AMD RYZEN discussion and benchmarks


#143

Decided to give RealBench a try and based on what I just saw I don’t trust it at all. It just ‘benchmarked’ my system yet only exceeded 50% processor utilization once (during encoding). Most of the tests didn’t exceed 15%.


I ran two back-to-back benchmarks and achieved very different results in the Image Editing and Encoding categories. What I find most humorous is that the "Heavy Multitasking’ barely broke %10 of my CPU utilization.

What is this supposed to tell me?

BTW, I am running a Plex server and Emby server on this system while running the benchmarks, but nothing was being streamed so the impact should be negligible.


#144

Thanks.
Regarding de-lidding, Its an expensive CPU, so if I do de-lid it, it will be with a tool for doing the job.

@Balthazar2k4
What Real Bench is telling you is that it uses consumer applications like HandBrake and GIMP to run the tests, which are real world tests. They do NOT scale well across multiple CPU cores. Handbrake for example only scales well up to 6 or 8 cores. That is why it isn’t utilising all the grunt that ThreadRipper has.

Regarding getting different results from test runs.
The most likely cause is because you are running the tests on a fully loaded Windows install, where things like AV applications will almost certainly steal CPU cycles. You would need a bare Windows install to get results that are (within margin of error) fairly comparable. That is what reviewers use when testing PC hardware.

The second reason could be your overlock is not as stable as you think it is. But I would bet its simply you are testing a fully loaded Windows install.

I’m going to use these Real Bench results and make some graphs for easy comparison. It would be great if more users here could post a screenshot of a RealBench test run. The more results I have, the better.


#145

Seems best temp to clock/voltage ratio for me is 5.1 ghz temps full load top out at 58c at 1.28 volts


#146

@Dee The GIMP results for image handling I have no issue with. The fact they vary so much between tests could very well be associated to background tasks. The Handbrake results I don’t feel are an accurate portrayal at all. They should be using more stringent encoding parameters to truly task ANY CPU. My Handbrake encoding profile pegs my processor at nearly 100% when converting 1080p AVC to 1080p AVC. To me that would provide a more accurate view across processors because only utilizing half of my CPU doesn’t offer the proper comparison (especially in the final score).

The ‘test’ that bugs me is the ‘heavy multitasking’. Clearly their version of heavy multitasking and mine are quite different. Seriously, it barely registers a blip on my machine. Once again, how does this offer the proper comparison for those of us with high core counts.

As for my overclock, it is stable. It has been run through the Prime95 FFT gauntlet for hours and it does video encoding around the clock now. I usually keep HWMonitor running in the background and the clock never drops below 3998mhz.

All I am saying is that a benchmark should max out the system in order to provide a proper comparison between machines. Otherwise, it would appear that my system isn’t significantly different from a Ryzen 7 and while I understand the basis is the same, there are x2 of them in the 1950x and from a multitasking and encoding perspective should (and does) wallop the Ryzen 7.


#147

Benchmarks are what they are. If you don’t think its relevant that’s fine.
What you have to remember is this benchmark is to test mainstream desktop PC’s, and that means it has to work on 2 core systems as well as CPU’s with more cores.
This time last year, top of the range mainstream desktops had probably 4 cores, 6 cores on what was classed as high end desktop.
Now we have mainstream desktops with 6 and 8 cores. Threadripper is not a mainstream product. Benchmarks will probably take some time to catch up with the latest and greatest CPU’s.

Having said that, your encoding score is well down on what some reviewers were getting with 1950X.


#148

The good old days when intel was the only option you either had a 4c/4t or 4c/8t for the mainstream Z series, for more you had to go for the X series, pricing was very high, also if you were a gamer you didn’t see any improvement from a 4C to a 6C CPU. Also AMD CPUs were outdated and their performance was mediocre.

Now things are moving forward and we have more options, if anyone had said to me that we could get a 16C/32T CPU for less 1000 euro or dollars I would have laugh and said that you need to see a doctor.
Just remember the pricing of the 10C/20T CPU that intel released for the X99, i remeber it was close to 2000 euros, or US dollars.

When it comes to benchmarks, they can be fun as long as you do it for fun. As @Dee said it will take sometime, not only for the realbench from ASUS, but also from software to catchup with the new multi core CPUs, I am sure that a lot of the software out there had poor multicore optimization, and most of the benefit form the higher clock speed.


#149

Well, my Geekbench scores are right inline with that same reviewer at 4648 single core and 32881 multi. I have run RealBench now six times with minimal background apps running and I am still getting wildly inconsistent scores. Well beyond margin of error.


#150

CoffeLake 8700K @ 4.8Ghz

Real Bench score


Click on the picture for full size image.


#151

Very interesting results tbh 166,333 vs a threadripper with a 1080Ti scoring 169.176 and vs your ryzens 143,325 score

some decent results :smiley:


#152

Ryzen vs Coffeelake
FutureMark PCMark 10 Pro.

RyZen 7 1700 @ 3.8Ghz


Click the picture for full size image

CoffeeLake 8700K @ 4.8Ghz


Click the picture for full size image


#153

Benchmarks have their limitations, especially ‘real world’ benchmarks.

The real world benchmarks i’m using in this post in most part, have a workload to mimic a typical consumer usage pattern. That means they have to work on a system with only two CPU cores, and all the way up to perhaps 16 CPU cores.

Consequentially, that means they will not in most part take advantage of a CPU with 8 or more cores. So keep that in mind when viewing these graphs. The only benchmark that i’m showing here that will measure the full potential of a high CPU core count is CineBench R15.

So lets start.

For all the results shown here, both CPU’s were overclocked to their maximum stable, and safe clock speed for 24/7 use. In the case of the AMD RyZen R1700 that was 3.8GHz, and in the case of the Intel i7 8700K 4.8GHz. Both systems used identical DDR4 3200MHz DRAM at 3200MHz. Both systems also used an identical NVidia GTX1060 GPU.

CineBench R15

cinebench graph

With its CPU clock speed advantage, the Intel 8700K was the fastest in the single threaded test. In the Multi threaded test, and RyZen 1700 with its 8 CPU cores easily outperformed the Intel 8700K even though the 8700K had a 1GHz clock speed advantage.

Asus RealBench

I have added a result for RyZen ThreadRipper in this test, kindly provided by Balthazar2k4 .

RealBench has a fairly typical consumer grade workload, and will therefore not take full advantage of any of the CPUs under test.
I will leave you to draw your own conclusions from the results.

FutureMark PCMark 10 Professional

PCmark 10 graph

PCMark 10’s workload is typical to one you would find an your home or office. Suffice to say, it will not take full advantage of these multi core CPU’s.

Once again, i will leave you to draw your own conclusions on the results.

My own thoughts.

For a typical home user, CPU clock speed and IPC are still the winner. The Intel 8700K with its 1GHz clock speed advantage is in most part the fastest. It is also more expensive than the RyZen 1700.

However, if you can take advantage of the RyZen 1700’s extra two CPU cores, like I have done in my studio, then RyZen 1700 is the more powerful CPU, and has given my studio a new lease of life.


#154

their all down to which use you wanna do mine is mainly a gaming orientated system so 8700K was the perfect choice for me with a 1080Ti


#155

I’ve ordered a de-lidding kit for my CoffeLake 8700K.

Anyone considering buying and doing any overlock on the 8700K is going to need very good cooling, and to get the most out of the 8700K will require de-lidding it, IMO.


#156

Thats about as easy as it can get Wendy. LOL, even with my Rheumatoid Arthritis I could probably do it.


#157

I figure it has to be easier and safer than doing it with a razor blade. Both for the CPU and my fingers. :slight_smile:


#158

very easy with that tool i gave mine away before i got my 8700k “doh!!”


#159

They sell them at Newegg in the USA for 44.00. I guess 44 bucks is a fair price to pay for something I might use once a year.


#160

Remember you will need to add liquid metal paste between the CPU die and the heat spreader, then a thermal paste such as Noctua NT-H1, between the heat spreader and your CPU cooler. Add also some high temperature silicon glue to stick the heat spreader back on after de-lidding.
You’ll also need 100% alcohol to remove the Intel TIM, and some plastic razor blades to remove the silicon glue that Intel applies to these CPU’s.

So the cost of de-lidding isn’t only a matter of buying a de-lidding tool. Unless you already have the other stuff.


#161

thermal grizzly is the best liquid metal to use


#162

Thanks Wendy, I have mostly everything but the delidder.