X variants originally boosted higher and had higher XFR which meant they had better clockspeeds in the scenarios you were mentioning.
There are many heavily threaded games today… so they don’t load just 1 core but 2-6 cores typically.
I will report back on how mine performs with my 1070. I don’t care to have 144hz with consistent frametimes but 90-100 with medium settings would be nice in most games. If not I will be ordering a 5700xt.
What games are you playing? I can tell you that my minimums in HoTS went to around 75 with averages around 110 now and peak 167, when before the minimums were around 50 with averages around 90 and peak around 144.
Pubg/Squad/Insurgency Sandstorm
Well yeah if you really want an Asus board,
with all the features then you have to pay for it indeed.
But of course Asus also has cheaper X570 options, which are still decent.
The Asus X570 TUF Gaming plus for example is not a bad board.
And comes with a reasonable price.
But of course you could also check out other brands as well.
To me the most interesting X570 boards that Asus is currently selling are the,
X570-E Strix, X570 TUF Gaming plus and the Pro X570 WS ACE.
But this kinda depends on where you life really.
In my area the price difference between the X570-E Strix and the Crosshair 8 Hero is still too big.
3 Likes
Yup.
Allthough some max out 4 cores (Fallout 4 for example), and you would then loose performance having something like a browser open on a 4c machine.
And this is what I don’t understand. Sure, multithreading is “hard” (harder than don’t care single core f-it).
But why not load stuff that is needed soon with a new core and then just change over?
I did that for a project once, worked and was realy simple to do.
HotS is a joke.
1700x + 5700XT yields me 3 to 300 FPS averageing arround 80.
Might want to take a look at my experience with one. Drivers and Windows (and some games) are not there yet.
1 Like
I wont be able to snag one til Christmas anyways so hopefully if needed by then most of the stuff is ironed out.
2 Likes
I am getting 75-166fps now w/ 110 avg (4K ultra settings) vs 50-144 w/ 90 avg.
My FPS will be different due to Vega + 4K Ultra but still…
Maybe by then they will be past stepping 0… I’m still not sure what is up with this. I’ve never received a non-ES processor with a stepping of 0.
It really seems like they decided to ship the models we saw at CES in january.
This is hardware porn at it’s best! Gotta love those Noctua colors repesented there!
1 Like
A Video disussing a sample of Results from DerBauer. Great Video with some interesting Points:
TLDW: between 50 and 95% of CPU’s (depending on SKU) aren’t hitting their advertised Boost Clocks, independent of MB. No MB Manufacturer or AMD have any clue how it could be fixed at the moment.
He also states, that all Ryzen CPU’s are great CPU’s and that AMD probably should have just advertised lower Boost Clocks. This would have prevented a lot of the “Problems” we now have.
4 Likes
Just by the video title, I see a problem: The test parameters are way out there. No control over anything except what CPU is used. Mainboard, Windows version, drivers, background tasks, nothing was controlled for.
Problem is, without knowing exact software that ran while boost clock was measured, there is no way to tell if the numbers are “wrong” or if the test is flawed.
He adresses the Mainboard variance in the video. If X570 is anything like Intels “Multicore enhancement”, then the test is a bunch of data points without any value.
Could also be down to baseclock variance.
Edit:
Bias is also addressed, but not accounted for.
Edit2: I think in the ASRock AM4 server mainboard thread, there was mention of PSU related problems. Any reports of similar behaviour on X570?
1 Like
Please watch the video. It was explicitly stated that
- AMD’s recommended settings and ways to measure where used (Cinebench R15 Single threaded and HWMonitor)
- Outlier results where thrown out (Chilled Water, anything with PBO)
- Windows Version wasn’t controlled as that wasn’t needed. AMD didn’t say Ryzen 3000 isn’t compatible with Win 7, so performance should be independent of OS
- For singlethreaded Maximum Boost Clock on 6-12 Core systems, Background tasks and such should have negligable impact. We aren’t looking for scores here, just for the absolut Maximum Clockspeed a Single Core can hit in a Cinebench Run.
- Mainboards impact is an unknow. He specfically adressed the Board that hardware unboxed said was the best out of all tested, and the results reflected the overall results. This means that we can assume boards have minimal impact.
In the end, he shared his results with all AIB’s and AMD. I’m pretty sure if there was ANYTHING wrong with the methodology, AMD would be happy to point that out to invalidate the results. Having a video out there stating that “something is wrong and AMD has no clue” after having talked to AMD leads me to believe it’s pretty reasonable those results are close to correct.
We’ll never get a lab test of that magnitude, but i also think that’s not needed really. After all, we care for what those Chips do in our PC’s, not in a lab.
Edit: Some of your Edits. All of those points are valid, but not really relevant. In the end all we as consumers care for is “do we get the advertised performance” and that’s what was measured. And at the moment, a large portion of customers don’t get the performance they paid for. This might be different in a controlled lab environment, but for us customers, that’s largely irrelevant.
1 Like
In progress.
Good
Without fierce control, “outlier” is a broad (or narrow) term.
I beg your pardon?
A 34 point swing may not seem big, but that is the same CPU on the “same” OS.
Even worse, Win10 has random whatever going on all the time.
That strongly depends on scheduler, power/VRM condition, and the boost algorythm itself.
What now? Unknown or minimal?
Just as one does not buy a car for engine RPM, one should not buy a CPU for MHz.
With all the beancounters hanging arround AMDs HQ, I wonder why they even bother with consumers.
Related
2 Likes
This whole test is not about points at all.
I have no clue. Depends on who you ask. Again, that’s not the point though.
AMD sells a Product saying it’s maximum Boost Clock is 4,2Ghz for example. They don’t qualify that with “On some boards, when you run it on Windows 10 with latest updates and aren’t running anything other than Cinebench”.
The point is, that those advertised speeds should be possible on any supported configuration, and so far, all the tested configurations are supported. All OS Versions, and all Mainboards tested are officially supported.
It’s also not the case that 100% of chips on 40$ Mainboards with stock coolers and no airflow should be able to do this.
But if 95% of 3900X Chips (where we reasonably can assume higher end Boards, RAM, cooling and recent OS Versions) aren’t hitting close to their advertised speeds, something is off. And AMD as well as Mainboard manufacturers have admitted that.
Yes, those results aren’t “scientific”. But they give a strong indication of the actual situation out there.
And the conclusion also wasn’t that those are bad chips. Quite the contrary actually. All Ryzen Chips are great in pure performance and value. AMD was just to aggressive on the marketing trying to push numbers they can’t reproduce reliably now.
1 Like
The test runs calculations, the resulting points represent compute work done in a time frame.
So points * time = IPC x clockspeed * time
I think that is my problem with it. It is a pseudo-scientific presentation of non-scientific data.
God forbid if Intel ever did that, people would probably not even click on the video.
2 Likes
I’m not sure why i’m here defending DerBauer… It was a video i found interesting that seemed to reflect some of the Problems we’ve had here too. I saw nothing wrong with the methodology or outcome, but if you do, that’s fine by me.
It’s already been shown that higher clockspeeds can net lower scores in cinebench. Many Reviewers have faced that “Problem” when overclocking Ryzen 3000.
He didn’t claim it was a scientific test. He just made sure to evaluate the data he had as close to good as possible. The graphs are done properly and the Sigma values are done properly too. The data isn’t perfect, but that wasn’t implied.
It’s just a look at what’s out there, what users are seeing and reporting that back at the community, AIB’s and AMD to have a basis to work on. I’m sure all MB Manufacturers are doing their own, much more strict, tests. And DerBauer said that all the Manufacturers he talked to saw similar results to what he got here.
Timestamped
So maybe Cinebench is the problem?
As much as I like Roman, I think he is blowing this up a bit… It’s like 25MHz below advertised and it’s still early days of this architecture. Could easily be old windows versions, if he likes it or not, that does make a difference.
Would be interesting to see cinebench in a VM. Or on a Ryzen 3000 hackintosh. @wendell? You probably tried that already, right?
5 Likes
The older agesa which boosts higher has worse scores in cinebench and geekbench. Which no one noticed.
9 Likes