AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

lopri

Elite Member
Jul 27, 2002
12,941
361
126
Looks to me the second chart is simply a division by core counts. e.g. 6950X: 21576 / 10 = 2157.6, 1800X: 20249 / 8 = 2531

Silliness abound, for a few extra clicks the site will do whatever it takes.
 

lolfail9001

Golden Member
Sep 9, 2016
1,056
353
96
I don't know why but maybe they tested with same Memory?
That was my first thought too, but it would mean that physics score only depends on memory. And that is obviously not true, since Kaby and Bdw-E both clearly scale with frequency. Scale linearly, in fact. Then there's an issue that physics test does not scale linearly, there is a drop off in efficiency going from 12 threads to 16 threads on both BDW and Zen. Now, they [drop offs] are different, but that difference can be explained with Zen going from 16MB L3 to 16 MB L3 and BDW going from 15 L3 to 20 L3.
Shortly, i would disregard the alleged 4Ghz score entirely, because no way in hell it is a 4Ghz score.
Looks to me the second chart is simply a division by core counts. e.g. 6950X: 21576 / 10 = 2157.6, 1800X: 20249 / 8 = 2531
Physics test is notorious for being unrealistically parallel.

Good news: at least in physics test, Zen is basically at Broadwell-E levels, difference is within margin of error.

Bad news: physics test has literally 0 relevance to actual gaming.
 

lopri

Elite Member
Jul 27, 2002
12,941
361
126
I am talking about how the chart was made, not how scores are calculated. The second chart's numbers are simply divisions of the first's.

Edit: To be fair it does not look the author is claiming otherwise. "Score per core" can be interpreted as simply "scores divided by number of cores." Where do these scores come from anyway?
 
Last edited:

Face2Face

Diamond Member
Jun 6, 2001
4,100
215
106
Just imagine, R5 1600 with an entry level mobo for $300 ... wowzers!

$239 at MC o_O


Seriously, the R7 1700 and R5 1500 are the true jewels of the lineup.
 

Agent-47

Senior member
Jan 17, 2017
290
249
76
i don get it. single thread results makes zero sense. its not memory dependent as i7 scales just fine with clock.

why can't they run the bench with cores disabled in the bios. or if thats not possible, at the very least set processor single core CPU affinity from task manager.
 

Glo.

Diamond Member
Apr 25, 2015
5,112
3,727
136
Based on the scores, you basically get 6800K, with AMD branding, with 65W TDP, and for 200$ less, when 6C/12T CPUs are considered.

That is actually mind blowing value...
 
  • Like
Reactions: Drazick

HurleyBird

Platinum Member
Apr 22, 2003
2,584
1,144
136
Am i the only one finding a claim that Zen has exact same physics score per core at 4Ghz,3.3Ghz and 3.2Ghz... a little bit hard to believe?
Per core scaling is significantly <100%. The 4GHz CPU has a lower per-core score because it's being divided by 8. Looking at the per-core performance of both 6-core CPUs (6800K vs. 37/33), which happen to be closest to one another (3.4GHz vs. 3.3GHz clock speed), there's a 4% difference in absolute performance between the two. There's some wiggle room depending on how turbo behaves for each CPU, or if turbo is even enabled, but it looks like Zen and Broadwell are pretty evenly matched in this benchmark as far as IPC.
 

unseenmorbidity

Golden Member
Nov 27, 2016
1,395
967
96
Based on the scores, you basically get 6800K, with AMD branding, with 65W TDP, and for 200$ less, when 6C/12T CPUs are considered.

That is actually mind blowing value...
Probably $300 less, because you aren't paying an extra $100 for a Quad channel memory x99 motherboard.
 

lolfail9001

Golden Member
Sep 9, 2016
1,056
353
96
Per core scaling is significantly <100%.
Yeah, i checked that literally 5 minutes ago, when i concluded that 4Ghz sample was not running on 4Ghz. Or the "stock" clocked eight and six core was not running on 3.4/3.3Ghz, but more like 3.5, making their throughput in this case about 5%lower than Broadwell.... that once again aligns with what we knew since last August. Funny.
Memory bound or some other bottleneck
I figured it has to be combination of cache structure and scaling deficiency of 3dmark, the one that puts 6900k 10% behind 6850k in spite of identical clocks.
 

Agent-47

Senior member
Jan 17, 2017
290
249
76
Then there's an issue that physics test does not scale linearly, there is a drop off in efficiency going from 12 threads to 16 threads on both BDW and Zen. Now, they [drop offs] are different, but that difference can be explained with Zen going from 16MB L3 to 16 MB L3 and BDW going from 15 L3 to 20 L3.
Shortly, i would disregard the alleged 4Ghz score entirely, because no way in hell it is a 4Ghz score.
actually that makes sense.

if its thread dependent, we can compare the single thread IPC by comparing the numbers from two 6c or 4c CPUs:

6c-> amd@3.3 = 2545 intel@3.4 = 2645 or amd 5% slower clock for clock.
4c-> amd@3.2 = 2544 intel@4.2 = 3552 or amd 6% slower clock for clock.

facinating

EDIT:
8c-> amd@3.4 = 2231 intel@3.2 = 2329 or amd 10% slower clock for clock. This seems to suggest amd's SMT is not as efficient?? or its cache is not as good??
 

lopri

Elite Member
Jul 27, 2002
12,941
361
126
I checked out the link where the author says:

I think the next chart is far my important. Notice how close all Ryzen CPUs are to each other if we take single-thread performance.
That is just wrong. His chart makes it look like 3.2, 3.3, 4.0 GHz CPUs had identical ST performance but somehow a 3.4 GHz one were 12% slower. Then he goes on to muse as if he discovered something deep. :rolleyes:
 
  • Like
Reactions: Magic Hate Ball

PhonakV30

Senior member
Oct 26, 2009
979
351
136
AMD Ryzen: ZD3406BAM88F4_38/34_Y — Eight-Core CPU @ 4Ghz = 20249
AMD Ryzen: ZD3406BAM88F4_38/34_Y — Eight-Core CPU @ 3.4Ghz = 17878

So for 13%(20249/17878) more , You need either 200mhz more ( 4.0/3.8 = 5.2% ) or 600Mhz ( 17% ).so We take 17%.
for 17% more clock You get 13% more score.now If Octa cores gets 4.5Ghz then

4.5/3.4 = 32% ==> ( 32 * 13 ) / 17 = 24% ==> 17878 * 1.24 =22252 Score
 

HurleyBird

Platinum Member
Apr 22, 2003
2,584
1,144
136
Yeah, i checked that literally 5 minutes ago, when i concluded that 4Ghz sample was not running on 4Ghz. Or the "stock" clocked eight and six core was not running on 3.4/3.3Ghz, but more like 3.5, making their throughput in this case about 5%lower than Broadwell.... that once again aligns with what we knew since last August. Funny.
The stock 6800K is probably running at a turbo of 3.6GHz.

So, assuming that turbo is disabled on the 37/33, Ryzen has something like 4% better IPC for this task.

But assuming that turbo is enabled on Ryzen and it's running all that way up at 3.7GHz, then the difference is something like 6.8% in favour of Broadwell.

If the all core turbo for Ryzen is actually lower than 3.7GHz, lets say, 3.5GHz, then the difference is about 1% in favour of Broadwell.

This is assuming frequency scaling is mostly linear. If that isn't the case, these values can be compressed a bit. Either way, both cores look pretty close in terms of IPC for this benchmark.
 
Last edited:

lopri

Elite Member
Jul 27, 2002
12,941
361
126
EDIT:
8c-> amd@3.4 = 2231 intel@3.2 = 2329 or amd 10% slower clock for clock. This seems to suggest amd's SMT is not as efficient?? or its cache is not as good??
See, that is why you cannot do this. Take for example 7700K:

7700K @4.8 GHz = 3893, 811.0/GHz
7700K @4.2 GHz = 3552, 845.7/GHz

So we have a situation where 7700K's IPC is 4% higher than.. 7700K. This is why you cannot derive ST numbers by simply dividing MT numbers by core count.

Edit: I picked a wrong example, but the point still stands. Let's pick 6800 and 6950.

6800 @3.4 GHz = 2645, 777.9/GHz
6950 @3.0 GHz = 2158, 719.3/GHz

We arrive at the conclusion that 6800's core is 8% better than 6950's core. ^^
 
Last edited:

.vodka

Golden Member
Dec 5, 2014
1,203
1,537
136
I believe it, because that's the really cheap line that cannot OC.


A300 and A320 are the non-OC chipsets. B350 can overclock. That price that's going around is for a B350 motherboard, and others too.

Also, there's the date February 24 in that link... something could be up next week.

Whether it's sensible to overclock on a $70 motherboard, well.. that's another thing. A 4c processor probably yes, without any problem... I suppose it'd be too much strain on the VRM when doing the same with a 6 core or 8 core CPU on these kind of motherboards.

We've seen such cheap motherboards survive FX8xxx OC'd chips, so anything's possible :D
 
  • Like
Reactions: Drazick
Status
Not open for further replies.

ASK THE COMMUNITY