Discussion Ryzen 3000 series benchmark thread ** Open **

[cortexa99]

But I thought the consensus was that AMD has the ipc advantage now; apparently not.

This is an easy math. Single-core is pretty closed for both intel&amd now(Clock*IPC), but multi-core is not, AMD has obvious advantage. Vastly lower the MC weight in benches would make AMD looks worse than before.

I think this site change SC weight up from 40% to 60%(50% difference) but MC down from 10% to just 2%(4 times lesser) is a clear screw up for a benchmark, nowaday benches has no reason to up SC weight & down MC weight at this significant ratio all of a sudden. This action is utterly suspicious if you ask me......

 

[beginner99]

Yea, intel has bought them off for sure. That is why the first thing you see on their homepage is huge pictures of 2 ryzen processors.

i don't see anything (ad-blocker) hence they are obviously ads and pages nowadays don't control what ads they serve, that is handled by the ad provider and it's obvious you get tech ads on that site, most likely for CPUs.

 

[CHADBOGA]

The 3200G is more impressive by the day.

Indeed.

Even though I don't really need to upgrade, I'm going to justify buying another system on the basis that if my current motherboard dies, I can't easily or maybe at all, get a replacement for it, due to its age.

I'm thinking that towards the end of this year, I will buy the 3200G and then I'll be able to wait and see what is happening on the discrete GPU and Zen 3 front, in 2020.

And at worst, I then only have to throw away a $100 or less processor, when I upgrade to a 16 or 24 core Zen 3.

 

[VirtualLarry]

I'm toying with the idea of going with some APUs, for my other rigs.

 

[lopri]

The effective CPU speed index measures performance with the following weights: 40% single-core, 58% quad-core and 2% multi-core.
Source

That has to be a joke or a mistake. Why even bother with 2%? Might as well take it out completely.

 

[Asterox]

 

[DrMrLordX]

Yea, intel has bought them off for sure. That is why the first thing you see on their homepage is huge pictures of 2 ryzen processors.

You win the "delightfully consistent" award.

That has to be a joke or a mistake. Why even bother with 2%? Might as well take it out completely.

It means userbenchmark now only takes quads seriously. Would be interesting to see how 9900k scores fare against heavily-overclocked (4.8GHz+) i7-7700k scores, for example. 9900k might still win on cache alone.

@Zucker2k

Matisse is extremely powerful in ST performance in anything that fits nicely in its L3 cache, and it's still very competitive in scenarios where thread migration might be an issue (especially for users that tune their RAM to reduce thread migration performance penalties). Take a look at CBR20 ST scores. CBR20 was tuned for Intel processors vs. CBR15 (CBR20 uses Embree), and yet AMD still takes the cake in ST performance. A 5 GHz 9900k gets around, what, 510? A 3900x @ 4.5 GHz can beat that, no problem.

https://forums.anandtech.com/threads/【maxon】cinebench-r20-benchmark-thread.2562153/post-39874480

Userbenchmark shifting away from MT just makes Matisse less-dominant. Realistically-speaking, the 3900x (at least) should beat most/all Intel chips in userbenchmark assuming it can boost properly, which is a sticking point for some users right now. Might help Matisse users to set thread priority to make it fit on one CCX just in case the program is bouncing between CCXs though.

 

[maddie]

You win the "delightfully consistent" award.



It means userbenchmark now only takes quads seriously. Would be interesting to see how 9900k scores fare against heavily-overclocked (4.8GHz+) i7-7700k scores, for example. 9900k might still win on cache alone.

@Zucker2k

Matisse is extremely powerful in ST performance in anything that fits nicely in its L3 cache, and it's still very competitive in scenarios where thread migration might be an issue (especially for users that tune their RAM to reduce thread migration performance penalties). Take a look at CBR20 ST scores. CBR20 was tuned for Intel processors vs. CBR15 (CBR20 uses Embree), and yet AMD still takes the cake in ST performance. A 5 GHz 9900k gets around, what, 510? A 3900x @ 4.5 GHz can beat that, no problem.

https://forums.anandtech.com/threads/【maxon】cinebench-r20-benchmark-thread.2562153/post-39874480

Userbenchmark shifting away from MT just makes Matisse less-dominant. Realistically-speaking, the 3900x (at least) should beat most/all Intel chips in userbenchmark assuming it can boost properly, which is a sticking point for some users right now. Might help Matisse users to set thread priority to make it fit on one CCX just in case the program is bouncing between CCXs though.

Logic in 2019? Pfft.

 

[Ottonomous]

Userbenchmark, as inaccurate as it is, has reduced the weighting of MC performance from 10% of the overall score to 2%, giving intel processors a hilarious advantage against amd counterparts

the old scoring system:
https://web.archive.org/web/2019060.../Faq/What-is-the-effective-CPU-speed-index/55

8350K having an effective speed of +7% against the 2990WX and surpassing it:
https://cpu.userbenchmark.com/Compare/Intel-Core-i3-8350K-vs-AMD-Ryzen-TR-2990WX/3935vsm560423

 

[DAPUNISHER]

Which is faster?
A 3900X that hits 196fps at 4.5GHz or a 3900X that hits 196fps at 4.4GHz?
Makes no difference where the performance comes from. If X is the product of Y and Z, with X being constant, it doesn't matter what Y and Z are individually. They hit max performance out of the box, which ultimately means that Y and Z are irrelevant in the overall context of things.
We could be talking about Intel CPUs here for what it matters.
I'm tired of people hanging onto any little shred of information as a means to burn AMD when it is the bigger picture that counts. The bigger picture is that these CPUs are competitive, and in very many cases the performance leaders. How they got there is neither here nor there in the grand scheme of things. Sure, from a technical point of view it is interesting, but certainly not something with which to scold them.

Quoted for emphasis.

It is a bit early for a serious roasting of AMD anyways. They launched CPUs, GPUs, and chipset with PCIe 4.0 concurrently. Thus far, it has been a solid launch, with some bugs. And given there is ample historical precedence for this being the rule and not the exception, hold off with the pitchforks and torches until we see which are features and which are bugs, definitively.

 

[ondma]

You win the "delightfully consistent" award.



It means userbenchmark now only takes quads seriously. Would be interesting to see how 9900k scores fare against heavily-overclocked (4.8GHz+) i7-7700k scores, for example. 9900k might still win on cache alone.

@Zucker2k

Matisse is extremely powerful in ST performance in anything that fits nicely in its L3 cache, and it's still very competitive in scenarios where thread migration might be an issue (especially for users that tune their RAM to reduce thread migration performance penalties). Take a look at CBR20 ST scores. CBR20 was tuned for Intel processors vs. CBR15 (CBR20 uses Embree), and yet AMD still takes the cake in ST performance. A 5 GHz 9900k gets around, what, 510? A 3900x @ 4.5 GHz can beat that, no problem.

https://forums.anandtech.com/threads/【maxon】cinebench-r20-benchmark-thread.2562153/post-39874480

Userbenchmark shifting away from MT just makes Matisse less-dominant. Realistically-speaking, the 3900x (at least) should beat most/all Intel chips in userbenchmark assuming it can boost properly, which is a sticking point for some users right now. Might help Matisse users to set thread priority to make it fit on one CCX just in case the program is bouncing between CCXs though.

There are plenty of "consistent" posters in this forum. Blaming intel yet again of cheating without a shred of proof, is a prime example of this "consistency".

 

[zinfamous]

There are plenty of "consistent" posters in this forum. Blaming intel yet again of cheating without a shred of proof, is a prime example of this "consistency".

Let's apply Occam's Razor: So, what non-Intel favoring reason would they have to suddenly change the performance calculations in the way that they did?

I also don't see that a notification was given that they change happened--just that some users noticed it and brought it to light. Usually, there is a notice and an explanation, but this didn't happen in this case. An important data point, that.

 

[Space Tyrant]

I did a bit of overclocking on the 3600X today.

First, a baseline: this is my best GB4 run of the 3600X at stock with tweaked memory settings at DDR3200.
https://browser.geekbench.com/v4/cpu/13863582
(single-core:6248 multi-core:31839)

This is my best OC run at 4.25GHz with almost the same latency:
https://browser.geekbench.com/v4/cpu/14046522
(single-core:6253 multi-core:33353)

And, lastly, here's the best 2700X run (4.275GHz) I had from last year (slightly different GB version):
https://browser.geekbench.com/v4/cpu/8392486
(single-core:5584 multi-score:33351)

The 3600X at 4.25G matches its stock single-core score and also matches the 2700X 4.275GHz multi-core score.

My 3600X sample at stock often boosts to 4.4Ghz under light loads but doesn't seem to be hitting & holding that speed in any benchmark runs.

Edit: I turned off 4 cores and SMT, turned up the vcore a bit, and clocked the 3600X to 4.4GHz. So, here's the score for the "full boost", 2-thread 3600X:
https://browser.geekbench.com/v4/cpu/14050066

 

[Shivansps]

What th..? That A320 is outperforming a X570. PBO is actually working on it as well... not even i expected this, thats far better than i would ever trought.

As a side note i would add, that A320 and Gigabyte A320 are ones of the best VRM on the market, along with the Asus one, matching low end B350/B450 VRMs. There are crappier A320 than that one, Asrock ones with 3+2 for instance.

 

[Kenmitch]

What th..? That A320 is outperforming a X570. PBO is actually working on it as well... not even i expected this, thats far better than i would ever trought.

As a side note i would add, that A320 and Gigabyte A320 are ones of the best VRM on the market, along with the Asus one, matching low end B350/B450 VRMs. There are crappier A320 than that one, Asrock ones with 3+2 for instance.

Surprising results that I'd imagine are repeatable. I guess the only real hurdle would be the bios flashing for somebody switching over to the platform.

Microcenter bundle pricing for that MB and a 3600 is $204.99

 

[SPBHM]

3600 + a320 is a perfectly valid combination, and the lack of OC is kind of irrelevant since the CPU boosts so high by default; good to see that AMD didn't do any silly AGESA locks or something,
they should do the same with PCIE4, if the motherboard manufacture wants to make it work and give their warranty, let them do it.

 

[lopri]

That dude's analysis is way off. It is not because X570 boards have "longer traces" but because Gigabyte's latest BIOS is lowering performance of these CPUs. I have experienced first hand with my Aorus X570 Elite board. Due to the issues I had with 1003AB BIOS I had to go back to 1003, and the performance is up and more consistent, matching the numbers in the reviews. Power usage is also higher. I do not know why there is such difference, but it is not an expensive board vs a cheap board thing.

 

[B-Riz]

Which is faster?
A 3900X that hits 196fps at 4.5GHz or a 3900X that hits 196fps at 4.4GHz?
Makes no difference where the performance comes from. If X is the product of Y and Z, with X being constant, it doesn't matter what Y and Z are individually. They hit max performance out of the box, which ultimately means that Y and Z are irrelevant in the overall context of things.
We could be talking about Intel CPUs here for what it matters.
I'm tired of people hanging onto any little shred of information as a means to burn AMD when it is the bigger picture that counts. The bigger picture is that these CPUs are competitive, and in very many cases the performance leaders. How they got there is neither here nor there in the grand scheme of things. Sure, from a technical point of view it is interesting, but certainly not something with which to scold them.

I told my wife, a non computer nerd, that the $500 3900X I got was faster than Intel's $1100 9920X; she was shocked the AMD chip was $500.

 

[B-Riz]

Quoted for emphasis.

It is a bit early for a serious roasting of AMD anyways. They launched CPUs, GPUs, and chipset with PCIe 4.0 concurrently. Thus far, it has been a solid launch, with some bugs. And given there is ample historical precedence for this being the rule and not the exception, hold off with the pitchforks and torches until we see which are features and which are bugs, definitively.

Do we not remember the Sandy Bridge P67 fiasco? LOL

I joined here around that launch, and was so excited to replace the janky i7 920 C0, but then, it all started to go sideways with motherboards, what a poop show.

A large issue, if you follow AMD Reddit posts, is, having to support all the older boards, if it was just X570, this would be a lot better.


This has been going great I think, all things considered. I followed the setup directions from below, and my X570 / 3900X has been ok, but the 3900X had better perf on the B350, LOL.

https://www.reddit.com/r/Amd/comments/cbls9g

 

[DrMrLordX]

There are plenty of "consistent" posters in this forum.

Then you're in good company!

Let's apply Occam's Razor: So, what non-Intel favoring reason would they have to suddenly change the performance calculations in the way that they did?
.

There isn't one. Oddly enough, the change lowers scores relative to quads for a lot of Intel CPU as well! Just not as much as it lowers scores for AMD chips.

 

[Abwx]

3600X/3800X review :

https://www.computerbase.de/2019-07/amd-ryzen-3600x-3800x-test/

 

[IntelUser2000]

Do we not remember the Sandy Bridge P67 fiasco? LOL

Not just P67, but the entire lineup.

The replacement policy around with the 6-series chipset was to be commended. I sent in my motherboard to Newegg, because it worked with Intel to make the replacement process easy. I sent in the request, and they sent me the shipping code and arrived in less than a week, with no cost to me.

The market rewarded them for it because their revenue went to grow by 20%. You think its obvious, but rarely you see it happen.

Contrast with the Samsung's policy on Note 7's when the phone's were found to have a risk of catching fire due to bad design. They were offering what, $200 on the next phone?



Although, most problems are not as clear cut. If you offered recalls on every single issue that existed, there wouldn't be a single business alive. This is especially with code, software and firmware.

 

[Thunder 57]

I told my wife, a non computer nerd, that the $500 3900X I got was faster than Intel's $1100 9920X; she was shocked the AMD chip was $500.

Something that the AdoredTV idiots never understood. Pricing a product too low can be just as bad as pricing it too high. "It's so cheap, something must be wrong with it".

 

[Atari2600]

Something that the AdoredTV idiots never understood. Pricing a product too low can be just as bad as pricing it too high. "It's so cheap, something must be wrong with it".

Conversely, it is not unreasonable to expect anyone spending $500 on a CPU to have even a rudimentary grasp on the state of the market.

 

[Dave3000]

Which CPU has the highest 4-core turbo boost (turbo's highest when 4 cores/8 threads are under full load)? 3600x, 3800x or the 3900x? Also since the 3800x has 4 cores in one CCX and the 3900x has 3 cores in one CCX, if a game uses up to 4 cores, will the 3900x be CCX hopping in that particular game causing it to perform worse than the 3800x in a situation like that or does the larger L3 cache of the 3900x compensate for this?