SAKaveri Versus Richland: A Performance Per Clock Comparison

csbin

Senior member
Feb 4, 2013
814
4
136
#1
http://semiaccurate.com/2014/01/23/kaveri-versus-richland-performance-per-clock-comparison/

When AMD launched its latest APU, Kaveri, almost two weeks ago it touted the gains in performance per clock that this new APU brought with it. The quad-core, dual module CPU on Kaveri is the first implementation of the company’s Steamroller core. The last generation APU, Richland used AMD’s Piledriver core which was notable for being about 6 percent faster per clock than AMD’s Bulldozer core which was the first design in the family tree.

When AMD launched the Bulldozer core and CPU architectural family in late 2011 the company presented a rather anemic roadmap for consistent ten to fifteen percent performance gains with each new generation of Bulldozer based core. Three years later it is both gratifying and disappointing to see that AMD has more or less stuck to its word on that roadmap.

In the Kaveri launch day slide deck AMD noted that on average the new Steamroller core provides 10 percent higher performance per clock than the outgoing Piledriver core and in some cases can be up to 20 percent faster at the same clock speed. Unfortunately for AMD and consumers alike Global Foundries’ new 28nm SHP manufacturing process is less frequency focused than the old 32nm SOI process. Thus at launch the highest performance Kaveri SKU is more or less a match for older Richland chips in terms of absolute CPU performance.

We decided to try and verify AMD’s claims of higher per clock performance. To that end we benchmarked the A10-6800K and the A10-7850K against one another. Both chips had turbo disabled and had a base clock speed of 4 Ghz. As always you can find our raw testing results on Mega. First we have our numbers in table format.

And now those numbers again in bar chart.

AMD claimed a 10 percent average PPC boost for Kaveri over Richland. According to our test results the company actually delivered an average improvement of 11 percent. There were a couple of minor regressions in Cinebench and LuxMark performance, but outside of those two cases Kaveri performed equally or better than Richland. The widest gap was in the wPrime 1024M benchmark where Kaveri was 38 percent faster than Richland.
So there you have it, AMD’s Steamroller core offers 11 percent higher per clock CPU performance than its old Piledriver core. It’s clear that Steamroller’s biggest advantage over Piledriver is its performance in multicore applications. There is a bump in single threaded performance, but the difference in multithreaded benchmarks is a much more compelling story.S|A
 
Mar 13, 2006
10,103
73
126
#2
So even AMD's own slides admit it's slow and late. Nice job.

There are so many ways you could have put that. You picked what's likely the least constructive and most flame-baity method.
-ViRGE
 
Last edited by a moderator:

know of fence

Senior member
May 28, 2009
555
0
71
#3
The caption on the old "staircase" chart, should be saying

~10-15% increases in performance per Watt each year.

I'm so getting tired of captions contradicting the graph shown, or omitting vital info as if to make things brief and concise. There isn't a better example for insulting someone's intelligence.


 
Last edited:

Ventanni

Golden Member
Jul 25, 2011
1,400
13
106
#4
Even though Kaveri doesn't blow our socks off in terms of performance, it's power efficiency is undoubtedly better when you compare the 6800k to the 7850k. I do have to admit though, the 45, 65w, 95w TDP designations for the new Kaveri processors seem strange given their actual power draw.
 

el etro

Golden Member
Jul 21, 2013
1,581
0
81
#5
~10-15% increases in performance per Watt each year.
I'm tired of the same performance increase over and over again when they need to boost hardly their IPC to bring competition again in High Performance markets(consumer or professional).
Anyway their efforts on perf/watt of lower TDP parts is plausible(but insufficient).
 

jpiniero

Diamond Member
Oct 1, 2010
5,987
124
126
#6
need to boost hardly their IPC to bring competition again in High Performance markets(consumer or professional).
Anyway their efforts on perf/watt of lower TDP parts is plausible(but insufficient).
They won't because they are getting out of those markets once Piledriver has run it's course.
 

Enigmoid

Platinum Member
Sep 27, 2012
2,909
0
81
#7
Even though Kaveri doesn't blow our socks off in terms of performance, it's power efficiency is undoubtedly better when you compare the 6800k to the 7850k. I do have to admit though, the 45, 65w, 95w TDP designations for the new Kaveri processors seem strange given their actual power draw.
True, though the better measure of efficiency would be downclocking the 6800k to the same clock and measuring performance and power draw relative to the 7850k. The 6800k is at the far end of the perf/watt curve.

Efficiency has increased though not as much as one would suspect.
 

Torn Mind

Diamond Member
Nov 25, 2012
3,244
0
91
#8
No one should ever believe in claims of huge IPC boosts(40%and above especially) because aside from Netburst to Conroe, big gains in IPC simply don't happen, especially if the architecture is the same. Once you think about it in terms of circuits and how the circuit architecture is more same than different, it makes sense that the percent increase in performance is also incremental. This applies to both Intel and AMD.
 

Ajay

Diamond Member
Jan 8, 2001
4,851
57
136
#9
They won't because they are getting out of those markets once Piledriver has run it's course.
That seems likely since it seems that AMD has neither the engineering staff or $$s to bring up a new uArch. It is vaguely possible that the meager gains in the BD uArch have occurred because AMD has diverted $$s and talent to a new uArch - but so far there are no signs of this.
 

Zstream

Diamond Member
Oct 24, 2005
3,165
0
116
#10
That seems likely since it seems that AMD has neither the engineering staff or $$s to bring up a new uArch. It is vaguely possible that the meager gains in the BD uArch have occurred because AMD has diverted $$s and talent to a new uArch - but so far there are no signs of this.
The Jaguar will be here for some time and with Windows 9 being slimmed down, it will be perf/watt that wins the day.
 

Ajay

Diamond Member
Jan 8, 2001
4,851
57
136
#11
The Jaguar will be here for some time and with Windows 9 being slimmed down, it will be perf/watt that wins the day.
Yes, I was just talking about big core. But it doesn't seem like AMD is having much success yet with Jaguar based implementations outside of consoles yet. The money they put into ARM development would have been better spent on x86 in the short run, IMHO. Long term, I don't know - AMD is in a tough spot wrt x86 DT/Mobile and facing an battle royal for ARM based servers. I wouldn't want to be R. Read.
 

CHADBOGA

Golden Member
Mar 31, 2009
1,770
26
126
#12
SemiAccurate really uses some crappy benchmarks.
 

inf64

Platinum Member
Mar 11, 2011
2,809
29
136
crazyworldofchips.blogspot.com
#16
Ok so you take away the wprime and you are at 7.7% (taking into account Cinebench odd results). That's not bad at all considering the fact that many workloads stress the shared FP units which didn't change for the better (but for the worse as one MMX pipe is gone).
 

Abwx

Diamond Member
Apr 2, 2011
8,774
100
126
#17
Wprime is a useless synthetical benchmark. And they used it two times to distort the average towards Kaveri.
As synthetical as CB R15 or 11.5 , the former being purely integer..

Wprime scores are indicative of integer throughput the same
way that winzip/7zip or X264 but since its scores are not
compatible with your point then the bench must be removed...
 

Enigmoid

Platinum Member
Sep 27, 2012
2,909
0
81
#18
As synthetical as CB R15 or 11.5 , the former being purely integer..

Wprime scores are indicative of integer throughput the same
way that winzip/7zip or X264 but since its scores are not
compatible with your point then the bench must be removed...
No. Not even close.

Wprime is useless, completely useless and irrelevant in terms of real world performance.

Cinema 4D is a piece of real world software which was used in films such as inception, narnia, tron, iron man, etc. Cinebench is an benchmark of how well a computer will run Maxon's software.
 

inf64

Platinum Member
Mar 11, 2011
2,809
29
136
crazyworldofchips.blogspot.com
#20
Check my last post before you reply ;). I got to average of 7.7% based on their numbers without wprime, it's solid improvement and not bad at all.
 

Abwx

Diamond Member
Apr 2, 2011
8,774
100
126
#21
Check my last post before you reply ;). I got to average of 7.7% based on their numbers without wprime, it's solid improvement and not bad at all.
And despite the two archivers 7% lower IPC , i see no protests because two similar benches where used, besides they got the worse scores of all reviews for theses two tests, so much that the reviewer mentioned that from his
point of view 10% on average is the most accurate number.
 

PPB

Golden Member
Jul 5, 2013
1,100
10
106
#22
No. Not even close.

Wprime is useless, completely useless and irrelevant in terms of real world performance.

Cinema 4D is a piece of real world software which was used in films such as inception, narnia, tron, iron man, etc. Cinebench is an benchmark of how well a computer will run Maxon's software.
CB is garbage, doesnt represent the real world renderers. So there you have a garbage bench that shows good improvement and another garbage bench with regressions to balance it out.
 

Enigmoid

Platinum Member
Sep 27, 2012
2,909
0
81
#23
CB is garbage, doesnt represent the real world renderers. So there you have a garbage bench that shows good improvement and another garbage bench with regressions to balance it out.
Yet its been used by many major movie studios......

The fact remains that it is representative of real world software (whatever quality it is) and is used by many.
 

Cookie Monster

Diamond Member
May 7, 2005
5,161
0
86
#24
So why won't they release an FX lineup based on steamroller cores? I still think its not a bad move because these new cores do bring some incremental performance/efficiency gains. Or is the business case for it just not worth the risk?
 

Homeles

Platinum Member
Dec 9, 2011
2,585
0
0
#25
Check my last post before you reply ;). I got to average of 7.7% based on their numbers without wprime, it's solid improvement and not bad at all.
Haswell's 10% was the end of the world, Kaveri's 7.7% was a "solid improvement."

I'm not calling you out or anything -- I don't recall your comments on Haswell. I'm just pointing out the hypocrisy here. The inconsistency is especially bad when you take into account that AMD's IPC was already low to begin with. It just goes to show that the accuracy is generally unimportant in hardware communities -- the most important things are facts that promote your agenda.

I really disagree that 7.7%, or the more favorable 11% average IPC improvement can be considered "solid." It's mediocre, regardless of brand.

These crappy improvements are essentially meaningless. We're living in a post-performance superiority world. Performance scaling is all but dead right now, and the industry has wisely focused on other ways to improve their products. I wish the enthusiast community would catch up already and focus on the things that matter today, and not the things that mattered years ago.

I'm not saying this discussion shouldn't be happening, but performance is still the end-all, be-all in communities like this, and it frankly shouldn't be. Thankfully, it seems that more people are figuring out that computers can be used more than just rendering frames for entertainment.
 
Last edited:


ASK THE COMMUNITY