AMD Zen 2 (Ryzen 3000) - * NO COMPETITOR HINTING *

Dirk Diggler · 2 Sep 2019 at 07:51

Robert896r1 said:
You’re still not making any sense but keep stating that testing might be flawed. Flawed in what way? Manufacturer provided guideline were followed which means that AMD themselves are expecting single core behavior to align with Max boost represented.

How is your sample size of ONE a better reference point?

Start putting facts together instead of drive by skepticism. It’s lazy.

It's obvious what the guy is saying, are you going to keep trolling until he apologizes or something?

Buffalo2102 · 2 Sep 2019 at 07:54

I'm not saying my sample of one is a better reference point. Just saying, based on my sample of one, the testing may be flawed. What's so difficult to understand?

I only have one fact - the behaviour of my CPU. I'm not trying to debunk anything here and I have no agenda. I'm just pointing out something that may be significant, but equally may well be an aberration. Up to you if you choose to believe me or even take notice of me.

Panos · 2 Sep 2019 at 12:31

MichaelM said:
Absolutely. The fact that he had to go through each entry to condition the data showed how uncontrolled it is. Junk in, junk out. I applaud his attempt and knowledge of statistical analysis but the execution was lacking.
Solve that and then you get to your point, or simply frame the results in the context of "real world variation of setups and settings" which would be helpful for AMD and customers.

No news on the 3950x? The rumored 7th release date is looking less likely then?

You forgot some dudes in Reddit boasting they did 10+ entries to show AMD is bad......

And nobody should forget that Roman isn't some independent researcher. He has vested interests in Intel sales due to deliding services & kits he is making money from. (and Thermal Grizzly).

humbug · 2 Sep 2019 at 13:23

Nghtmare said:
Does anyone know if Windows updates or Chipset Driver updates can affect RAM performance, and latency. I used to get 64.3ns on the same setting I am on now, but now for some reason that has gone up to 66ns without me touching anything.

This one from a few weeks ago, ignore the CL16-17-17 timings, that was not properly stable, 75.0ns

This one just now, with some hours of Memtest testing pass, 75.5ns

Windows and Chipset drivers are upto date.

Twinz · 2 Sep 2019 at 13:26

I think the test parameters AMD gave him are bad.

My 3800x boosts to 4.5 during games (4.525 in the max column last time I played) but rarely have I seen it do so under a full single core load like CB.

After watching core speeds during a SC CB run, I think I know why AMD recommends their "ryzen" power plans with crazy fast response times...It switches between cores during heavy single core load.

I think 7nm heat density is difficult to deal with and a load like that gets moved around the die like a hot potato. The benchmark score is still good, but the fastest core in the CPU needs help from other (slower) cores to get through the entire benchmark.

I'm surprised this behavior hasn't been talked about more. It seems like a clever way to deal with an engineering challenge. (Fast switching + hot-potato SC hand-offs) ...but I suspect it doesn't make for good marketing hype. (Becuase the "star core" gets benched a lot)

Nghtmare · 2 Sep 2019 at 13:27

@humbug My current RAM settings used to give me 64.3ns consistently and managed to get it to 62.9ns with some adjustments. Now the same 64.3ns settings give me 65.6ns, I have no idea why :/

It seems like yours went up a little too, L3 cache increase slightly on mine as well. Used to get 9.8ns and now I get 10ns.

humbug · 2 Sep 2019 at 13:33

Nghtmare said:
@humbug My current RAM settings used to give me 64.3ns consistently and managed to get it to 62.9ns with some adjustments. Now the same 64.3ns settings give me 65.6ns, I have no idea why :/

It seems like yours went up a little too, L3 cache increase slightly on mine as well. Used to get 9.8ns and now I get 10ns.

There have been 2 or 3 updates to AIDA64 in the last couple of weeks, did you update it? i have.

Buffalo2102 said:
But that whole test is based on the assumption that Cinebench will provoke the max boost clock. My 3800X doesn't boost to the advertised 4.5GHz during Cinebench but often does during general use.

Kinda makes all of that useless doesn't it?

Agreed His testing methodology completely ignores the fact that Ryzen 3000 boost behaviour depends on what its doing, its not necessarily that it will never boost to its advertised boost clocks, all we can say is most of them don't do it in Cenebench.

Mine for example in Cinebench ST runs at 4.125Ghz, yet i have just run an AIDA64 benchmark and the CPU ran at 4.2Ghz.

humbug · 2 Sep 2019 at 13:43

One thing i found that many may not know about, if you want your FSB clock to stay at 100Mhz, not 99.67Mhz or whatever..... turn off Spread Spectrum.

humbug · 2 Sep 2019 at 13:54

@Nightmare I just updated https://support.microsoft.com/en-gb/help/4511555/august-30-2019-kb4511555

On rebooting i ran AIDA64 again, first run 77ns, second run 76ns, third run 75.5ns

Nghtmare · 2 Sep 2019 at 13:55

humbug said:
@Nightmare I just updated https://support.microsoft.com/en-gb/help/4511555/august-30-2019-kb4511555

On rebooting i ran AIDA64 again, first run 77ns, second run 76ns, third run 75.5ns

I'll give that a go when I get home, and I am also going to check Aida64 since I don't think its updated.

humbug · 2 Sep 2019 at 13:57

Just saw this, i'll do this later, i have to go out shortly.

humbug · 2 Sep 2019 at 15:47

*Right thread this time, too many Ryzen threads*

Installed the update, no change...

Just purely out of interest i ran the FarCry 5 benchmark to compare it with Online reviews of my 3600, just to see if my memory overclocking actually makes a difference.

At 720P to make sure the GPU is not the bottleneck, other that that everything is the same as Toms Hardware, and well yes, the difference is huge.... even tho my RAM while better than stock is still not good.

Toms: 82/109
Mine: 94/120 (+15%) on the minimums.

Toms Hardware

https://www.tomshardware.com/reviews/amd-ryzen-5-3600-review,6287-6.html

chroniclard · 2 Sep 2019 at 15:54

You ran at 720p?

humbug · 2 Sep 2019 at 15:55

chroniclard said:
You ran at 720p?

Yes. At 1080P the GPU is the bottleneck, even at 720P there is still some GPU bottlenecking.

Smak · 2 Sep 2019 at 16:38

Nghtmare said:
@humbug My current RAM settings used to give me 64.3ns consistently and managed to get it to 62.9ns with some adjustments. Now the same 64.3ns settings give me 65.6ns, I have no idea why :/

It seems like yours went up a little too, L3 cache increase slightly on mine as well. Used to get 9.8ns and now I get 10ns.

I have same issue my 63.5ns settings are now 64.1ns but no idea what change is causing this.

Nghtmare · 2 Sep 2019 at 17:27

Smak said:
I have same issue my 63.5ns settings are now 64.1ns but no idea what change is causing this.

Im glad its just not me, and that there is something going on since I was going paranoid. I managed to get 62.9ns before but didnt want to put more voltage on the RAM so I left it at 3733Mhz.

This is the best I got

Nghtmare · 2 Sep 2019 at 17:29

@humbug
That is a pretty big difference.
14% on lows and 10% on average? Correct me if I messed up the math lol

humbug · 2 Sep 2019 at 17:44

Nghtmare said:
@humbug
That is a pretty big difference.
14% on lows and 10% on average? Correct me if I messed up the math lol

Between 14 and 15% on the lows, 10% on the averages yes

Its similar to what HUB found, tweaking the timing manually with DRam Calc.

3800MT/s, 1900Mhz IF: 57 FPS minimums
3800MT/s, 1900Mhz IF: Manual (DRam Calc) 68 FPS minimums +19%

Murphy · 2 Sep 2019 at 17:47

humbug said:
Agreed His testing methodology completely ignores the fact that Ryzen 3000 boost behaviour depends on what its doing, its not necessarily that it will never boost to its advertised boost clocks, all we can say is most of them don't do it in Cenebench.

If they don't it seems strange for AMD to recommend that he should use that.

I don't understand why people are defending AMD for this, don't get me wrong Zen 2 are great CPUs and 100Mhz makes next to no difference in the real-world performance of them, for me this is simply a matter of AMD misleading people, whether they set out to do that is irrelevant, it would've been simple to lower the max advertised boosts by 100Mhz and probably no one would've batted an eye, it probably would've even done them some favors as it wouldn't have made PBO redundant.

I'm not directing this at you personally BTW humbug, it's just a general moan about how trying to excuse AMD for over promising and under delivering does them (AMD) no favors.

jigger · 2 Sep 2019 at 17:56

Murphy said:
If they don't it seems strange for AMD to recommend that he should use that.

I don't understand why people are defending AMD for this, don't get me wrong Zen 2 are great CPUs and 100Mhz makes next to no difference in the real-world performance of them, for me this is simply a matter of AMD misleading people, whether they set out to do that is irrelevant, it would've been simple to lower the max advertised boosts by 100Mhz and probably no one would've batted an eye, it probably would've even done them some favors as it wouldn't have made PBO redundant.

I'm not directing this at you personally BTW humbug, it's just a general moan about how trying to excuse AMD for over promising and under delivering does them (AMD) no favors.

How do you feel about Intel? People lost 30% performance over night and Intel knew it would happen.

Competitor rules

AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***

AMD Zen 2 (Ryzen 3000) - * NO COMPETITOR HINTING *