AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[LTC8K6]

I think 4c/8t should be APU only.

 

[moonbogg]

I'd consider a 6/12 core Zen, but I'd feel like I was missing out on something, 2 more cores to be specific. I have seen enough gaming benchmarks where 8 cores pulls enough ahead of everything else to make me unhappy with less than 8. It all comes down to OC performance for me now.

 

[The Stilt]

Yup. Coupled with cheaper-then-2011v3 AM4 boards, that will be pretty competitive from what we have seen so far.

I wouldn't exactly "count the chickens" yet regarding the pricing of AM4 motherboards, at least the proper ones.
Take a look at Z270 boards for example. They cost generally the same as X99 boards of the same segment.

 

[LTC8K6]

Well, there are lots of possible combos.
4/4
4/8
6/6
6/12
8/8
8/16

But I don't think AMD should go that route.

 

[crashtech]

I'd consider a 6/12 core Zen, but I'd feel like I was missing out on something, 2 more cores to be specific. I have seen enough gaming benchmarks where 8 cores pulls enough ahead of everything else to make me unhappy with less than 8. It all comes down to OC performance for me now.

Which games pull ahead with 8 cores?

 

[moonbogg]

Which games pull ahead with 8 cores?

I can't remember but they've been shown many times. Many many times and everyone knows. This is well known.

 

[jpiniero]

I think 4c/8t should be APU only.

It likely will be eventually but until the APUs are released it does make sense. Guess it will depend on yield and how many dies are salvageable.

 

[zinfamous]

I can't remember but they've been shown many times. Many many times and everyone knows. This is well known.

They are the best games, believe me! I will tell you all about them when the time is right.

AFAIK, Watchdogs 2? and pretty much any RTS that is well-optimized for DX12...which is basically one example right now and it isn't all that well optimized.

 

[Insert_Nickname]

I wouldn't exactly "count the chickens" yet regarding the pricing of AM4 motherboards, at least the proper ones.
Take a look at Z270 boards for example. They cost generally the same as X99 boards of the same segment.

Don't know US pricing but here you can get a decent Z270 board for 2/3 of an entry level LGA-2011v3 board. A Z170 board can be had for half.

 

[moonbogg]

I solemnly swear to execute all 16 threads as faithfully as possible, to all games developed foreign and domestic.

 

[crashtech]

I can't remember but they've been shown many times. Many many times and everyone knows. This is well known.

Everyone minus one, apparently. There's this:

BUT, there is no hexacore on there, so it's hard to say for sure how much advantage the two extra cores confer. I was genuinely curious and wanting to agree with you. Maybe someone else would be more helpful, I've been trying to find examples of octacores beating hexacores in games.

EDIT: Oh, okay, that was a Trump joke. Sorry, I don't keep up on that crap and don't really expect to see it on this side of the forum.

 

[moonbogg]

There's more games than that. Besides, an 8 core with similar clocks as a quad would at least equal the quad (or hex) but pull ahead in those programs that support more threads. Its time to leave quads in the dust. They are ancient. Zen 8/16 or nothing for me.

 

[frozentundra123456]

I can't remember but they've been shown many times. Many many times and everyone knows. This is well known.

Yea, the universal "everybody". Personally, i am not so sure there is reliable data on that. Now the 5960x shows some very impressive results, especially on the game.gpu tests, which I consider much less reliable than I used to, but even then, one must remember it has a huge cache as well. The best way to test this would be with the 5960x, and disable 2 or 4 cores. Dont really know if there is much data on this. My personal feeling is that there is very little gaming benefit with more than 6 hyperthreaded cores.

 

[crashtech]

Yea, the universal "everybody". Personally, i am not so sure there is reliable data on that. Now the 5960x shows some very impressive results, especially on the game.gpu tests, which I consider much less reliable than I used to, but even then, one must remember it has a huge cache as well. The best way to test this would be with the 5960x, and disable 2 or 4 cores. Dont really know if there is much data on this. My personal feeling is that there is very little gaming benefit with more than 6 hyperthreaded cores.

Hyperthreading seems to suffer higher likelihood of performance regressions with higher core counts, which makes one wonder if a highly OCed octacore with HT off might be the killer setup for gaming. So much of this is uncharted territory at this time.

 

[dfk7677]

Battlefield 1(so the Frostbite engine in general) is very thread number sensitive and can use up to 16 threads. Also IPC from haswell 4c/4t to Skylake 4c/8t goes from 1.14 to 2.08.

I think games will follow the multithreading path massively in the next few years.

 

[CentroX]

Battlefield 1(so the Frostbite engine in general) is very thread number sensitive and can use up to 16 threads. Also IPC from haswell 4c/4t to Skylake 4c/8t goes from 1.14 to 2.08.

I think games will follow the multithreading path massively in the next few years.

because DICE are among very few developers that always tried to drive tech forward. Battlefield 3 was one of the first games to be built for DX11 only.

 

[DrMrLordX]

According to Chiphell user Wjm47196:

- Launch in March

March 1st-2nd then? Otherwise that doesn't square with the leaked GDC presentation by AMD, which could be as late as March 3rd . . .

 

[DeletedMember377562]

Everyone minus one, apparently. There's this:

BUT, there is no hexacore on there, so it's hard to say for sure how much advantage the two extra cores confer. I was genuinely curious and wanting to agree with you. Maybe someone else would be more helpful, I've been trying to find examples of octacores beating hexacores in games.

EDIT: Oh, okay, that was a Trump joke. Sorry, I don't keep up on that crap and don't really expect to see it on this side of the forum.

1. Let's look at one test, a Russian test site that's shady, has many numbers that deviate from other sites, and that doesn't even bother to test chips at same speeds, give better info about what kind of RAM is used and more. You want a proper 1080p tests of chips? Look at Eurogamer's (by DigitalFoundry's) test of Haswell vs. Skylake. Watch Dogs 2 is not included, but a interesting fact is that those very games testet by Eurogamer are also having these big gaps with the same chips by GameGPU. But as DF proved, the 6700K is 5% ahead of the 5820K and 5960X, which is also how much it is ahead in terms of architecture improvements from HW to SL.

2. Let's ignore the fact that they're testing a GTX 1080 at 1080p. Because that makes a fuck ton of sense, right?

 

[StrangerGuy]

For me I would rather see 1st percentile FPS than true minimums which are very likely to be irrelevant outliers.

 

[crashtech]

1. Let's look at one test, a Russian test site that's shady, has many numbers that deviate from other sites, and that doesn't even bother to test chips at same speeds, give better info about what kind of RAM is used and more. You want a proper 1080p tests of chips? Look at Eurogamer's (by DigitalFoundry's) test of Haswell vs. Skylake. Watch Dogs 2 is not included, but a interesting fact is that those very games testet by Eurogamer are also having these big gaps with the same chips by GameGPU. But as DF proved, the 6700K is 5% ahead of the 5820K and 5960X, which is also how much it is ahead in terms of architecture improvements from HW to SL.

2. Let's ignore the fact that they're testing a GTX 1080 at 1080p. Because that makes a fuck ton of sense, right?

Thanks for the link! I'm having trouble discerning what the underlying hostility is about, though. What conclusions do you draw from the Eurogamer test?

 

[DeletedMember377562]

Battlefield 1(so the Frostbite engine in general) is very thread number sensitive and can use up to 16 threads. Also IPC from haswell 4c/4t to Skylake 4c/8t goes from 1.14 to 2.08.

I think games will follow the multithreading path massively in the next few years.

Heard that story thousands of times. What you and many others don't seem to understand is just because more threads are used, doesn't mean the game is reliant on them. Even the developers themselves admit to only 4 threads being necessary, whereas the rest are "worker cores", that they can split the tasks up on. And you even see in the performance gains. You mention Battlefield, for example. I don't know of any proper CPU tests of that game. What I do know, however, is that BF4 (running the same Frostbite engine) will give you the same exact performance on 4 cores as 6 and 8 cores, even at 1080p, with a GTX 1080. As proven here: http://www.tomshardware.com/reviews/multi-core-cpu-scaling-directx-11,4768-2.html

 

[DeletedMember377562]

Thanks for the link! I'm having trouble discerning what the underlying hostility is about, though. What conclusions doyou draw from the Eurogamer test?

My conclusion is the same. That when you use a GTX 1080, it's most likely that you play at 1440p resolution (even one of the most advanced shooters, the type of games you need high frame rates at, BF1, gets about 110-120 FPS at this resolution with a GTX 1080). And if you do that, even the most CPU intensive games out there will give you the same minimum and general frame rates on a 6600K with 4c/4t (they tested with 6700K with HT off, which is basically the same), as any CPUs with higher core counts: http://www.tomshardware.com/reviews/multi-core-cpu-scaling-directx-11,4768-2.html

And thats' not taking the factor of frequency into account. A modern KL can do 5 GHz. A modern Broadwell-E can to 4.3 GHz tops, in general (of course 4.4 can be achieved, but it's as unrealistic as 5.1 for KL). That's a staggering 700 MHz, or 16%, difference, If we did that, it would go even more so in favor of the fewer cores, as these things matter more for games than core counts. Not to mention the fact that Intel's 4 core chips are always 1-2 generations ahead, always giving them the edge in terms of IPC and other aspects.

Just read what the developers of the tested games admit to in the TH article.

For me I would rather see 1st percentile FPS than true minimums which are very likely to be irrelevant outliers.

Excactly. One of many areas where GameGPU is a horrible place to discern any facts. All their CPU benches deviate from other sites. The fact that people keep using them as a reference is astonishing. I guess they are the only place for people who purchased Intel-E chips to "prove that my chip is best for gaming", as the overwhelming data points towards the opposite.

 

[deasd]

According to Chiphell user Wjm47196:

- Launch in March
- 4C/8T four months after 8C/16T
- (Fastest?) 8C/16T might not be cheap - he expects 3999-4999 RMB ($580-720)

https://www.chiphell.com/thread-1697525-1-1.html

I used google translate and get the guy said 4C8T will launch in April. Looks like you guys mix up '4 months' and '4th month'.

Bitsandchips was still correct that said 'No Four Cores Zen CPUs at least initially'

The guy also said game performance is 'expected' which means it should be good.

 

[DeletedMember377562]

I used google translate and get the guy said 4C8T will launch in April. Looks like you guys mix up '4 months' and '4th month'.

Bitsandchips was still correct that said 'No Four Cores Zen CPUs at least initially'

The guy also said game performance is 'expected' which means it should be good.

It's literally the same architecture. Not like a 4 core Ryzen wil magically perform better than 8 core Ryzen at same speeds. If they make their 6 core chips at managable prices, and they can easily clock to 4 GHz, I would very quickly buy them. Do not have the need for higher frequency (and too little gain in performance vs. temps/power usage), nor 8c/16t. It's of course all down to the price. But if they can price it a bit lower than current KL 4c/8t for their 6c/12, they have me hooked.

 

[crashtech]

For me I would rather see 1st percentile FPS than true minimums which are very likely to be irrelevant outliers.

I agree with this. Low percentile frame rates are where higher core counts seem to shine though, possibly due to poorly controlled variables like background processes.