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

Page 220 - 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.

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
So interesting stuff, after about 25 fails to post and about 15 hard locks i found something that worked higher then DDR3-2000
...
I made a chart using your data:
TjAGb4X.png
 

Justinbaileyman

Golden Member
Aug 17, 2013
1,980
249
106
Any chance those benchmarks are fake?? I mean I seen the Zen presentation,and it showed Zen beating a 6900k not a 5930k. Something is off or something is very fishy about those scores. There is just no way a fx-8350 is going to match the new Ryzen in anything considering the 6900k is 2-3 times faster and stronger then a fx-8350 in most the benchmarks I've seen including the ones on the front page of A/T main website and the Ryzen was shown beating the 6900k.
 

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136
The key to keep in mind is that every micro architecture has its design choices and compromises.

It is entirely possible that Zen beats Broadwell-E clock for clock and is competitive with Skylake in certain workloads, while also being closer to Ivy-Bridge clock for clock in other workloads.

So we will have to Wait for Benchmarks™ with some in-depth reviews after NDA expires to get the full performance profile.
 

Agent-47

Senior member
Jan 17, 2017
290
249
76
In one test. 8c 16t has a lot over 4c 8t.

People really like hyperbole in here.

Lol. People are really losing focus and context.

You said "Prime numbers are still.better then kaby lake." I said that's not true for prime numbers.

So yes in one tests. A test that requies4mb per core and has quite a bit of branching while using int64

Please don't pull a Conway on me. I want and to do good. 4 out of my 5 builds were amd.
 

Nothingness

Platinum Member
Jul 3, 2013
2,371
713
136
From PassMark's website:

The Prime Number Test aims to test how fast the CPU can search for Prime numbers, reported as operations per second. A prime number is a number that can only be divided by itself and 1. For example, 1, 2, 3, 5, 7, 11 etc. This algorithm uses loops and CPU operations that are common in computer software, the most intensive being multiplication and modulo operations. All operations are performed using 64-bit integers. This test uses about 4MB of memory per core. The specific formula used for this test is the Sieve of Atkin with a limit of 32 million.​

Emphasis mine. If these modulo operations are done using the DIV/IDIV opcodes, then it wouldn't be surprising to see them poorly optimized. These are very slow instructions even on modern Intel CPUs. I doubt that this would have been a high priority for AMD's development team, and it would not be likely to have a substantial impact on a wide variety of real-world software. Most applications that need high-speed division and modulus operations use various tricks to speed this up.
IIRC Atkin sieve modulo operations are all by constants and so should be optimized away to faster operations by any decent compiler.
 
  • Like
Reactions: Dresdenboy

.vodka

Golden Member
Dec 5, 2014
1,203
1,537
136

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,686
136
I made a chart using your data
Any thoughts on frequency and core count scaling?

Here's how the Ryzen chip fares against some other common CPUs—and I'll note that I've run PerformanceTest on my own systems here, with two options, stock and clocked at a static 3.4GHz (to match the Ryzen clock). (The AMD Ryzen, FX, and A10 results are from Passmark's database.)

9MGsVSuStcGjbbWDradh5U-650-80.png


Frequency data is all over the place:
  • i3 7350 stock score is almost equal to 3.4Ghz, essentially 0 scaling
  • i7 7700K stock score shows linear scaling vs. 3.4Ghz
  • i7 5930K stock shows 15% negative scaling vs. 3.4Ghz
Number of cores next:
  • a 100% core count increase from 7350K to 7700K (both at 3.4Ghz) brings a 100% increase in score - 100% scaling
  • a 33% core count increase from 5930K to 5960X (both at 3.4Ghz) brings a 15% increase in score - 50% scaling
It seems to me that at least the stock i3-7350K and 5930K @ 3.4Ghz data points are incorrect. The author was understandably in a hurry, as we all are when Zen is concerned :)
 
Last edited:
  • Like
Reactions: Drazick
Status
Not open for further replies.