Overclock.net banner

1 - 17 of 17 Posts

·
Registered
Joined
·
345 Posts
Discussion Starter #1 (Edited)
Greetings! Over the past 6 months i've been assembling and selling computers, so i had lots of processors to test! This benchmark is dedicated to productivity applications only!!!

This CPU benchmark is dedicated to folks who are looking a for a budget friendly workstation build - for those who can't afford new AMD X399 or Intel X299 systems with Ryzen Threadripper or Core i9 processors, but are checking out Z370 and X470 platforms with flagship Core i7 8700K and Ryzen 7 2700X processors. Not only this will be a massive direct battle between these two chips, but also included are the other alternative chips, like Ryzen 7 1700 and Core i5 8600K, which cost less and can provide enough performance for your WS rig. Without a doubt, older Core i5, Core i7 and Xeon processors have been included in the benchmark to provide comparison and perhaps competition to the modern chips.

PROCESSORS and MOTHERBOARDS

All overclockable processors have been overclocked to their maximum capabilities using Noctua NH-D14!!!

Core i7 8700K - all cores overclocked to 5 GHz and 4.8 GHz for AVX tasks {Asrock Z370 Taichi}
Core i5 8600K - all cores overclocked to 5 GHz and 4.8 GHz for AVX tasks {Asrock Z370 Taichi}
Core i7 6700 - all cores default turbo boost 3.7 GHz {Gigabyte X170-Extreme ECC}
Core i7 5775C - all cores overclocked to 4.3 GHz {Gigabyte Z97-Gaming 3}
Core i5 4570 - all cores default turbo boost 3.4 GHz {Gigabyte Z97-Gaming 3}
Ryzen 7 2700X - all cores overclocked to 4.2 GHz {MSI X370 Gaming Plus}
Ryzen 7 1700 - all cores overclocked to 3.8 GHz {MSI X370 Gaming Plus}
Xeon E5-2680 V4 ES 14 core/28 thread - all cores default turbo boost 2.3 GHz {Asrock X99 Extreme6}
Xeon E5-2630 V4 ES 10 core/20 thread - all cores default turbo boost 2.3 GHz {Asrock X99 Taichi}
Xeon E5-2620 V4 8 core/16 thread - all cores default turbo boost 2.3 GHz {Asus X99 M-WS}

OTHER PARTS

RAM: 4X8 DDR3 1600 MHz CL8 and 2X16 GB DDR4 2133 MHz CL14.
HDD: WD SE WD2000F9YZ 2 TB SATA3 7200 RPM 64 MB
GPU: COLORFUL GEFORCE GTX 1080 TOP V2 8 GB
WINDOWS 7 X64 SP1
NVIDIA FORCEWERE 398.36


At first i was assembling workstations with Broadwell-EP Xeon ES processors, since they are being sold real cheap at ebay. Even the 14 core/28 thread CPU can be found for 400 - 500 $, while 6 months ago it was going on sale for 700 $. The 18 core/36 thread Broadwell-EP Xeon was being sold for 1000 $, so it was a too expensive shot. That got me thinking:

- ''what if it ain't worth buying those Xeons, what if Core i7 8700K or Ryzen 7 2700X are actually faster, what if i am making a mistake by assembling workstation computers with Xeon chips''?

Ye, you get the idea where this benchmark came from now...

These processors have been tested in real time programs with custom presets as well as in standard synthetic benchmarks:

CPU-Z
Cinebech R15
Vray
Corona
Handbrake video compression (multiple tests)
Blender rendering tests (multiple tests)
Irfan photo conversion tests (multiple tests)
SPEC 3D rendering simulation tests (multiple tests)


OPEN THE VIDEO TO SEE THE RESULTS:

 

·
PC Enthusiast
Joined
·
1,075 Posts
Worth noting that 2133Mhz DDR4 slows down a stock 6700K let alone a 6 or 8 core cpu. We saw this in the anandtech launch article where a 4790K with 1600mhz DDR3 was actually faster than a 6700K with 2133mhz DDR4.

if you divide the rated speed by the CAS latency then you can get a rough guide to the performance of the ram:

1600 / 8 = 200
2133 / 15 = 142.2
2133 / 14 = 152.36~

To get DDR4 that's as fast as your 1600C8 DDR3 you need something like 3000C15.
 

·
Registered
Joined
·
345 Posts
Discussion Starter #3 (Edited)
Worth noting that 2133Mhz DDR4 slows down a stock 6700K let alone a 6 or 8 core cpu. We saw this in the anandtech launch article where a 4790K with 1600mhz DDR3 was actually faster than a 6700K with 2133mhz DDR4.

if you divide the rated speed by the CAS latency then you can get a rough guide to the performance of the ram:

1600 / 8 = 200
2133 / 15 = 142.2

To get DDR4 that's as fast as your 1600C8 DDR3 you need something like 3000C15.
That's right! Only your calculations are a bit bizzare. It goes like this:

For DDR4: 1/2133 MHz x CL14 = 0,00656 seconds.
For DDR3: 1/1600 MHz x CL8 = 0,005 seconds.

So you could argue that Core i7 5775C and Core i5 4570 had a bit of an advantage! ;)

I still have that Core i7 8700K, if i get the chance to test all this again with DDR4 3000 MHz CL15, i will report immediately how much faster Core i7 8700K can be!!!
 

·
PC Enthusiast
Joined
·
1,075 Posts
I have a 4GHz R7 1700 and a 5.2Ghz i5 8600K, both with ram at 3333C14.. I can turn them down a bit to match your clocks and run the ram at 2800C14 for a rough comparison (different boards, OS installs etc will change the result slightly)

The reason for the funky simpler ram calculation is just to make it easy to remember and give a nice higher is better result to compare with. It works well enough for showing the performance difference at glance when comparing ram kits online.
 

·
Registered
Joined
·
345 Posts
Discussion Starter #5
I have a 4GHz R7 1700 and a 5.2Ghz i5 8600K, both with ram at 3333C14.. I can turn them down a bit to match your clocks and run the ram at 2800C14 for a rough comparison (different boards, OS installs etc will change the result slightly)
So you can at least make the 3 synthetic benchmarks - CPU-Z, Cinebench R15, Vray. You could do Corona as well if you had a GTX 1080. GPU impacts CPU performance in that program.

Leave your RAM at 3333 MHz CL14 and do the tests with Ryzen 7 1700 3.8 GHz (all cores) and Core i5 8600K 5 GHz (all cores). It will be interesting to know how pure memory bandwidth improves the performance when the latencies are the same (2133 MHz CL14 vs. 3333 MHz CL14)!!!
 

·
PC Enthusiast
Joined
·
1,075 Posts
Well I did run a bunch of tests.. and the day after the i5 pc doesn't want to turn on. I think something shorted somewhere because it shut off after a few minutes of idling on the desktop and the psu seems to trip and turn off half a second after I power it on =\

Here's the screenshots from the R7 1700 testing at 3.8Ghz with ram at 2133c14 and 3333c14.. not much in it for cpuz but a few seconds in vray and a bump up for cinebench, along with a noticable difference in 7zip and superposition (with gpu swapped to a 1080ti and held at 1924Mhz for both tests to rule out cooling boost)

https://imgur.com/a/LS1uASL
 

·
DRAM Obsessor
Joined
·
1,033 Posts
Memory frequencies work very closely to the cpu with how Ryzen was designed (Soc and Infinity Fabric). The higher your Memory clock the lower the latencies and the faster the cpu becomes and the faster the cpu becomes the lower the memory latencies become. Using MLC my latency is around 57ns and using Passmark Performance Test or Memtest86 latency tests they are around 46ns with Ryzen 2700x @ 4.45ghz. Ill post some benchmarks in a bit at that freq with 3533 cl14. Which are all my daily setup but can do 4.55ghz as well later for max OC setup for comparison.
 

·
Registered
Joined
·
345 Posts
Discussion Starter #8 (Edited)
Well I did run a bunch of tests.. and the day after the i5 pc doesn't want to turn on. I think something shorted somewhere because it shut off after a few minutes of idling on the desktop and the psu seems to trip and turn off half a second after I power it on =\

Here's the screenshots from the R7 1700 testing at 3.8Ghz with ram at 2133c14 and 3333c14.. not much in it for cpuz but a few seconds in vray and a bump up for cinebench, along with a noticable difference in 7zip and superposition (with gpu swapped to a 1080ti and held at 1924Mhz for both tests to rule out cooling boost)

https://imgur.com/a/LS1uASL


Thank you anyway!!! Superposition seemed to have the only real improvement. I can not believe how lame the performance was in other synth tests. I expected the 3333 MHz RAM to vam-bast 2133 MHz RAM...

But of course this tells nothing until some real productivity benchmarks are in place. Unfortunately, by this time i have sold every last CPU from this benchmark.

Memory frequencies work very closely to the cpu with how Ryzen was designed (Soc and Infinity Fabric). The higher your Memory clock the lower the latencies and the faster the cpu becomes and the faster the cpu becomes the lower the memory latencies become. Using MLC my latency is around 57ns and using Passmark Performance Test or Memtest86 latency tests they are around 46ns with Ryzen 2700x @ 4.45ghz. Ill post some benchmarks in a bit at that freq with 3533 cl14. Which are all my daily setup but can do 4.55ghz as well later for max OC setup for comparison.
Perhaps you can provide some real productivity tests with different RAM speeds on same CPU OC. What tests can you provide? Can you do some Blender demo rendering, image conversion or video compression?
 

·
DRAM Obsessor
Joined
·
1,033 Posts
Well I did run a bunch of tests.. and the day after the i5 pc doesn't want to turn on. I think something shorted somewhere because it shut off after a few minutes of idling on the desktop and the psu seems to trip and turn off half a second after I power it on =\

Here's the screenshots from the R7 1700 testing at 3.8Ghz with ram at 2133c14 and 3333c14.. not much in it for cpuz but a few seconds in vray and a bump up for cinebench, along with a noticable difference in 7zip and superposition (with gpu swapped to a 1080ti and held at 1924Mhz for both tests to rule out cooling boost)

https://imgur.com/a/LS1uASL


Thank you anyway!!! Superposition seemed to have the only real improvement. I can not believe how lame the performance was in other synth tests. I expected the 3333 MHz RAM to vam-bast 2133 MHz RAM...

But of course this tells nothing until some real productivity benchmarks are in place. Unfortunately, by this time i have sold every last CPU from this benchmark.

Memory frequencies work very closely to the cpu with how Ryzen was designed (Soc and Infinity Fabric). The higher your Memory clock the lower the latencies and the faster the cpu becomes and the faster the cpu becomes the lower the memory latencies become. Using MLC my latency is around 57ns and using Passmark Performance Test or Memtest86 latency tests they are around 46ns with Ryzen 2700x @ 4.45ghz. Ill post some benchmarks in a bit at that freq with 3533 cl14. Which are all my daily setup but can do 4.55ghz as well later for max OC setup for comparison.
Perhaps you can provide some real productivity tests with different RAM speeds on same CPU OC. What tests can you provide? Can you do some Blender demo rendering, image conversion or video compression?
Yes, I’ll run some rendering at 4.4ghz+ with various ram speeds from 3200-3533 cl14 and get the information back. Already running the 3466 run right now but blender classroom takes a bit so to do all the test will take some time.
 

·
DRAM Obsessor
Joined
·
1,033 Posts
Ok, well I ran benchmarks most of the day as well as made memory setups from scratch and tested them for errors and the results were quite surprising to say the least. As far as rendering is concerned, there is absolutely no impact whatsoever from memory frequency. Well, at least from my setup there isnt. My setup is a 2700x @4.45ghz most of the time as with XFR/PBO on Ryzen the maximum overclock is determined at bootup from how cool your system is. The cooler the system is the higher the max All core and 1-4 core overclocks will be up to a maximum of 43.5x the base clock which mine was 102mhz during these benchmarks. All core clocks were 4333mhz-4384mhz as it got cooler in the evening my system ran cooler giving 50mhz more to the all core. The 1-4 core boost was 4435mhz and that remained as such as it was at the 43.5x max, the only way to increase it is through the base clock once you have met a hard cooling limit for the extended boost clocks. Anyway, aside from cpu Im running a 1080 ti with an unlocked bios from the XOC card allowing the max of 1.2v and the only way to go further as far as I know is to hard mod the card through variable resistors soldered to the card. Memory is 3200c14 g.skill TridentZ Samsung BDie, I started with 3200c12 then jumped to 3466c14 then to 3333c14. I had planned on testing literally every strap from 3200c12 all the way to 3533c14 but I soon found out that was just going to be a waste of time. As far as my system goes, I am running about as a system like this can run on AIO cooling and would get into more extreme setup going further so maybe that contributed to the absence of increased performance from increased frequency on the memory. On blender there was like 3 seconds change from 3200c12 to 3466c14 and around 100 point difference in Timespy which I feel is close to margin of error for both. Only place I saw anything noticeable was decompression/compression and file transfers but that was at best, 1.5% increases. This was with me tuning all timings to maximum performace and testing each for roughly an hour for memory errors. While they are not considered "rock solid" Ive tuned timings on everyday Ryzen for nearly 2 years on various cpus and I can say with 99% certainty that these are good to go. Aside from the lack of change which I do feel like is due to my system already nearly maxed out so any gains should be minimal coming from something such as frequency increase to memory, I did notice most of all the change in system response and boot times seemed to be a bit faster although I have a 960 pro as my boot drive so im not sure how much can be improved on that either. There is definitely a difference with application response and windows load times that is noticeable. Aside from that there was nothing massive that changed at all from 3200c12 to 3466c14 and after spending that long on those results I determined there was no reason to test the other straps between as they would yield something similar. Now there may be something noticeable from stock memory settings (2133mhz) and 3533c14 tuned and I could test that if you wanted but I still feel it wouldnt be something that would make you gasp. I will leave the screenshots I took before I stopped in spoilers below for reference along with the timings I used. I think someone with a more mild system would be able to better show what impact ram has on a Ryzen system, getting more performance from mine is like trying to squeeze water from a rock unless I clock the cpu higher and get sub-ambient cooling. Sorry for the lack of a better result for you but at the very least you have something to reference for information on higher clocked systems.


3200c12



3333c14 (this is the strap where i noticed things werent changing and just stopped screenshotting)



3466c14
 

·
⤷ αC
Joined
·
11,240 Posts
@ Marshock Can't believe I missed this post (as it involves workstation benchmarks), but if your Ryzen is running DDR4 2133 it's being severely held back. 3200MHz CL16 at a minimum , 3200C14 and 3466C14 if your chip can do it.
 

·
Registered
Joined
·
345 Posts
Discussion Starter #12 (Edited)
Ok, well I ran benchmarks most of the day
CJ, you put alot of effort, and seems it did not pay of with minor improvements.... But what i really wanted you to do is test 2133 MHz DDR4 vs. 3200 MHz DDR4 at the same CL14.

We need to know how much of an improvement does faster RAM offer in:

Blender CPU based rendering tests
Handbarke video compression tests with H.264 1080p, H.264 2160p, H.265 10 bit 1080p presets.
Photo archive conversion tests - preferably 8K resolution - especially from jpg to png.
Perhaps some archiving stuff
SPEC View Perf tests (even though these are purely video card dependent, still would be nice how RAM speed changes things)

Possible?

Anyway, you realize how great Ryzen 7 2700X is? In my tests with lowish RAM speeds it was in most of the tests ahead of Core i7 8700K {with all cores at 5 GHz} and the 28 thread Xeon E5-2680 V4 {with all cores at 2.3 GHz}.

@ Marshock Can't believe I missed this post (as it involves workstation benchmarks), but if your Ryzen is running DDR4 2133 it's being severely held back. 3200MHz CL16 at a minimum , 3200C14 and 3466C14 if your chip can do it.
Yes, you realize that all processors were equally bottle-necked? Which means it was fair game, because both the Intel and AMD were tested with basic 2133 MHz DDR4, so non had the advantage anyway. We all realize both Intel and AMD could show even better results with higher RAM speeds.
 

·
DRAM Obsessor
Joined
·
1,033 Posts
Ok, well I ran benchmarks most of the day
CJ, you put alot of effort, and seems it did not pay of with minor improvements.... But what i really wanted you to do is test 2133 MHz DDR4 vs. 3200 MHz DDR4 at the same CL14.

We need to know how much of an improvement does faster RAM offer in:

Blender CPU based rendering tests
Handbarke video compression tests with H.264 1080p, H.264 2160p, H.265 10 bit 1080p presets.
Photo archive conversion tests - preferably 8K resolution - especially from jpg to png.
Perhaps some archiving stuff
SPEC View Perf tests (even though these are purely video card dependent, still would be nice how RAM speed changes things)

Possible?

Anyway, you realize how great Ryzen 7 2700X is? In my tests with lowish RAM speeds it was in most of the tests ahead of Core i7 8700K {with all cores at 5 GHz} and the 28 thread Xeon E5-2680 V4 {with all cores at 2.3 GHz}.
I do realize how great it is. I have used both flagships since they launched and will get Ryzen 2 flagship once it’s ready. Here’s my take on ram and Ryzen. It makes great improvements but those improvements aren’t going to show in rendering times near as much as file transfers and compression/decompression. I can tell a great difference is made with ram just from 3200 to 3466 from the feel of the cpu response and overall system response. As far as showing that on paper...it seems difficult to do. I’ll try to see if more of a difference is made vs stock ram settings seeing as 3200 to 3466 with both profiles optimized aren’t really going to make worlds of difference anyway. It’s really the latency changes that show the most dramatic improvements. Bandwidth numbers are already very good so sacrificing a bit of bandwidth for lower latency is always a good idea when tuning timings. Lowest Aida latency I get is 57ns and I’m hoping for 45ns on Ryzen 2 @ 3800c14. It will be blazing fast and only get faster in multithreading as the latency drops. Frequency just isn’t as much of a factor vs having the late cues and clock cycles low. I don’t see a point in even traveling higher than cas 14 until 4000mhz or better is achieved with tight timings. There’s just no proof that the higher frequency is a benefit over the slower clock cycles that you will most certainly have with cas 15 or higher at those frequencies. Most depends on the ability to get those subtimings optimized and I just haven’t seen that yet.
 

·
Registered
Joined
·
4,613 Posts
Did you run any single-threaded benchmarks?
 

·
DRAM Obsessor
Joined
·
1,033 Posts
Did you run any single-threaded benchmarks?
I did run cinebench on single threaded and cpuz single threaded bench and saw little no change. Really, what it is looking like is that rendering just isnt relying on memory enough to see anything dramatic within a small (3200 to 3466) change in frequency. It surely relies on it for bandwidth and to a smaller extent, latency but the bandwidth on Ryzen at 3200 is already really good and its likely that performance from bandwidth has reached a plateau for those particular tasks hence having little to no change in results. Things do get more apparent in benchmarks designed to run specific memory sensitive tasks like SiSoft Sandra or PassMarks memory test within their benchmark software. There arent many programs that are designed to do so beyond those from what ive seen. Real World tests outside of gaming would more or less be ones with file transfers and compression/decompression tools. Benchmarks are easily found as anything like what was mentioned above and others that rely heavily on mathematical processing like SuperPi. As stated before, the real noticeable areas in the real world are app responses and boot times and overall just how the system feels much snappier and load times are considerably shorter. I wouldve done more testing that day but benching that long and seeing those results were disappointing. What was a bit odd though is that in gaming benchmarks such as Firestrike, specifically the Physics portion reacts quite well to memory which made me think that rendering would yield similar results but that isnt the case. Different workload is the reason i presume. There is no doubt that memory OC affects Ryzen quite well but that specific margin I tested didnt. The results are probably more prominently shown with stock settings and max OC on memory with optimal timings. Ill try that soon and see what the difference is. theres surely going to be a difference since the bandwidth and latency is miles apart
 

·
Registered
Joined
·
345 Posts
Discussion Starter #16
UPDATE

Intel Xeon E5-2683 V4 tested. 16 cores/32 threads, 4 MB L2, 40 MB L3. Another engineer sample, that manages 2.3 GHz turbo boost on cores. Since i am on a new Windows 10 computer, i could not test the full list of the previous benchmarks, so here what's been tested:

CPU-Z MULTI THREAD 5070 points.
CINEBENCH R15 1940 points.
V RAY 1:09 s.
CORONA 1:58 s.
BLENDER BMW 4:47 s.
BLENDER CLASSROOM 15:19 s.
BLENDER PAVILLON 16:46 s.
BLENDER SPLASH 9:05 s.
BLENDER BENCHMARK 36:58 s.
 

·
Banned
Joined
·
403 Posts
Well they need to do Something..Gonna have 9900k Killers lol
 
1 - 17 of 17 Posts
Top