Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[Xpage]

I don't see many people's problem. We never expected 7700k performance. Going from Excavator to this is a HUGE jump, more than I expected from AMD. Their MT pe4rformance is great and was expected since they are aiming for workstation/server. This was never going to be a top of the line choice for gamers. All this does is make this a potential solution if you like AMD and want AMD and not Intel then this will bring you in from the cold to something competitive unlike AMD's Piledriver FX offering which is very outdated.

So great CPU, typical launch issues. I learned my lesson from the Intel SB motherboard recall about buying day 1. Hence why I get something after 2-3 months. Though Ill be keeping my 2500k for awhile longer since I only do 1440p and don't game much. I am waiting for Vega to see what that does as I need a GPU more than a CPU.

 

[french toast]

So, with more reviews pouring, we can have a retrospect to events and "leaks" leading to the launch and discuss finer points of Zen:

1) AMD has been mum about CCX interconnects and L3 because they are hurting and will continue to hurt performance. 16MB of L3 per CPU is nice and dandy, but each core can access 8MB of those @ limited speed and that speed is also being hurt by coherency traffic - not optimal. I'd take Intel's multi ring HCC CPU any day over this. They better have some extra souce (in form of extra links?) for rumoured 32 core server chip, coherency traffic between 8 CCX chunks will kill it.
2) SMT has 10-15% perf penalty for being enabled, Intel has sorted out these things ~Nehalem gen. Very disappointing for desktop loads and okayish for servers.
3) OC rumours are pretty much shot down too, 1 core OC or not, this is another Polaris chip from AMD, loaded with features to extract the most out of process and bin, leaving scraps on table for overclockers. ~4Ghz is nice for 1700, but not so nice for top end CPU.
4) Larger L2 cache is nice, a lot of perf is coming out of it, goes a long way of showing how Intel has been milking customers with 256KB/s of L2. It turned out a lot of workloads can benefit very much from extra L2 and 4x as slow L3 is a gap too wide. SKL-X could have some major surprises for performance, some workloads will benefit nicely.

What stays the same between AMD releases is the tune of "wait for OS/game/software updates". We were supposed to get extra performance from Phenom when TLB bug was sorted out by recompiles, Bulldozer was about to shine after OS patching and scheduling to one thread of module and counts of execution threads increased overall. Now OS needs to be aware of CCX and treat it as NUMA node to extract optimal performance?

Memory reading software is not reading this nw uarch properly, bios is really bad right now, games are optimized for intels uarch, amds SMT is slightly different, windows drivers might need sorting.
Besides if you look outside of gaming performance is great and reviews are glowing.

Edit basically AMD cocked up the launch once again, they underestimated the focus on gaming performance, perhaps they figured they would be judged like intels HEDT, they were wrong.
The actual product though looks great.

 

[DarkKnightDude]

If I recall correctly, when Intel introduced Hyperthreading, wasn't there issues with it for awhile? Was that fixed by software or in another release?

 

[Whitestar127]

Why is Linus only testing 4K?

 

[rvborgh]

what's amazing to me is that AMD's first full go at SMT seems to perform better on average than Intels... i really am not sure why folks are complaining... this is a server chip that plays games well but is going to be an even better server chip... whereas previously AMD had a server chip that didn't play games all that great, and to top it off wasn't even a good server chip.

Most importantly... because of this AMD is going to survive.

So, with more reviews pouring, we can have a retrospect to events and "leaks" leading to the launch and discuss finer points of Zen:

1) AMD has been mum about CCX interconnects and L3 because they are hurting and will continue to hurt performance. 16MB of L3 per CPU is nice and dandy, but each core can access 8MB of those @ limited speed and that speed is also being hurt by coherency traffic - not optimal. I'd take Intel's multi ring HCC CPU any day over this. They better have some extra souce (in form of extra links?) for rumoured 32 core server chip, coherency traffic between 8 CCX chunks will kill it.
2) SMT has 10-15% perf penalty for being enabled, Intel has sorted out these things ~Nehalem gen. Very disappointing for desktop loads and okayish for servers.
3) OC rumours are pretty much shot down too, 1 core OC or not, this is another Polaris chip from AMD, loaded with features to extract the most out of process and bin, leaving scraps on table for overclockers. ~4Ghz is nice for 1700, but not so nice for top end CPU.
4) Larger L2 cache is nice, a lot of perf is coming out of it, goes a long way of showing how Intel has been milking customers with 256KB/s of L2. It turned out a lot of workloads can benefit very much from extra L2 and 4x as slow L3 is a gap too wide. SKL-X could have some major surprises for performance, some workloads will benefit nicely.

What stays the same between AMD releases is the tune of "wait for OS/game/software updates". We were supposed to get extra performance from Phenom when TLB bug was sorted out by recompiles, Bulldozer was about to shine after OS patching and scheduling to one thread of module and counts of execution threads increased overall. Now OS needs to be aware of CCX and treat it as NUMA node to extract optimal performance?

 

[french toast]

Majincry draw call performance looks shocking, i wonder what is causing all this?

 

[tidy]

.

 

[HurleyBird]

what's amazing to me is that AMD's first full go at SMT seems to perform better on average than Intels...

I'm not sure if it does or not. Unlike Intel's implementation, AMD's SMT seems like it may have a tangibly negative impact on performance even when only one thread is running per physical core. We certainly shouldn't take any penalty into account when talking about overall uplift, since that's just inflating the value and making a negative sound like a positive. Instead, the baseline figure should have SMT disabled entirely.

 

[SPBHM]

I don't see many people's problem. We never expected 7700k performance.

we expected it (1800X) to perform like at least a 6900K in gaming given that it performs like one in many other things and it's higher clocked; but it's not doing that right now, in part because of the loss of performance with SMT (which is something that really doesn't happen with the 6900K with HT), but not only that, if AMD can fix the performance loss with SMT it's going to be closing the gap significantly to what I expected.

If I recall correctly, when Intel introduced Hyperthreading, wasn't there issues with it for awhile? Was that fixed by software or in another release?

I think that's difficult to compare, Intel enabled HT on the P4 in 2002 (and they had it in the silicon since the original P4 in 2000), back then no games really used more than 1core/thread (and even excluding games, almost no desktop software took advantage of more than 1 core) Windows XP was not that well prepared to deal with it, while now most games use lots of threads and windows is prepared to deal with lots of cores and SMT (in the form of HT)

but look here of all the games tested no game lost performance with HT enabled
http://www.anandtech.com/show/1031/6

now I understand there were reports at the time (and much later, some people still claim HT penalizes gaming performance to this day), but it's a very different picture, look here, every single game lost a clear amount of performance:
http://www.hardware.fr/articles/956-7/impact-smt-ht.html

 

[tidy]

.

 

[MentalIlness]

Yeap, FX8150 today is as good or better vs Core i5 2500K in games and still faster in MT workloads

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/60-minutes-–-cpu-performance-and-energy-consumption-in-gaming-atenra.2459963/

Indeed About time to retire this FX 8120.....It has been installed since launch only to be reseated just once over the years. Never any issues, and it has served me well. It has done everything I needed it to pretty well.

What kind of performance jump would I be looking at by going with a RyZen 1600X or a 1700 ? Then Id probably just go to the Zen+ upgrade route to 8 core.

 

[tamz_msc]

Any advice, My main application is matlab,
With a variety of parallel and sequential work loads,

the 3 build I'm looking at are
~$900 for Ryzen 1700 ---The performance of the Ryzen in compute is tempting, but the platform seems like a risk at the moment.
~$900 for i7-7700K ---- Boring safe choice, might be the best
~$1200 for I7-6800K ---- more than I want to spend, but may be worth it for the compute and option to expand memory and gpu in future

any advice ?

That would depend on what you do with MATLAB. If you're doing signal/image processing, then FFTs would benefit from AVX on Intel. If all you do is data visualization, then the difference would be smaller. So, depending on how effectively you can utilize more cores, and what type of things you do, then the order would be 6800K~1700>7700K.

Though the Ryzen 7 1700 would be much better in perf/watt than either of those.

 

[flopper]

I don't see many people's problem. We never expected 7700k performance.

same fps performance in gaming as a 7700k and way way better in everything else.
7700k is dead.
Intel has nothing now, too expensive and not enough cores.

 

[lolfail9001]

same fps performance in gaming as a 7700k and way way better in everything else.
7700k is dead.
Intel has nothing now, too expensive and not enough cores.

How about performance in games people actually play? Like CS:GO on mins and MOBAs?
http://www.phoronix.com/scan.php?page=article&item=amd-ryzen-gaming&num=3

(Hint: CS:GO on Windows is not much better).
Oh, never mind, 1800X was a CPU bottleneck in 4k, 10/10.

Reality of it is: AMD are freaking liars, trying to market VM-farm reject as "gaming and enthusiast product". Won't stop me from upgrading to it, because on my merits it is still a decent value/efficiency product should they workaround the present issues.

 

[french toast]

How about performance in games people actually play? Like CS:GO on mins and MOBAs?
http://www.phoronix.com/scan.php?page=article&item=amd-ryzen-gaming&num=3

(Hint: CS:GO on Windows is not much better).
Oh, never mind, 1800X was a CPU bottleneck in 4k, 10/10.

Reality of it is: AMD are freaking liars, trying to market VM-farm reject as "gaming and enthusiast product". Won't stop me from upgrading to it, because on my merits it is still a decent value/efficiency product should they workaround the present issues.

What is your view on fritzchess? Did we get some benchmarks in the end?

 

[tamz_msc]

How about performance in games people actually play? Like CS:GO on mins and MOBAs?
http://www.phoronix.com/scan.php?page=article&item=amd-ryzen-gaming&num=3

(Hint: CS:GO on Windows is not much better).
Oh, never mind, 1800X was a CPU bottleneck in 4k, 10/10.

Reality of it is: AMD are freaking liars, trying to market VM-farm reject as "gaming and enthusiast product". Won't stop me from upgrading to it, because on my merits it is still a decent value/efficiency product should they workaround the present issues.

So people don't 'actually' play GTA V or Battlefield 1? On top of that you choose Linux to demonstrate gaming? SMH

 

[leoneazzurro]

Hmm... are you saying that people don't play BF1 , Overwatch, Witcher 3, The division to say at least something, And in 4k on the same review you linked the difference in CS:GO in 4K is between 130 FPS and 150 FPS, measurable, noticeable and having practically ZERO difference in real world experience.

 

[french toast]

People buy 8 core 1000$ cpus to play mobas dont you know, if it cant get 200fps there its obviously not an enthusiast product.

 

[Riek]

1) AMD has been mum about CCX interconnects and L3 because they are hurting and will continue to hurt performance. 16MB of L3 per CPU is nice and dandy, but each core can access 8MB of those @ limited speed and that speed is also being hurt by coherency traffic - not optimal. I'd take Intel's multi ring HCC CPU any day over this. They better have some extra souce (in form of extra links?) for rumoured 32 core server chip, coherency traffic between 8 CCX chunks will kill it.

Giving it does not seem to have an impact in a lot of well threaded applications, I wouldn't say it will be a bottleneck before we actually see it in action..

2) SMT has 10-15% perf penalty for being enabled, Intel has sorted out these things ~Nehalem gen. Very disappointing for desktop loads and okayish for servers.

The penalty seems to be very limited to gaming. In the average cases AMD HT scaling is far superior than Intels. See the stilt benchmarks. (It can also be questionned wether it is HT that is cullprit or your first point... it is not because you modify one parameter (HT on/off) that this might not impact the other bottleneck

What stays the same between AMD releases is the tune of "wait for OS/game/software updates". We were supposed to get extra performance from Phenom when TLB bug was sorted out by recompiles, Bulldozer was about to shine after OS patching and scheduling to one thread of module and counts of execution threads increased overall. Now OS needs to be aware of CCX and treat it as NUMA node to extract optimal performance?

True, but on bulldozer it was critical since it performed bad. For Ryzen the worst cases seem to indicate good performance instead of superior. So even without any fix, Ryzen will not bottleneck gaming, will still be far superior in its price class for many applications while being competitive in those it does 'badly'. Not sure if you can speak ill about such a design.

 

[lolfail9001]

What is your view on fritzchess? Did we get some benchmarks in the end?

http://www.pcgameshardware.de/Ryzen-7-1800X-CPU-265804/Tests/Test-Review-1222033/

Of course we did, about 10% behind 6900k in per clock. Could be worse, imho, the gaming came out of nowhere, however.

So people don't 'actually' play GTA V or Battlefield 1? On top of that you choose Linux to demonstrate gaming? SMH

Want to me quote the list of most played games on steam? Also, if you have Dota 2 benchmarks done on Windows, you are welcome to share those, i just know that they perform identically on Windows and Linux.

Hmm... are you saying that people don't play BF1 , Overwatch, Witcher 3, The division to say at least something, And in 4k on the same review you linked the difference in CS:GO in 4K is between 130 FPS and 150 FPS, measurable, noticeable and having practically ZERO difference in real world experience.

No, i am saying that these are far from the most played games in the world, even if they are way more popular than stuff like AotS, Hitman or Project CARS. Also, if CS:GO in 4k is CPU bottlenecked at 130 fps, guess what FPS will it have at CS:GO settings.

 

[Zucker2k]

Joker did a followup video about his 1700 @ 3.9GHZ ~ 7700k @ 5GHz

The Ryzen system also hiccups around .10, 1:48, and what was that at 250? At 3:30 he says "The 1700 falls behind a little....." then goes on to say he's taking that back; he's not going to give an opinion. I stopped watching at that point.

 

[tamz_msc]

The Ryzen system also hiccups around .10, 1:48, and what was that at 250? At 3:30 he says "The 1700 falls behind a little....." then goes on to say he's taking that back; he's not going to give an opinion. I stopped watching at that point.

So what? Witcher 3 has a noticeable frame time spike at the end of the benchmark sequence on the i7 7700K that isn't there on the 1700.

 

[french toast]

http://www.pcgameshardware.de/Ryzen-7-1800X-CPU-265804/Tests/Test-Review-1222033/

Of course we did, about 10% behind 6900k in per clock. Could be worse, imho, the gaming came out of nowhere, however.

Want to me quote the list of most played games on steam?

That would be great if you could, i cant post links else i would.
Oh could you pull up the data on what percentage of those are 8 core HEDT processors? Thanks.

Of course you wont because it will show most people are happy with a pentium or an i3, no one is stupid enough to buy a 2000$ pc just to play moba games when they can buy a 70$ pentium that does the job just fine.
If you can find a correlation between CS go and moba games and 8 core HEDT cpus on steam you will have a point.

 

[lolfail9001]

Oh could you pull up the data on what percentage of those are 8 core HEDT processors? Thanks.

Of those games? You're words here, mate.

Of course you wont because it will show most people are happy with a pentium or an i3, no one is stupid enough to buy a 2000$ pc just to play moba games when they can buy a 70$ pentium that does the job just fine.

You are exactly right: majority of people are happy with pentium or i3 or even laptops. Ryzen is not for those people either, right now. We are not majority of people, hence why we are talking about high end Skylakes and Ryzens.

If you can find a correlation between CS go and moba games and 8 core HEDT cpus on steam you will have a point.

You will have a point only if you prove that most people buying AAA games have HEDT rigs. We're tied, as such.

 

[PotatoWithEarsOnSide]

Damn. There was me preparing to do just that. LoL at 1000fps was my goal. Apparently that'll make me good at it.

Trolling is not allowed
Markfw
Anandtech Moderator