Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[guskline]

I hope imported.jjj. I had the 2 RX480s in my 6700k rig running at 4.6Ghz so I have a lot of benchmarks saved on that.

 

[Joric]

Not sure what's your point as all of those test at 1080p with a high end card and that's exactly the problem.
In real gaming you are GPU bound and the CPU's impact on perf is very different than when you create a CPU bottleneck.
It's like testing seq write for a SSD with random write and then claming that seq write sucks.

I'm not really interested in real world gaming scenarios so much as CPU performance in different applications.
The huge differences in comparative performance between productivity applications and heavily multithreaded games needs explaining in order for me to know whether it's likely to effect my current or future use scenarios.
Presumably the same would apply to Gamers that do frequent GPU upgrades though.

 

[Crumpet]

The review we've all been waiting for.

Considering how much minecraft can rag pc's.. i'm quite happy to see this

 

[nathanddrews]

Hardware Unboxed detailed Ryzen GPU Bottleneck Testing:

Around the 2min mark, you can read some of the extended testing done that hasn't been published yet...

 

[dfk7677]

That induced bottleneck is used too show their incompetence.
They don't test the CPU, the goal is to show gaming perf with that CPU and they create a situation that is not representative for gaming.
It's like testing a car's 0 to 60 with deflated tires.
They don't get valid data for 99% of users as it is a corner case.

If they would test at reasonable res the picture changes in a fundamental way.
Even for the 6900k, that thing is better than the 7700k in gaming(cost aside) but people don't know it because reviewers test at low res with a high end card.

I respectfully disagree. 7700K is much better than 6900K because of higher clocks (and some IPC) in most games (which are not multithreaded). Don't get me wrong, I would buy a 1700 over a 7700K anytime, as it is much better all around and also better in a properly multithreaded game like BF1, that I play.

Still if you want to measure CPU performance in a game, you have to make sure that the GPU is not bottlenecking the system. I don't know if that means 720p or 1080p on low settings, you just have to make sure that GPU usage is less than 100%.

Another way to measure CPU (or GPU for that matter) performance that would be much closer to real world situations, would be to count time that the system doesn't achieve vsync framerates on different resolutions/refresh rates (e.g. 720p@60Hz-144Hz, 1080p@60-144Hz, 1440p@60Hz-144Hz, 4k@60Hz) all with low medium, high and ultra presets. But this I guess would be very time consuming.

 

[imported_jjj]

I respectfully disagree. 7700K is much better than 6900K because of higher clocks (and some IPC) in most games (which are not multithreaded). Don't get me wrong, I would buy a 1700 over a 7700K anytime, as it is much better all around and also better in a properly multithreaded game like BF1, that I play.

Still if you want to measure CPU performance in a game, you have to make sure that the GPU is not bottlenecking the system. I don't know if that means 720p or 1080p on low settings, you just have to make sure that GPU usage is less than 100%.

Another way to measure CPU (or GPU for that matter) performance that would be much closer to real world situations, would be to count time that the system doesn't achieve vsync framerates on different resolutions/refresh rates (e.g. 720p@60Hz-144Hz, 1080p@60-144Hz, 1440p@60Hz-144Hz, 4k@60Hz) all with low medium, high and ultra presets. But this I guess would be very time consuming.

Reviewrs are trying to measure gaming perf not CPU perf and that's the problem as they don't do that at low res.
Measuring CPU with a game would be like that at low res and that's how this idiocy started but it turned into measuring gaming perf. Ofc measuring CPU perf with a game is pointless as it's worse than synthetics but w/e.
Reviewers have measured at low res and concluded that Ryzen sucks at gaming, that's the problem. as they haven't even tested gaming except for the tiny niche that needs very high FPS.
90% of discrete desktop GPU sales are bellow 250$ and even the bulk of the 10% that buys higher end, won't game at 1080p

Ryzen does have a problem in games that seems latency related, not saying it doesn't but the impact might be almost irrelevant, hard to say for sure as almost nobody looked at gaming at realistic resolutions and many reviewers had a crippled BIOS.

As for 7700k vs 6900k, do some research and look at results at realistic res and you will agree with me.Because some games scale, overall the 6900k is already better, value aside.

 

[cytg111]

many many walls of text later i've given up on staying "current" on this launch.
Can someone sum up for me(and others), has the bottleneck(s) regarding gaming performance been identified? What is the current hypothesis?

 

[imported_jjj]

Great news! https://www.techpowerup.com/231204/...s-and-fortis-series-ddr4-memory-for-amd-ryzen
G.SKILL's R&D team achieved an impressive overclock on a 64GB (4x16GB) kit to frequencies of DDR4-3200MHz at ultra-low timings of CL14-14-14-34.
Further pushing the boundaries of 16GB (2x8GB) memory kits on the AM4 platform, G.SKILL tuned it to a blistering fast DDR4-3466MHz CL 16-16-16-36

 

[KTE]

I guess I will write more later upon finding time...

Pure performance wise, AMD has done an amazing job in every respect. No one can sniff at that. It is AMDs Conroe, albeit only vs AMD themselves.

For anything Media/DAC/Encryption related, Ryzen is a sure bet. MT performance in a lot of applications is absolutely awesome. You have got to admit. It gives BDe a very good challenge at a fraction of the cost.

The cache, coherence and mem latencies are very poor. Along with SMT and driver problems, they will produce sucky results in many mem sensitive and nonoptimized benchmarks. Games, for certain. Going forward this should improve somewhat but don't hold your breath on it. AMD should've known all of this.


Debunk That Hype

AMD has a strong internet fanbase. I consider myself a fan.

Then there are what we call AMD fanatics, far removed from reality. Like their counterparts from Intel, they don't understand science, data or reason. Their purpose is just to try and spin everything AMD to the best, craziest, light existable.

Posting support in hoardes, doesn't bolster the accuracy of your belief.

They were seriously delusional on many fronts for the past 4 months on here, creating this hugely wishful hype that has inturn made Ryzen look average upon release. They pushed unrealistic expectations in everyone's face, which has hurt AMDs image in the end.

Upon reviews, they post frenetically trying to make the same excuses to defend AMD, excuses we've heard since Phenom. This is a sorry state.

All of the unrealistic nonsense I kept seeing, reviews have debunked:

1. Blender/POVRay was AMDs best case. Selected marketing. All of Horizon was pure marketing.

2. Doing everything altogether in one uarch, it's obvious the platform has A LOT of teething issues, and clocks were problematic. No wonder the delays. The platform is a beta. End users and reviewers NEVER have to wait for all this to be sorted. It is judged how it is sold.

3. Low Power Plus is Low Power Plus! I heard so much irrational pseudoscience nonsense in the buildup here. Every one of it has been debunked by data now.

It is obvious power or process is absolutely no where close to Intel. That 1800X is choking being pumped +30W from the model below.

Clocks/volts/currents are ceiling, OC minimal, XFR a gimmick suited to mobile and power way above 90W, and above Intels 140W chips when properly tested.

No, sorry to all irrational pseudoscience. No magic 0.9v 4GHz at less than 80W because of a Neon FPU.

4. Piledriver vs Exc tests for IPC show 2% average difference now.

And Ryzen ST isn't 1-7% like the hype, but 10-20% behind Intel.

5. For the average guy, Ryzen is certainly not the gamers CPU. 4C, high IPC is still king. Intel has better buys, especially for futureproofing. Excuses don't mitigate that CPU load tests - which give a proper picture at all ranges - show it well behind.

And seriously. Argue all you like but...110fps vs 100fps is NO DIFFERENCE to a gamer! I played competitive FPS for years since Quake. Charts showing +100FPS are only good to ascertain the technical 'better' but not for actual playability.

6. BitsandChips fed all the wrong zealous hype trains. Seems apparent they just wanted to cash in. Their latest linking a 1% runtime variation in CB to 'Neural Net Prediction'is equally ludicrous. It's called margin of error, for Christ's sake.

7. HEDT doesn't care for price or power. It cares about absolute performance. Which is, still, ruled by Intel.

AMD has now given Intel a challenger for certain workloads, however.

8. TheStilt did an awesome job! Should be renamed TheKanterStilt.

wow Amazing Review Phenom vs Ryzen !

https://www.computerbase.de/2017-03/benchmarks-phenom-ii-x6-ryzen-7-vergleich/

Now that's actually a quality data point.

Sent from HTC 10
(Opinions are own)

 

[PotatoWithEarsOnSide]

Review shows 1700x ~ 7700k with gtx1080

1080p very high no aa

The most noteworthy thing that I saw in that video was the minimum framerates for the R7 1700x and the 7700k. Time after time, the benchmarks seem to be showing that the Ryzen CPUs do have lower framrates on average, by they sure as damnit don't have the lowest minimum framerates. On that point, how important is framerate consistency to gamers?

 

[OTG]

whoops

 

[noneis]

There seems to be some kind of issue with lower clocked memory
https://www.computerbase.de/2017-03...4/#abschnitt_ddr43200_vs_ddr42666_vs_ddr41866

Similiar performance with @2666 CL14, @2666 CL16 and @3200 CL16, but 15% gaming performance drop when @1866 CL16 is used.

 

[PotatoWithEarsOnSide]

Ok, so more interesting viewpoints from that Hardware Unboxed video.
I get the 1080p testing, but I don't see the conclusions that others have drawn as a result of it.
Gamers seem to be asking "what is the best gaming CPU now?" but using a metric that tries to foresee the future of gaming, i.e. "yes, there's little difference in 4k now, but with better GPU tech going forward, the Ryzen CPUs weakness is exposed."
What this fails to address is:
a) future upgrade path from 7700k,
b) increased multi-threaded games moving forward (they were recommending i5s only last year, but coudn't now)
The analysis is both right and wrong at the same time.

IMO, the recommendations given to those currently looking at an upgrade should be along the lines of:
"Ryzen, yes, but..."
and...
"KL, yes, but..."

Neither are perfect for gaming in the future, but the Ryzen option seems to be the most future-proof. Obviously this could change should Intel introduce more cores at a sensible price level, whilst also not insisting on changing platforms every 2nd iteration.

All told, the R7 1700 OCed to 3.9GHz looks a snippet, especially seeing as it is achieving that on the B350 MOBOs.

 

[french toast]

I guess I will write more later upon finding time...

Pure performance wise, AMD has done an amazing job in every respect. No one can sniff at that. It is AMDs Conroe, albeit only vs AMD themselves.

For anything Media/DAC/Encryption related, Ryzen is a sure bet. MT performance in a lot of applications is absolutely awesome. You have got to admit. It gives BDe a very good challenge at a fraction of the cost.

The cache, coherence and mem latencies are very poor. Along with SMT and driver problems, they will produce sucky results in many mem sensitive and nonoptimized benchmarks. Games, for certain. Going forward this should improve somewhat but don't hold your breath on it. AMD should've known all of this.


Debunk That Hype

AMD has a strong internet fanbase. I consider myself a fan.

Then there are what we call AMD fanatics, far removed from reality. Like their counterparts from Intel, they don't understand science, data or reason. Their purpose is just to try and spin everything AMD to the best, craziest, light existable.

Posting support in hoardes, doesn't bolster the accuracy of your belief.

They were seriously delusional on many fronts for the past 4 months on here, creating this hugely wishful hype that has inturn made Ryzen look average upon release. They pushed unrealistic expectations in everyone's face, which has hurt AMDs image in the end.

Upon reviews, they post frenetically trying to make the same excuses to defend AMD, excuses we've heard since Phenom. This is a sorry state.

All of the unrealistic nonsense I kept seeing, reviews have debunked:

1. Blender/POVRay was AMDs best case. Selected marketing. All of Horizon was pure marketing.

2. Doing everything altogether in one uarch, it's obvious the platform has A LOT of teething issues, and clocks were problematic. No wonder the delays. The platform is a beta. End users and reviewers NEVER have to wait for all this to be sorted. It is judged how it is sold.

3. Low Power Plus is Low Power Plus! I heard so much irrational pseudoscience nonsense in the buildup here. Every one of it has been debunked by data now.

It is obvious power or process is absolutely no where close to Intel. That 1800X is choking being pumped +30W from the model below.

Clocks/volts/currents are ceiling, OC minimal, XFR a gimmick suited to mobile and power way above 90W, and above Intels 140W chips when properly tested.

No, sorry to all irrational pseudoscience. No magic 0.9v 4GHz at less than 80W because of a Neon FPU.

4. Piledriver vs Exc tests for IPC show 2% average difference now.

And Ryzen ST isn't 1-7% like the hype, but 10-20% behind Intel.

5. For the average guy, Ryzen is certainly not the gamers CPU. 4C, high IPC is still king. Intel has better buys, especially for futureproofing. Excuses don't mitigate that CPU load tests - which give a proper picture at all ranges - show it well behind.

And seriously. Argue all you like but...110fps vs 100fps is NO DIFFERENCE to a gamer! I played competitive FPS for years since Quake. Charts showing +100FPS are only good to ascertain the technical 'better' but not for actual playability.

6. BitsandChips fed all the wrong zealous hype trains. Seems apparent they just wanted to cash in. Their latest linking a 1% runtime variation in CB to 'Neural Net Prediction'is equally ludicrous. It's called margin of error, for Christ's sake.

7. HEDT doesn't care for price or power. It cares about absolute performance. Which is, still, ruled by Intel.

AMD has now given Intel a challenger for certain workloads, however.

8. TheStilt did an awesome job! Should be renamed TheKanterStilt.


Now that's actually a quality data point.

Sent from HTC 10
(Opinions are own)

A large chunk of what you said i largely agree with, there was alot of hype and in the build up, ryzen SHOULD be judged on when it is sold, beta stage or not, 14lpp IS a low power optimised process, the stilt proved that,below < 3.5ghz Ryzen is amazingly efficient even on crap bios and drivers, past 3.5ghz it falls off of a cliff.
So bravo for cutting through some of the hype. #claps hands rather vigorously.

Right, i disagree with other parts, against broadwell it show similar ipc, sometimes a bit lower and sometimes abit better, SMT outside of gaming is superior with ryzen, as is efficiency.
Broadwell however is better with avx 2 (not used much), overclocks better, offers quad channel memory and as things stand- is better in gaming (until games are updated, then i expect equality)
If you look at productivity benchmarks they are practically equal, so its fair to say broadwell ~ Ryzen.
Skylake is obviously better still, but currently offers half as many threads, dampening its appeal for productivity, skylake x fixes this but will likely be twice the price.

Then you get to pricing of both cpu and mobo, no contest amd has nailed this one.
Overall ryzen is very competitive and a complete bargain, even if it was overhyped a bit and still buggy on launch like you say.
My picks would be 6800k or r7 1700, both great choices imo.

 

[Gikaseixas]

Interesting review from Tech Deals
https://www.youtube.com/watch?v=sciuiEcrnzg

 

[CentroX]

Here is my system setup:

it is OC to 4025 mhz all cores.

1800X @ 4025 mhz + Asus Crosshair VI Hero + 16GB DDR4 3000mhz memory.

- my system

- ryzen master setup

- validated cpu-z

- single thread cpu-z

- multi thread cpu-z

- cpu z

cinebench R15

- cinebench R15

http://i.imgur.com/fKN5sht.png - old system

When I set the memory to 2933 mhz my system froze. Doing 2667 for now.

 

[tential]

I respectfully disagree. 7700K is much better than 6900K because of higher clocks (and some IPC) in most games (which are not multithreaded). Don't get me wrong, I would buy a 1700 over a 7700K anytime, as it is much better all around and also better in a properly multithreaded game like BF1, that I play.

Still if you want to measure CPU performance in a game, you have to make sure that the GPU is not bottlenecking the system. I don't know if that means 720p or 1080p on low settings, you just have to make sure that GPU usage is less than 100%.

Another way to measure CPU (or GPU for that matter) performance that would be much closer to real world situations, would be to count time that the system doesn't achieve vsync framerates on different resolutions/refresh rates (e.g. 720p@60Hz-144Hz, 1080p@60-144Hz, 1440p@60Hz-144Hz, 4k@60Hz) all with low medium, high and ultra presets. But this I guess would be very time consuming.

This is my issue DFK. You say most games are not multithreaded.
So an i7 7700k is a better choice than the 6900k.
If this is the case, why in gods name are you guys trying to compare the i7 7700k to the 1700?
It's an unfair fight because MOST GAMES aren't multi threaded.
Just because the i7 7700k and 1700 are the same price does NOT mean they are comparable products.

Ryzen 5 and 3 should be compared to the i7 7700k. Ryzen 7 should ONLY be compared to Intel's HEDT platform. Doing anything else is completely misleading to consumers.

Ryzen 5 is the far better gaming choice if you want an AMD system. You get a cheaper product that doesn't have extra unused cores that give off more heat and my guess is it will clock higher even.

Again, I don't understand why no one has simply disabled the cores on their 1700 and tested how Ryzen 5 does against the i7 7700k.

If you're ea GAMER that's what matters.

Ryzen 7 should be only compared to the intel HEDT platform PERIOD. Doing anything else is disingenuous.

 

[tential]

Also, if SMT actually worked correctly, then Ryzen would be a completely different processor. If you turn SMT off for games and turn Ryzen 7 into an effective 8 threaded CPU, it does better. So SMT and scheduling must not be working well together right now.

Fix that, and the Ryzen 3 CPU should do 95% as well as the native Ryzen 7 with SMT disabled.
Just like if you turn hyper threading off on an i7 it'd be like an i5.....

 

[KTE]

To be very honest, review quality en-masse -- for AMD related reasons -- has actually been very poor compared to what we have come to expect.

I'll give it a few more days before saying much.

I usually wait for ht4u's article, as for me, they are rarely matched on authenticity and thoroughness. In-line with this, they did not release a Ryzen review, saying (translated): http://ht4u.net/news/32692_amd_ryzen_startet_-_but_ryzen_musst_wait/

Today, AMD officially launches the sale of the long-felt Ryzen processor. Mainboards and processors in the form of the R7 1800X, R7 1700X and R7 1700 go on the market. Information about the basic data was already in the run-up, because AMD did not hold behind the mountain. The Ryzen 7 1800X is, according to AMD promise, the fastest current eight core processor. But today everything is still a bit stuck, and our article will be left out for the time being.

We had already announced last year that we will not accept any more 3-day Launchs. Unfortunately, the Ryzen test set was promised to us on February 23, 2017, but it did not reach us until 28th February 2017, as with most editorial offices. In the short time a reasonable test to create a completely new generation of AMD processors is simply impossible for us, and we do not want to present half-life work. We have benchmark results, but we do not show them yet. We have measurements for the power consumption, but we do not show yet - Ryzen must wait. Just until we have all the results and these are evaluated.

We can confirm that the AMD information on the basic data is true. We must also note that the final mainboards have only arrived at the partners (all obviously). It is hailing new BIOS updates and e-mails from AMD, which should be considered.

This includes, among other things, the fact that you have to disable the "balance mode" in the energy-saving options in Windows in order to get the best performance. And also the hint that the SMT capability of the processor can push the performance among games. These should be disabled in appropriate benchmarks if possible. However, our present test pattern did not yet offer an appropriate option. The memory support is partly very limited by the hardware - one knows however of the AMD processors and their memory controller, so that one must first approach here. But there should be a new BIOS update, and the tools do not support Ryzen correctly. AMD describes its tool Ryzen Master Utility as an early beta. The testers are therefore quite a bit on the fly or powerful in the stress.

What we can definitely say is that our kit arrived with the last BIOS update very fluently and apparently smoothly performs its services and feels fast loading times. The few, evaluated scores so far, however, are less pleasing because, with the SMT option activated, AMD is often behind its targets.

Ryzen reads from the key data highly attractive, the whole environment needs but apparently still care with regard to performance (games) and compatibility (memory), which might have been better in advance. We present our article as soon as we are through the work through.

 

[unseenmorbidity]

Well the induced bottleneck is used to show how high the CPU performance is. Of course almost nobody will use it in 720p, but at the same time, who will buy it to run synthetic benches all day long?

Yah, that's what they say, but how is this relevant?

How is 720p on low settings with a OC'd GTX 1080 relevant!? It's not even a niche situation... It's completely detached from reality and utterly meaningless.

The only defense I have heard thus far is "well, I might go buy another GPU later". Okay... Except, it's the 8 core that is more futureproof, so yah...

If someone can explain how this is relevant information, then I would be most grateful, because I don't see it. It looks like meaningless data. Like seeing how many cookies you can stuff in your mouth. Okay, great! You got six cookies in there... But how is this information useful?

EDIT: Actually, not totally worthless. There are points where Watch Dogs 2 is hammering the CPU and GPU at 720P Low, and getting less than 70 FPS. lol! I figured that game was optimized poorly, but wow! I am surprised people even include it in benchmarks.

 

[AMDisTheBEST]

Yep, these people care so much about benchmarks at 720p because they can't wait to game on console resolution. I bet they have 4 SLI Pascal Titans just to output another hundred or so 720p frames on their 720p, 60 hertz refresh monitor

 

[cytg111]

...

The cache, coherence and mem latencies are very poor. Along with SMT and driver problems, they will produce sucky results in many mem sensitive and nonoptimized benchmarks. Games, for certain. Going forward this should improve somewhat but don't hold your breath on it. AMD should've known all of this.

- Very nice summary, thanks . So consensus is that Ryzens achilles heel is 1. clocks and power and 2. memory and cache latencies.
Number one can grow over time but the second is worse I gueess, no amount of driver/firmware updates is gonna reduce those latencies, right? So for a better gaming SKU from AMD we will have to wait and see what Ryzen++ brings forth. Right?

Given the current OC headroom, or lack thereof, suggests that its pushed pretty close to the limit... Would there not be a point to getting a low leakage 65watt part, clock it 500MHz down and see how far you can undervolt it?

(also, linkage to the stilts work?)

 

[AMDisTheBEST]

Doed anyone of a review using a Polaris or Fiji? Or any Radeon?

I have one review of ryzen paired with fury nano and Rx 480 benchmarking on Linux.

 

[OrangeKhrush]

Again the issues here are plenty and they are power state related.
Over the weekend not only have I spoken to a DRAM manufacturing company insider responsible for the high end kits, but at least two individuals from board vendors working in RnD/Development.
I have shared my findings with them and they will investigate them thoroughly. As stated before, reproducing the low performance figures is easy enough, but also the high ones as well with consistency.
If I get time I will post several screenshots here which show this.
All of the suggestions and points made in the OCN thread are true no question, but the primary issue here I still believe and have tested to be power management.
I recovered higher than 2133MHz frequencies again, then I lost them because I changed one setting and the CPU/DRAM/IMC couldn't handle that change. So "d2" and that was it. Not able to get back to high frequency DRAM again. I have to leave the CPU out of the system for a good 30min, and try again and it should be sorted.

With Ryzen 4GHZ is not 4GHz. As to the scaling though of the clock, it is down to design but the process node as well and I have this on good info from someone who works at another fab. Thay process will not scale unless it is vastly changed, to the point where it isn't a LLP anymore. The tooling for this was taken from SAMSUNG's Low power process in the first place. Clocks will not magically increase until AMD goes to a completely different node/process or both.

 

[ccronos]

in new For Honor 1800x looks good
http://pclab.pl/art73043-4.html