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

Page 216 - 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.

Agent-47

Senior member
Jan 17, 2017
290
249
76
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.
you qouted the 8 core results which had similar clocks. IPC do not scale linearly at higher clocks, but i scaled down, so if my scores were biased, they were biased towards intel marginally :p

also the more important results are the 4c and even the 6 core. the 6 core was clocked similarly and still points towards the same result
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.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,670
1,250
136
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.

That's more of a good example of why you can't take the reported frequency on some graph at face value in the age of turbo clocks, and maybe the dangers of mixing and matching test environments too. In general, the stock results probably have less uncertainty because we can assume certain things about how turbo is behaving.
 

Crumpet

Senior member
Jan 15, 2017
745
539
96
VideoCardz.com is claiming that those benchmarks are done on - ASUS B350 PRIME PLUS motherboard
 

Agent-47

Senior member
Jan 17, 2017
290
249
76
What is 7700k's 4 core turbo bin? 3dmark physics is load light enough for generally all CPUs to use their max turbo bin. Napkin math suggests that it should 4.4Ghz.

3.5Ghz, actually, checked wiki.

does TH run benches with turbo active? its a little strange if they do because than the stated clocks are misleading considering overclocked CPUs do not turbo any higher.

if so, in these benchmarks zen is faster than BWE and 1.5% slower than SK.
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
That's more of a good example of why you can't take the reported frequency on some graph at face value in the age of turbo clocks, and maybe the dangers of mixing and matching test environments too. In general, the stock results probably have less uncertainty because we can assume certain things about how turbo is behaving.
I am in agreement. That just adds to the reason why the 2nd graph is silly and misleading.
 

HurleyBird

Platinum Member
Apr 22, 2003
2,670
1,250
136
I am in agreement. That just adds to the reason why the 2nd graph is silly and misleading.

I agree that it is silly and misleading. If all processors involved had the same number of cores/threads then I wouldn't have as much of a problem with it.
 
Last edited:

CatMerc

Golden Member
Jul 16, 2016
1,114
1,149
136
Looks like all but the 8 core non OC are running at 4GHz. Probably XFR at work.

20249 / 8 (cores) / 4 (GHz) = 632.7

Now if we plop those numbers onto the 4 and 6 core, we get...
632.7*6*4 = 15,184 --- The score in the benchmark is 15,271
632.7*4*4 = 10,123 --- The score in the benchmark is 10,177


The only chip that doesn't line up is the 8 core without the 4GHz OC.
17,878 / (632.7*8) = 3.53GHz

Edit:
That implies a near 100% linear scaling, which is somewhat difficult to accept. Not impossible depending on other factors, but very rare indeed. Clearly the Intel ones do not scale linearly.

Thinking about it SMT might also factor into it and I have no idea how that works out.
Well, taking into account that the 6900K scored about 13% less than my calculations would predict, meaning there's 13% penalty for two extra cores, I worked back to get the 6 core clockspeed.

3.56GHz, which oddly lines up with the 8 core non OC speed I calculated.
Could be we just found the all core turbo of Ryzen.

(18,961 / 6 / 4.2) = 741.7
741.7 * 8 * 3.5 (all core turbo) = 21,067
21,067 / 18,635 = 1.13

Ryzen 6 Core:
15,271 / (632.7 * 6 * 1.13) = 3.56GHz
 
Last edited:

HurleyBird

Platinum Member
Apr 22, 2003
2,670
1,250
136
Wikipedia, it usually has a good track record on these.

I checked Wikipedia, and it lists 3.6GHz, but as far as I can tell it does not say how many cores that is. Because the base frequency is only 3.4GHz, I'm assuming it's all cores, but I could be wrong.
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
you qouted the 8 core results which had similar clocks. IPC do not scale linearly at higher clocks,
Neither does throughput per core. (and what 8 core results? I initially used a 7700K in my example)
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
Looks like all but the 8 core non OC are running at 4GHz. Probably XFR at work.

20249 / 8 (cores) / 4 (GHz) = 632.7

Now if we plop those numbers onto the 4 and 6 core, we get...
632.7*6*4 = 15,184 --- The score in the benchmark is 15,271
632.7*4*4 = 10,123 --- The score in the benchmark is 10,177


The only chip that doesn't line up is the 8 core without the 4GHz OC.
17,878 / (632.7*8) = 3.53GHz
That implies a near 100% linear scaling, which is somewhat difficult to accept. Not impossible depending on other factors, but very rare indeed. Clearly the Intel ones do not scale linearly.

Thinking about it SMT might also factor into it and I have no idea how that works out.
 

Crumpet

Senior member
Jan 15, 2017
745
539
96
I'm kind of interested in Asus's 3d printing mounts on the top end motherboards, so that may sway it for me. I have a good friend with a good quality 3d printer, so the ability to make my own mount may swing it. Especially if I can design a shroud that holds a CPU cooling fan over an AIO waterblock for VRM cooling.
 

CatMerc

Golden Member
Jul 16, 2016
1,114
1,149
136
That implies a near 100% linear scaling, which is somewhat difficult to accept. Not impossible depending on other factors, but very rare indeed. Clearly the Intel ones do not scale linearly.

Thinking about it SMT might also factor into it and I have no idea how that works out.
Well, taking into account that the 6900K scored about 13% less than my calculations would predict, meaning there's 13% penalty for two extra cores, I worked back to get the 6 core clockspeed.

3.56GHz, which oddly lines up with the 8 core non OC speed I calculated.
Could be we just found the all core turbo of Ryzen.

(18,961 / 6 / 4.2) = 741.7
741.7 * 8 * 3.5 (all core turbo) = 21,067
21,067 / 18,635 = 1.13

Ryzen 6 Core:
15,271 / (632.7 * 6 * 1.13) = 3.56GHz
 

Agent-47

Senior member
Jan 17, 2017
290
249
76
That implies a near 100% linear scaling, which is somewhat difficult to accept. .

Based on your math:

7700K @4.8 GHz = 3893/4.8, 811.0/GHz
7700K @ststockHz = 3552/4.4, 807/GHz

Or a 0.5 % difference. That's winthing uncertainty bounds.
 

CatMerc

Golden Member
Jul 16, 2016
1,114
1,149
136
Based on your math:

7700K @4.8 GHz = 3893/4.8, 811.0/GHz
7700K @ststockHz = 3552/4.4, 807/GHz

Or a 0.5 % difference. That's winthing uncertainty bounds.
Core scaling, rather than clock scaling. See my second post.
 

unseenmorbidity

Golden Member
Nov 27, 2016
1,395
967
96
9bXzF5B.png
 
  • Like
Reactions: psolord

Agent-47

Senior member
Jan 17, 2017
290
249
76
Core scaling, rather than clock scaling. See my second post.
Yes. I agree with that.

That's why I compared cpuz with same number of core to compute IPC difference .

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.

Granted a 6c ryzen with +100mhz clock scores same as the 4c ryzen, but the percentage nonlinearity is minimal and technically eliminating under the assumption that both and and Intel scales similarly with number of cores.

If you factor in boost clocks the difference is less than 2%.
 
Status
Not open for further replies.