AMD CPU speculation... and expert conjecture

[-Fran-]

Since you can't look at the assembler code used for AMD and Intel for each game, you can not conclude Intel is 2X IPC over AMD. The only reasonable thing con conclude from that information is "Intel around 1.5 times the framerate than AMD at the same speed on average for those games".

You're one of those sore losers that love to shove the other team's defeat and blow out of proportions the score when they have a victory.

Cheers!

 

[8350rocks]

http://forums.anandtech.com/showthread.php?t=2297052
Judging by superpi an fx8350 at 4ghz is equivalent to a 2.4ghz core 2 duo so a qx6700 my first ever quad at 2.66ghz beats the fx8350 at macro makes sense.

SuperPi?! AGAIN?!?!?

Do you have anything else to be doing? Rearranging your sock drawer maybe? Cleaning your room? Go outside and stare at the sun for a few hours? Anything?

EDIT:

http://en.wikipedia.org/wiki/Super_PI#Significance

Super PI is used by many overclockers to test the performance and stability of their computers. In the overclocking community, the standard program provides a benchmark for enthusiasts to compare "world record" pi calculation times and demonstrate their overclocking abilities. The program can also be used to test the stability of a certain overclock speed. If a computer is able to calculate pi to the 32 millionth place after the decimal without mistake, it is considered to be moderately stable in terms of RAM and CPU. However, longer tests with other CPU/RAM intensive calculation programs will run for hours instead of minutes and may give and/or provide better stress system stability. Super PI is not the fastest program for calculating pi (see software for calculating π for faster alternatives), and it is losing favor in overclocking and benchmarking circles because of it's poor representation of real world benchmarks. SuperPi is a better indicator of RAM performance than CPU performance, as x87 floating point instructions are a legacy code left over from the x486/x586 days. This coding method is no longer supported on a wide scale and is considered archaic and inefficient by modern coding standards.

Are we done with SuperPi yet? Yes? kthxbai! (to borrow from yuka)

 

[Ags1]

Yegods, we finally get past the ARM thing, now the trolls are back!

 

[8350rocks]

Yegods, we finally get past the ARM thing, now the trolls are back!

If it isn't one thing, it's another...isn't it?

 

[Cazalan]

This forum really needs a F&%*($& ignore button for garbage like this...

Can we get one of those? Please...?

+1000

 

[The Q6660 Inside]

This forum really needs a F&%*($& ignore button for garbage like this...

Can we get one of those? Please...?

+1000

4200+

 

[Cazalan]

Wait, a $300 processor is faster than a $200 processor? OMG!

 

[8350rocks]

Wait, a $300 processor is faster than a $200 processor? OMG!

You know what's funny?

It's not always faster!!! LOL.

(see 3770k vs. 8350)

 

[tracker45]

how good will the steamroller be compared to an i5 ?

 

[8350rocks]

how good will the steamroller be compared to an i5 ?

Should be about same/same single thread performance, and considerably better at multitasking.

 

[tracker45]

hopefully they are on am3

 

[jdwii]

Okay amd users explain where improving lets say an a10-6800k or 5800k by 30% will get to an ivy bridge i5 at lets say 3.2ghz performance wise with a steamroller 4 core kaveri. I know its will be 70% slower going by amds projected gains but prove this is wrong. You simply can't as thats why you ignore the benchmarks and make made up figures to make amd look closer then reality. No one would buy a core i5 if the fx8320 was only 30% slower per core then an i5 ib. That would make the fx8320 beat the i5 ib cpu lets say both at 3.5ghz easily. An i5 at same clock speed would equal an fx8320 with 5 cores 100% loaded and 20% of the 6 core loaded and 0% load of the other 2 cores. That would make the fx8320 sell like hotcakes if this was the case but we all know the i5 equals the 8 cores if at same clock.

So if you theory is correct an i5 at 3.5ghz would be equal to 5.2 cores of a piledriver cpu at 3.5ghz. Thats means its 60-70% slower then an fx8320.

Now the reality is ivy bridge is 2x better per core then piledriver likewise haswell is 2.1x better per core then piledriver which makes the i5 ivy bridge at 3.5ghz 4 cores equal 8 cores of amd piledriver cores at 3.5ghz and richland is not any better really.

Again you fail to realize what i said 10 pages ago about the scaling of cores when it comes to PD/BD

 

[palladin9479]

Hmm depends on generation. Clock for Clock the SB+ uArch's will always be better, they have more resources (assuming AMD stays with 2 ALU per core and shared FPU). AMD merely has more independently addressable processing units, aka cores, which lends to selective performance advantages for specific tasks. It's really a server / workstation type of design consideration, but hey it works for some of us. AMD attempts to make up for the performance shortfall by increasing clock speed, which works to a degree. You really should compare system components on an overall price vs performance vs opportunity cost instead of on absolute performance. When people build or purchase a solution they tend to have a total budget in mind and want to maximize the value on that budget.

 

[noob2222]

noob222 funny how an i5 4670k running at 3.4ghz scores 634 vs the fx8350 at 4.03ghz scoring 643.
http://www.cbscores.com/
In fact the i5 2500k scored higher which obviously means whoever runs an i5 4670k with turbo with no load on windows will score around 700.

One known fact is cb doesn't properly show actual clocks of intel cpus. With i5 4670k do you think is stock, the 643 score or the other one at 528?

Same page 100+pts different.

 

[Sigmanick]

AMD's 3rd Quarter results are in and they beat the Market expectation of $1.42Bn by selling $1.46Bn worth of goods. So good news for me as I bought some stock in them, i'm really hoping for $8.00 a share by next June. A toast To upcoming Steamroller and the SteamBox!

Lets compare more modern tests. One is free for everyone (boinc) the other is great if you already have the software (Tomb Raider).
[strike]DieHard Intel fan[/strike] Hafijur, will you run the same Benchmarks as i have? Or to put it in another way, Will you show me yours if I show you mine?

Obviously not showing off e-peen, as these are modest components.
Stock speeds with a 7200 rpm HDD
FX 6300 ($120.00 at newegg)
8 gb ddr3 1600
HD 7850
Screen native resolution is 1600 x 900

Tomb Raider Bench Mark
Scores: Min / AVG / Max
Low :172 / 216 / 260.7 fps
Ultra : 54 / 70.2 / 86 fps
Ultimate (tress FX) : 31.1 / 41.4 / 52.4 fps

I haven't Folded in a while, and I feel bad about that (BOINC that is, not Towels... but not them either though).

Here are my test results from the BOINC CPU test http:/Nicks PC results/ and I hope you can view the page. If not, it says that my Scores are -

Float Point Speed: 3014.42 Million ops/sec
Integer Speed: 10243.61 Million ops/sec

We will compare my score to a Core I7 3930K $569 AT NEWEGG (6 core 12 thread Sandy Bridge e 130W processor)
http:/Boinc CPU test/
Float Point Speed: 3178.26 Million ops/sec
Integer Speed: 12215.14 Million ops/sec


The lowly 95W FX6300 gives 94.8% of the Float point power, 83.8% of the Integer Power for less than 20% of the price. So tell me how intel has 200% better IPC?

IFF Steamroller gets a 30% ipc boost and can maintain clocks for DT Enthusiast chips, THEN Steamroller will out pace the SB-e chip.

I'm not really sure how to calculate the effect of a 3rd ALU on the BOINC Float test.

 

[griptwister]

Sigmanick with the ownage :lol:

After reading what I've missed, this thread has turned into quite the Poo Fling war. Let's just forget about ARM and the unInteligent derping and lets continue with steamroller. Speculation is fine, how ever, using speculation as a fact is not fine.

 

[blackkstar]

You guys really need to learn how to ignore you know who.

He has gone out of his way to actually private message me to try and convince me Intel is superior to AMD in every way.

He is either getting money for forum sliding or he's got some sort of personal issues that I am not going to touch on in this thread.

Regardless, I fail to see why everyone continues to address him. At best you're giving some Intel PIE marketing guy money for each post and at worst you're encouraging some sort of mentally unhealthy behavior.

Also, I found it kind of interesting, but I was looking back at Steamroller news from 2012 and there were two prominent rumors floating around. The first was that there was going to be a Steamroller that came out in early 2013 (which clearly didn't happen).

The second was that we'd see DT SR in mid to late 2014.

Oddly enough, it seems like the release info for Steamroller has been around for a year and a half, just seems like everyone forgot about it.

Read: https://www.google.com/search?q=amd+steamroller+roadmap+2012

Tons of conflicting issues, but if we look at what was supposed to happen and what has actually happened, things might become a little more clear.

http://www.xbitlabs.com/news/cpu/display/20121105000202_AMD_s_Roadmap_Slide_Does_Not_Predict_Steamroller_in_2013.html

That one is quite interesting to me. Yes, they were wrong about no SR in 2013, however they were right about Richland.

So, I guess what I am asking is, how do you guys feel about there being two versions of Steamroller (like seronx has mentioned previously) considering how delays and new products have crept up? I am kind of feeling that Steamroller 1 was supposed to be out now if it wasn't scrapped and it was then replaced with a Steamroller 2.

It makes sense to me, the old slides had Steamroller marked as "greater parallelism", which is something AMD doesn't need right now. It needs better IPC, and if they can do it while keeping the module design AND keeping the module die sizes about the same, they will have a massive winner of a product on their hand.

I think that maybe this is what has happened. We might see the "greater parallelism" show up in some sort of PD refresh with more cores, and then SR will release a year later than what we would have normally expected (October release) and it will have a far different goal than "greater parallelism".

Just my 2 cents, but it sounds about right. And I am a strong believer that AMD saying they are done competing with Intel means that AMD will still go to HEDT market, that is no longer something Intel is concerned with and it's why it's taken to strongly to internet campaigns against power usage and spreading FUD related to it, when power consumption is irrelevant for desktop CPUs in gaming PCs. (and as a side note, I had to tell some guy who was all excited about Broadwell doing 4.8ghz at an estimated 100w or so that he had three 170w GTX 760s in that very same computer, and that saving 80w on your CPU when you have 510w of graphics is a waste).

 

[Cazalan]

In today's Q3 earnings call Kaveri came up only once. That is will start shipping in Q4, but no clarification beyond that.

At least they did return to profitability at 48M, vs a loss of 74M last quarter.

 

[Cazalan]

So, I guess what I am asking is, how do you guys feel about there being two versions of Steamroller (like seronx has mentioned previously) considering how delays and new products have crept up? I am kind of feeling that Steamroller 1 was supposed to be out now if it wasn't scrapped and it was then replaced with a Steamroller 2.

That's the general rumor. S|A went into more detail a while ago. The thread in their forum is called "Kaveri 2.0" because the first Kaveri failed qualification.

http://semiaccurate.com/2012/11/06/amds-kaveri-apu-slips-again-2014-now/

 

[de5_Roy]

good for amd, coming out of 'red' this quarter. 😛 rory reed delivered on his promise. isn't it temporary though?

hopefully they are on am3

since bulldozer (fx8150) cpus amd officially doesn't support am3 socket. you can use am3 cpus on am3+ socket.
new kaveri apus will support a new socket - FM2+. companies have unveiled their fm2+ lieups already.

@kaveri rumors: i noticed 3 kinds.
the ones before mark papermaster's keynote speech,
the ones after the keynote speech,
the ones after rumors of cancellation.
since we've already seen kaveri silicon in physical form, earlier rumors are moot. however, steamroller cpus (likee fx ones) are whole different story.

 

[noob2222]

You guys really need to learn how to ignore you know who.

He has gone out of his way to actually private message me to try and convince me Intel is superior to AMD in every way.

Same here. Point I'm making, Maxon quit using icc and look how fast AMD cpus got with the new bench.

amazing when you optimize for both hardware camps amd isn't so far behind.

 

[juanrga]

Secondly, on PC's, there isn't a need for Mantel. The worse case overhead for DX11 is "maybe" 5% performance loss, which for a game running 60 FPS equates to ~3 FPS worth of performance loss, which is acceptable. There simply isn't much overhead left in the API.

So is re-writing the software and adding a secondary graphical backend, that only benefits the minority player in the GPU market, worth maybe a 5 FPS gain for developers? No, it isn't.

I wonder why you continue writing mantel, when it is MANTLE. If you cannot learn the correct name how could you learn its goals and characteristics?

The worst case scenario for DirectX is something as 100x. That is the reason why ID Software had so many trouble trying to surpass the 30 fps in a PC when their engine/game got 60 fps in a 10x less powerful console.

Nobody is rewriting the software, but porting the console code to the PC. I predicted that MANTLE version of BF4 would be about 30%--50% faster than the DirectX version

https://twitter.com/juanrga/status/383193358086049792

Time will say which is the exact performance gain, but I can sure you that you are doomed if you believe it will be only a 5% more.

 

[juanrga]

http://www.amazon.co.uk/review/R6NLTFRGDJECE/ref=cm_cr_pr_cmt?ie=UTF8&ASIN=B00CV3E3WK&linkCode=&nodeID=&tag=#wasThisHelpful
read the review, an amd user going to intel and suprise suprise he gets double the performance over a phenom 2 and we all know phenom 2 ipc is slightly better then piledriver so with my 2.1xipc theory the i5 4570 with turbo is on par with an fx8350.

i5 4670k:
http://www.amazon.co.uk/review/R29REJDY4R8AUR/ref=cm_cr_dp_cmt?ie=UTF8&ASIN=B00CO8TBOW&channel=detail-glance&nodeID=340831031&store=computers#wasThisHelpful
http://www.amazon.co.uk/review/R3KDIPLP26G9QD/ref=cm_cr_pr_cmt?ie=UTF8&ASIN=B00CV3E3UC&linkCode=&nodeID=&tag=#wasThisHelpful

Now lets see a 4 core kaveri get to 80% of an i5 haswell cpu at same clock or 70% of an i5 ivy bridge cpu. Even that will be on amds at best target for steamroller.

This forum really needs a F&%*($& ignore button for garbage like this...

Can we get one of those? Please...?

My brain comes with one of those buttons installed. I can switch on or off when I want, and I can ignore the garbage or just reply for fun.

Mantel is NOT an open standard; it REQUIRES a GCN based GPU. Which means mantle dies when AMD eventually replaces GCN.

Secondly: Ignore the marketers. Remember what AMD claimed about BD performance?

Thirdly: Let the API wars commence: http://www.guru3d.com/news_story/nvidia_does_5_million_deal_with_ubisoft.html

NVIDIA now throwing money around. Lets see who buys Activation (probably NVIDIA).

Still amazed why you insist on writing 'mantel' instead MANTLE.

MANTLE is not an open standard in the sense of being developed by a consortium or a standards body, but it is open. AMD already stated that MANTLE is not AMD CUDA.

AMD has also stated that MANTLE could be used by Intel and Nvidia if they want. My guess is that they don't.

Therefore if AMD did a mistake once with BD performance but was right with any performance claim made for Piledriver, jaguar, Richland, Centurion... you only take the single exception and ignore the rule?

Also, if your logic is to ignore anyone who makes a mistake then you couldn't post here anything legit, because you did more than one mistake in the past. Do you apply your own logic to yourself?

 

[juanrga]

cbscores is probably the most well known site that lets users submit there cinebench scores.

I love how their 'database' says that both a FX-8350 and an A10 have four cores.

Also loving how it reports that the Xeon E5-2690 is a 20 cores/40 threads chip, when it is 8-core/16 threads.

Finally, their mixing of scores obtained with different versions or even different OSes confirms that this is a very professional database.

 

[juanrga]

how good will the steamroller be compared to an i5 ?

It depends.

For ordinary, old software, Kaveri 4C will be about as fast as a Sandy Bridge i5. In some tests Steamroller will be faster and in others i5 will be faster.

Kaveri will be about 3x faster than i5 with HSA software.

Regarding old graphics code, the iGPU of Kaveri will be faster than anything made by Intel, including expensive Iris Pro used in some Haswell i7 models.

Kaveri graphics would be much faster with MANTLE enabled software. I am anxious to see the performance of the MANTLE edition of BF4 on a Kaveri A10.