AMD CPU speculation... and expert conjecture

Page 583 - 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.

noob2222

Distinguished
Nov 19, 2007
2,722
0
20,860

Wow, back to this discussion ...

Juan can never be wrong.

He didn't include the a10 6800k is because he doesn't know what a base line is, and pretending that the base on all of his calculations is 0. meaning that the a10 6800k has 0 performance in all benchmarks. just don't tell him that the 6800k only does that when turned off. I could go into detail what a baseline is, but he still thinks I was way off when I predicted that AMD's stated 20% minus 11% clock speed means that kaveri was only going to be 9% faster than richland and in some cases, bencmarks prefer the clock speed.

don't tell him that margin of error != absolute error.

In short, he didn't include the 6800k because he doesn't want anyone to know just how far off he was.
 

interesting. i didn't thoroughly check the igpu model numbers because i've always found those to be very convoluted, especially ulp, mobile and oem ones. the recent r5/r7/r9 gpu model numbering is worse than 7xxx numbering.
as for the a6 parts, i dunno.. it could be possible to make a 192 shader igpu. however, the 7400k has 256 shaders (4CUs) so this is a bit contradictory.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790
@colinp, if I had seek for "universal praise", I had used ICC-based benchmarketing on an operative system which I couldn't care less.

@jdwii, in those three benchmarks that you mention, the deviation from my predictions was -7%, 5%, and 1% which averages to 0%. Your own computations are off by one order of magnitude.

I note that you didn't bother to answer my question about PassMark. I will do: the deviation between my prediction and final measurement is again in the single digit percent.

@noob2222, my work was based on a baseline and the computation of error is standard. The baseline used was mentioned in my article and published before Kaveri was released.

http://www.brightsideofnews.com/news/2013/11/4/what-to-expect-from-kaveri-a-detailed-predictive-analysis.aspx

I will be nice and will not remind everyone which were your laughable predictions.
 


Kavari's biggest boost was in it's iGPU. It's efficiency increase on the CPU side were largely mitigated by the clock speed reduction. Overall the CPU is faster then the 6800K when we consider it'll be used as an APU and not a CPU + dGPU combo. I still believe the price makes it non-competitive and AMD is taking entirely too long to release the A8-7600.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790
After this short return to the old times (Fight Juan! Fight Juan!), let us return to AMD products and look to the future again.

The HBM dream is gone. Despite what several sites published, including Toms, HBM is not coming to Carrizo. A friend of mine insists on that Carrizo will be only released for mobile and that Kaveri will be maintained in desktops during all 2015. Sad, very sad!

As some of you know ;-) I was expecting a clock reduction from the change from 28SHP to GF28A, but I was expecting this to be compensated by a large improvement in IPC of about ~40%

AMD-Bulldozer-family-performance.jpg


in good agreement with the famous die-shot showing a Steamroller like module but with doubled execution units

Excavator-die-shot.jpg

However, it seems all that was canceled and final Excavator module will be Steamroller with minor tweaks on caches and so, providing only 5--10% IPC gains. This is not enough for compensating for the clock reduction on desktops and the reason why Excavator is not coming to desktop.

On mobile, GF28A provides high enough frequencies and Carrizo will be a real improvement over mobile Kaveri, which was already very competitive against Haswell i5/i7s.

This friend also claims that Carrizo will be released late, Q3'2015. K12 will be released in 2016 and its x86 sister latter. This former roadmap for 2015

https://twitter.com/juanrga/status/406390600804466689

has been changed by a new rodmap showing Kaveri on desktop extending to 2015 and 2016.

As said it is all sad!
 

colinp

Honorable
Jun 27, 2012
217
0
10,680


Ah-ha, so you ignored the results that you couldn't care less about. Glad you finally admitted it. Feels good to let it out, I bet.
 

etayorius

Honorable
Jan 17, 2013
331
1
10,780
So basically Carrizo will be used only on Mobile, Kaveri will see few improvements therefor it should be Kaveri+.

AMD is stagnant in CPU. very sad indeed.

This basically means more Piledriver crap till 2016 on the highend.

Does this has to do with GloFo and TSMC or something else?
 

blackkstar

Honorable
Sep 30, 2012
468
0
10,780
Relying on PD for the next two years in HEDT and Kaveri for mid-range desktop seems like suicide to me. AMD is basically going to pin the next two years on OEMs putting their products into quality devices and console sales. It might help since there's rumors of them getting a new Nintendo handheld as well, but that's not the greatest market to be in. A lot of people buy phones that can play crappy games and they're satisfied with those crappy games.
 

juanrga

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


I only can admit you and others are trying hard to negate that the deviation from my predictions was in the single digit percent.
 

juanrga

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


So sad as it seems, the responsible is AMD and nobody else. Bulldozer is unfixable and Keller needs time for a new architecture from scratch. But this also mean lots of money for R&D and AMD has returned to red numbers again this quarter. Therefore, I suppose the desktop Carrizo could be also canceled due to lack of funds.
 

juanrga

Distinguished
BANNED
Mar 19, 2013
5,278
0
17,790
Update: He says that Kaveri will be maintained in desktops during all 2015 and part of 2016!

This makes sense if K12 is an early 2016 launch and the x86 core derived from it is a late 2016 launch.
 

juanrga

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


Maybe these Kaveri vs trinity benches can help you to decide if upgrade or not

351sopw.jpg

25qdn5x.jpg

w8nhg2.jpg

qn01np.jpg

jl035t.jpg

335gv0m.jpg


http://www.3dmark.com/pcm8/3028085
http://www.3dmark.com/pcm8/3645888
http://www.3dmark.com/3dm/3028186
http://www.3dmark.com/3dm/3629180

Since it seems that Kaveri will be with us some time. It is likely that we will see new HSA announcements in next months.
 

Cazalan

Distinguished
Sep 4, 2011
2,672
0
20,810


If you stopped making narcissistic posts about predicting the future you'd get less flack.

I can predict here that ACMP...
I can also predict that SSDs...
I could even predict the size...
I can also predict which will be the architecture...
I did a set of precise predictions ...

You know the saying if everyone is causing me problems, maybe it's me not them.
 

noob2222

Distinguished
Nov 19, 2007
2,722
0
20,860


ya ... my laughable predictions for kaveri being 9% faster than the 6800k ... where did it end up? oh ya, 8%

your predictions ... lol


Kaveri scores for the CPU are obtained from taking Trinity/Richland scores as base and utilizing the assumed 20% gain from doubling the decoder per module.... bla bla bla ramble, ramble, ramble

Post the numbers for your baseline so we can see how far off you really were since richland's numbers vary from site to site.

oh that's right, that would mean you would have to admit you were wrong. your just like my ex who never sais anything concrete, just a bunch of half baked ideas that are not solid or factual, therefor can't be proven wrong since you didn't really say anything in the first place.
 

jdwii

Splendid
The A10 7850K(7750 which is even a little better than a 7850K iGPU) is as fast as a video card from 2008(4870)
http://www.anandtech.com/bench/product/513?vs=535
Seems to me APU's are 6 years behind GPU's in terms of performance. Also a little sad a I7 920 came out in 2008(6 years ago) to. I doubt any APU has the CPU performance of this yet either.
http://www.anandtech.com/bench/product/47?vs=1200

I think that explains enough to juan on why APU's are kinda for mainstream not for gamers or extremist.
 

Fidgetmaster

Reputable
May 28, 2014
548
0
5,010
None of the APU's are even close to first gen... their best one is like closer Athlon 760k/fx 4300 performance...

920 is underrated as hell still man....performance wise its more into the fx 6300-8350 range....

You do have to considerably Overclock it though to see anything remotely significant....
 

Fidgetmaster

Reputable
May 28, 2014
548
0
5,010
So really for the price they are still asking for PD and those APU's is ridiculous....

When someone could easily spend about the same on low end i3/haswell or a little more on better Haswell i3/i5s....and get ridiculously better performance....

AMD is screwing themselves in my opinion still trying to rely too much on APU/FX....

They are in dire need of something better/new.
 
To be fair, APUs make sense up to about $300 or so; even with a Pentium, its hard to get a CPU and dGPU that beats an APU in gaming for under $300. That's the market AMD should focus on.

Hence why if the Intel iGPU gets "good enough", AMD is in trouble.
 

don't worry, then. intel's only igpu(processor) that barely qualifies as "good enough" comes in bga, soldered to motherboard, h-series chipset (no o.c.) and costs higher than $220 to up to $360+. the broadwell/skylake delay/launch mix/up will give a bit of breathing room, for a while. if intel makes socketed bdw cpus with iris pro availabile under $200, then amd will be in trouble as they've priced their sole kaveri apus right under $180. i doubt that will happen.
don't get me wrong, even hd 2000/2500 will be "good enough" in some areas. by "good enough" i mean an igpu that can play games (edit: smoothly, no image q corruption or intermittent stuttering) at 720p-900p at low/medium settings, not an igpu capable of browsing internet and word processing (although depending on drivers either of those tasks can be nightmarish).
 

wh3resmycar

Distinguished


well he's predicting that HSA will fill that big performance void. but i could care less as this juan dude can never be specific as to which performance delta he's attributing his predictions to, whether it's server or consumer class APUs he's talking about.

as far as video games are concerned, his predictions are way off.

 
Status
Not open for further replies.