Ryzen benchmarking and overclocking results

[thesmokingman]

^^Yep. Stock boost does not run all cores at max frequency. It never has. It's surprising how many ppl don't realize the specifics.

 

[Kanan]

Intel now has a max boost for 1 core that is very high, generally speaking, and a all core boost that isn't so high, maybe +200--300mhz (talking about the 6800/6900k models with turbo 3.0). AMD had a 3 core turbo (3 of 6) when they first introduced turbo in Phenom II X6. Later it was more like Intels 2.0 turbo, only boostIng 1-2 cores, FX 8350 from 4 ghz to 4.2 ghz for example. My own 3960X simply boosted to 3.9 ghz all the time without me touching it, so it's a odd one, maybe because it's a X cpu I don't know, but I think that is it. So expect 6950X to run relatively high as well also explains it's high score among all games despite the low base clock and despite most games not really utilising more than 4 cores.

 

[r9]

Can someone do side by side gaming benchmark with SMT and one Ccx disabled. So we can see what ryzen quad core w/o SMT can do. With all the windows scheduler speculation going around. Benching with four cores and no SMT should give better results in gaming.

 

[EarthDog]

Intel now has a max boost for 1 core that is very high, generally speaking, and a all core boost that isn't so high, maybe +200--300mhz (talking about the 6800/6900k models with turbo 3.0). AMD had a 3 core turbo (3 of 6) when they first introduced turbo in Phenom II X6. Later it was more like Intels 2.0 turbo, only boostIng 1-2 cores, FX 8350 from 4 ghz to 4.2 ghz for example. My own 3960X simply boosted to 3.9 ghz all the time without me touching it, so it's a odd one, maybe because it's a X cpu I don't know, but I think that is it. So expect 6950X to run relatively high as well also explains it's high score among all games despite the low base clock and despite most games not really utilising more than 4 cores.

It was the motherboard. Many do this on the Intel side. The 3960X boosted all 6 cores 3 bins (300MHz), then up to 4 cores 5 bins, then up to 2 cores, 6 bins. Should be in the Intel Turbo Boost tables.

The 6950x boosts to 3.5 GHz...

"Turbo Boost 2.0 is what Intel calls its maximum Turbo or ‘peak’ frequency. So in the case of the i7-6950X, the base frequency is 3.0 GHz and the Turbo Boost 2.0 frequency is 3.5 GHz. The CPU will use that frequency when light workloads are in play and decrease the frequency of the cores as the load increases in order to keep the power consumption more consistent."
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

When I 'load defaults' on my board. I sit at 3.5Ghz all cores.

 

[OneCool]

Do we have any results on crossfire/SLI scaling? Kinda curious about that.

 

[cdawall]

It was the motherboard. Many do this on the Intel side. The 3960X boosted all 6 cores 3 bins (300MHz), then up to 4 cores 5 bins, then up to 2 cores, 6 bins. Should be in the Intel Turbo Boost tables.

The 6950x boosts to 3.5 GHz...

"Turbo Boost 2.0 is what Intel calls its maximum Turbo or ‘peak’ frequency. So in the case of the i7-6950X, the base frequency is 3.0 GHz and the Turbo Boost 2.0 frequency is 3.5 GHz. The CPU will use that frequency when light workloads are in play and decrease the frequency of the cores as the load increases in order to keep the power consumption more consistent."
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2

When I 'load defaults' on my board. I sit at 3.5Ghz all cores.

If you have TBMT3.0 installed the 6950X, 6900K and 6850K all turbo to 4.0

 

[EarthDog]

If you have TBMT3.0 installed the 6950X, 6900K and 6850K all turbo to 4.0

absolutely right. I deleted that part on an edit before posting.

It's also one core only.

 

[phanbuey]

Memory bandwith makes such a huge difference on this platform.... wondering if it's just because the cache is so much slower than on the intels that it is acting like a bottleneck... and the less you can keep the bottleneck waiting the faster the system responds.

Going to be putting some 3200Mhz c15 gskills in this weekend.

I am really curious why AMD didn't send 3200Mhz memory with their test kits.

 

[EarthDog]

Because it seems there is only a couple of boards out currently that support such speeds... at release time, it seemed only the CH6 supported those speeds.

 

[TheoneandonlyMrK]

Intel now has a max boost for 1 core that is very high, generally speaking, and a all core boost that isn't so high, maybe +200--300mhz (talking about the 6800/6900k models with turbo 3.0). AMD had a 3 core turbo (3 of 6) when they first introduced turbo in Phenom II X6. Later it was more like Intels 2.0 turbo, only boostIng 1-2 cores, FX 8350 from 4 ghz to 4.2 ghz for example. My own 3960X simply boosted to 3.9 ghz all the time without me touching it, so it's a odd one, maybe because it's a X cpu I don't know, but I think that is it. So expect 6950X to run relatively high as well also explains it's high score among all games despite the low base clock and despite most games not really utilising more than 4 cores.

Yeah but you can run upto 6 cores boosted with tweaks I run flat clocks but do use overdrive sometimes to boost my CPU clocks for benching, not for max clocks , just when I want to check something isn't wrong for example.
I think it and AMDs aborted fusion melarky where going to , well technically you can ,set a software overclock per application ,very much a precursor for what we see today because imho they and Intel have been struggling to maintain the right amount of performance across various loads without generating loads of heat ,easily seen in both camps varying power draw and heat output on varying work loads.

 

[Crap Daddy]

Weird result or possible future gaming performance when things work well?

http://www.eteknix.com/amd-ryzen-7-1800x-am4-8-core-processor-review/

This is the test system:

Ryzen of the tomb raider via eteknix parallel universe. The 980Ti on very high preset would not be even close to those framerates. Something is either done wrong or purposely misleading.

 

[phanbuey]

Because it seems there is only a couple of boards out currently that support such speeds... at release time, it seemed only the CH6 supported those speeds.

If they just stuck to that board and the fast ram, they would have had substantially better gaming numbers...

Between the faster ram, and the incoming SMT patch, the chip will game great.

 

[EarthDog]

Has faster ram (say 2400-3000) shown to significantly improve FPS on Ryzen? It doesn't in most titles on Intel...

 

[phanbuey]

Has faster ram (say 2400-3000) shown to significantly improve FPS on Ryzen? It doesn't in most titles on Intel...

Huge gains...
http://www.eteknix.com/memory-speed-large-impact-ryzen-performance/
http://www.legitreviews.com/ddr4-me...latform-best-memory-kit-amd-ryzen-cpus_192259

I can see it on the 1800X from 2133 to 2400 ... but will post some screenies when the 3200mhz kit comes in.

I think its because the cache is too slow to keep the proc fed otherwise. Intel's current caching is so efficient that the ram speed is trivial.

This reminds me of back in the day when memory speed mattered lol.

 

[EarthDog]

So one game at 1080p MEDIUM settings with a 1080 in the second link and TW3 in the first link which was one of the titles increasing Intel memory responds to as well. Gotcha.

EDIT: And that only seems to hold for 1080p. Anything higher and its nominal.

The bad news is that by the time you read 2560 x 1440 (1440P), the system is more GPU botttlenecked, so memory clock speed didn’t impact performance at all. The same thing proved true for 4K gaming performance, so if you aren’t GPU limited
Read more at http://www.legitreviews.com/ddr4-me...t-amd-ryzen-cpus_192259/4#yJXr8PtI3tKKVB69.99

 

[phanbuey]

So one game at 1080p MEDIUM settings with a 1080 in the second link and TW3 which was one of the titles increasing Intel memory responds to as well. Gotcha.

Better than nothing. Also you want the lowest graphical settings since its for CPU speed.

Trying to find more but doesn't seem to be a whole lot out there... might have to do some very unscientific homebrew benchmarking this weekend.

So one game at 1080p MEDIUM settings with a 1080 in the second link and TW3 in the first link which was one of the titles increasing Intel memory responds to as well. Gotcha.

EDIT: And that only seems to hold for 1080p. Anything higher and its nominal.

That's just because the bottleneck shifts, so there is no longer any point in trying to measure something that isn't a bottleneck anymore. Bottom line is ryzen is getting dinged for 1080P gaming at high FPS - and memory scaling seems to bump that up substantially from the very limited sample set that's out there.

 

[EarthDog]

That doesn't sound right... to test showing a situation, and exaggerating an issue, I don't use with a 1080. Sounds logical. If I have a 1080, I'm running everything full out at 1080p. THis puts more load on the GPU and would theoretically show less increase from ram speeds. Also, if you have a midrange GPU like a 1060 or 480... does it still show the same increases?

 

[phanbuey]

That doesn't sound right... to test showing a situation, and exaggerating an issue, I don't use with a 1080. Sounds logical. If I have a 1080, I'm running everything full out at 1080p. THis puts more load on the GPU and would theoretically show less increase from ram speeds. Also, if you have a midrange GPU like a 1060 or 480... does it still show the same increases?

Well no... but the point of the bench is not to show a balanced gaming rig, but to highlight the CPU bottlenecking as it is analogous to other games/similar workloads. It's not going to bottleneck a 1060 @ 1080P so the impact is not going to be as pronounced.

 

[EarthDog]

You proved my point. Its a 'highlight' BECAUSE its an unrealistic situation (1080, medium settings 1080p)... Put it back in the realm of normal (1080 ultra settings with AA...etc), it can be less pronounced and can even go away with lesser cards. So why report that way? Why not show how people actually play or use that card?

 

[Vya Domus]

Memory bandwith makes such a huge difference on this platform.... wondering if it's just because the cache is so much slower than on the intels that it is acting like a bottleneck... and the less you can keep the bottleneck waiting the faster the system responds.

Going to be putting some 3200Mhz c15 gskills in this weekend.

I am really curious why AMD didn't send 3200Mhz memory with their test kits.

Apparently Zen has some trouble with the way Windows handles thread scheduling and has some latency issues accessing the cache as a result.

 

[phanbuey]

You proved my point. Its a 'highlight' BECAUSE its an unrealistic situation (1080, medium settings 1080p)... Put it back in the realm of normal (1080 ultra settings with AA...etc), it can be less pronounced and can even go away with lesser cards. So why report that way? Why not show how people actually play or use that card?

Because you're trying to compare processor performance...

I am not understanding your point.

 

[EarthDog]

I'm not comparing processor performance.

What I'm trying to say is, IMO, the testing there of a 1080 using medium settings isn't a realistic test environment and exaggerates any differences that are found with more appropriate settings or resolutions. I'd be more interested in the results at ultra and 1080p with a 1080 since, that is presumably where people will set things... or that same 1080 at 2560x1440 and ultra as it can easily handle most titles there too.

As it stands, the type of testing done magnifies any differences found in more realistic settings.

 

[TheoneandonlyMrK]

Apparently Zen has some trouble with the way Windows handles thread scheduling and has some latency issues accessing the cache as a result.

I mentioned it ,was largely ignored.
Patch please.

 

[Kanan]

If you have TBMT3.0 installed the 6950X, 6900K and 6850K all turbo to 4.0

That's exactly what I meant.

Btw, here's a video from a software dev talking about Ryzen architecture in-depth, and what problems it has and how it's solvable:

Worth the time.

 

[phanbuey]

not sure if this is true... but:

https://www.reddit.com/r/Amd/comments/5xjmau/ryzens_data_fabric_is_locked_12_ram_speed_faster/

i think someone posted this already but worth doing again.