. why not get the devs compile on non-intel compilers and then bench? shouldn't be that hard for amd or hsa foundation or cpa. intel maintains hardware and (unfair) compiler advantages, that's that.

ICC is a faster compiler for AMD as well, but it's something like 5-7% for intel it's a little over 20% more efficient. That was the grounds for a lawsuit that AMD won, and intel has still not complied to the terms...3 years later.

what that says is that intel has better turbo and amd doesn't. both fx and sb/ib cpus have auto-overclocking turbo features.

It's not really a drastic advantage, but it is an advantage none the less.

you really shouldn't throw around prices like that when prices may change everyday. checked newegg, fx8350 is going for $200 and 3570k is for $221.
i am not gonna go into what destroys what. benchmarks can be made to favor either way. i can find gaming benches where fx8350 beats a 3570k across the board as well as multithreaded benches (if that's what one calls 'destroying').
although, if you're gonna pit 3570k/3770k vs 8350, solely cpu vs cpu, the fx loses instantly because it has no igpu. no amount of cheapness can fix that. :lol: well.. may be you could run the fx box using remote desktop... i think most of the am3 mobos with built in gfx don't support fx... there was something about old am3 mobos not supporting fx or something... i don't remember well.. may be remote desktop can help.
assuming [strike]both intel and amd[/strike] the amd pc (see it's not solely about cpus anymore) has discreet gfx (only for the sake of comparing the cpu performance), fx can gain it's usual upperhand and such. still, the cost of a discreet card significantly lowers fx's perf/dollar.
sensationalizing goes both ways.

FX8350 is $189 on outlet PC and i7-3770k is $309. $120 is 63% of $189.

No one who buys an intel with onboard graphics for a desktop actually uses them...if you can afford $300 for a CPU, you can certainly afford $100 minimum for a 200% improvement in graphics over intel's iGPU...if anything...you have to count that as a negative for intel, because they might cost less without that rubbish built onto the chip.

Besides FX8350 + HD 7750 costs less than i7-3770k by itself, and you get nearly equal CPU performance and 100% better graphics...so your cost argument is unjustified. Sure, you could talk about the i5-3570k, but the FX8350 is a better CPU to begin with, so you're paying more for less.

that 30% number is sorta vague. if we take 30%, assuming amd achieves perfect scaling, sr should be 30% more efficient(perf/watt) than bulldozer (not pd) according to their own cpu roadmap. then again, amd hasn't uttered the word 'steamroller' in months. they call their sr-based apu kaveri and never mentions what kind of cpu core will be in it.

The architecture diagrams have been out since January of this year. The 30% figure is quoted against piledriver not bulldozer. Does intel quote haswell as an improvement % of sandy bridge or ivy bridge? No one compares their improvement to architecture that is 2 generations ago...only the current generation. Kaveri is coming with Steamroller cores as well...it has been announced by AMD that it will likely be the first steamroller product out, with FX series desktop CPUs on the heels of it in Q1 2014.

if i remember glofo's recent process roadmap, their focus is on ultramobile socs (their future 14nm-xm node's - xm stands for extreme mobility, not 'we'll pwn skymont with this awesomeness'), not high performance high end cpus(amd fx). and there's no mention of 3d(...?) for 20nm node. i could be wrong in reading the roadmap, perhaps someone with better knowledge can explain it better.
http://semiaccurate.com/2013/02/08/global-foundries-lays-out-their-finfet-plans/
http://www.xbitlabs.com/news/other/display/20120920113524_Globalfoundries_Accelerates_Roadmap_Expects_First_14nm_Tape_Outs_in_2013.html
good thing is, this will help amd fiercely compete in mobile sector (jaguar's successors).
tbh, i haven't seen anyone underestimate amd. they came from behind and build llano and brazos, made gcn and are a major proponent of hsa. if any one says bad things, it's (ironically) the amd fanboys.

Google Global Foundries 20nm 3D TSV...just let me know what happens.

I am not talking poorly about AMD, I am being brutally honest about strengths and weaknesses, though it seems the intel guys don't like to talk about the weakness of the product they prefer too much...sore spot maybe?

 

[griptwister]

Features dynamic L3 cache, for 8350 and Calazan. This is not that surprising as AMD pushes its HSA the sacrifice of L3 is temporary but in some instances ie: SC 2 the absence of L3 hurts the A10 by over 10FPS relative to a 965BE and 4300, in BF3 the absence of L3 is less pronounced and the A10 holds pace with the above SKU's. Since Excavator will likely be a unified socket sacrificing L3 long term would impare performance. It is just a matter of AMD evolving its IMC which SR seems to be heavily focused on.

And if this whole 6 core thing is true...

AMD, take my money!!!

So... wait... where are they putting the GPU?

Interesting thought... doesn't this 40% GPU performance increase on richland still make Intel's iGPU suck still? Yes... Yes I think It does hahaha! It's amazing considering AMD is still using DDR3. Can't wait for Kaveri!!!