AMD Ryzen's First Game Optimization: 'Ashes Of The Singularity: Escalation,' Tested

[jsoosr]

This just goes to show that comparing and established platform that has had yrs of optimizations to a new platform and drawing hard conclusions is a questionable endeavor. Stardock, Bungie, and Bethesda have either already started or plan on optimizing for the Ryzen platform. These preliminary results of beta realease optimizations show that Ryzen is in the same ballpark as Intel in pure performance and certainly superior on a cost basis as a gaming CPU. Stardock spent a little over 400 hrs optimizing for the Ryzen platform so far compared to Yrs on the Intel platform so still has headroom there to grow I would think. That said, as intel transitions to cannon lake and beyond I am sure they will respond forcefully to the challenge from AMD as they have in the past. My take is that AMD is likely to remain truly competitive with Intel not just a budget alternative until Intell transitions to 10nm process then all bets are off. That said if you are in the market for upgrading now because your system is aging badly there is no reason not jump on the AMD bandwagon. The more interesting battle is yet to come when the R5 1600x comes out clocked at the same level as the R7 1700x with 2 cores disabled. If that partn can remain in the same neighborhood performance wise Then AMD will have hit a home run.

 

[Josh_194]

Please stop testing CPUs by benchmarking video games. Bottlenecking the GPU doesn't eliminate enough variables to analyze how well the CPU performs. This is bad test engineering that showed performance issues stemming from software/hardware interaction. The test community falsely claimed Ryzen was "bad" for gaming, and gave millions of consumers bad data points.

If you're saying they should test non-gaming application performance, there's plenty of that in the actual Ryzen 7 CPU reviews.
If you're saying benchmarking video games is not a good way to measure video game performance... what? Or if there's something specific about the way that they're benchmarking that you don't like, you'll have to elaborate.

Most people didn't say that Ryzen was bad for gaming. They said that Ryzen 7 CPUs are typically beat by Intel i5/i7 quads (which are cheaper), meaning that Ryzen 7 CPUs offer worse performance per dollar if all you're doing is gaming. Which is exactly what is said about Intel's LGA 2011 CPUs for the most part. Once Ryzen 5 comes out, I imagine we'll see more recommendations for Ryzen CPUs in gaming-centric builds.

You're not getting the point. We see the CPU performance in synthetic benchmarks such as Cinebench and Handbrake. These synthetic benchmarks are made to drive the CPU cores independently and test the response. When testing a video game, you are essentially testing the entire system along with how the software and hardware interact. Ryzen 7 will eventually run software with similar performance as shown in the synthetic benchmarks; but it will take time for software developers to take advantage of the new architecture. When it comes to the CPUs, Ryzen 7 is basically on par with the i7-6900k in multiple threaded and single threaded applications. Ryzen 7 beats the i7-7700k in multiple threaded applications and loses in single threaded applications. The overwhelming verdict that Ryzen 7 is essentially a workstation CPU is ridiculous. Ryzen 7 performs well, relative to competition, on all tasks with the ability to run multi-threaded applications while maintaining a high clock speed for single-threaded applications. Match that with Ryzen 7's energy efficiency and you may have the best CPU to hit the market in years. But instead of the correct verdict, we are stuck on 1080P gaming with top of the line GPUs... this is not only a viciously stupid testing environment, it's not even a real scenario. It's neither a decent synthetic benchmark or real-world benchmark.

 

[Soaptrail]

Bonscott, and Tom's first review barely mentioned much about RAM speed and Rank and i had to go back and hunt for the DDR speeds used. I know they were rushed but Tom's is usually good about explaining that stuff.

 

[InvalidError]

You're not getting the point. We see the CPU performance in synthetic benchmarks such as Cinebench and Handbrake.

Cinebench and Handbrake aren't synthetic benchmarks, they are real applications used by real people to do real work and their benchmark results are based on the time it takes to complete a given task or set thereof. The only "synthetic" thing about them is repeating the exact same job as in real life, you'd do the exact same job only once.

 

[spdragoo]

Please stop testing CPUs by benchmarking video games. Bottlenecking the GPU doesn't eliminate enough variables to analyze how well the CPU performs. This is bad test engineering that showed performance issues stemming from software/hardware interaction. The test community falsely claimed Ryzen was "bad" for gaming, and gave millions of consumers bad data points.

If you're saying they should test non-gaming application performance, there's plenty of that in the actual Ryzen 7 CPU reviews.
If you're saying benchmarking video games is not a good way to measure video game performance... what? Or if there's something specific about the way that they're benchmarking that you don't like, you'll have to elaborate.

Most people didn't say that Ryzen was bad for gaming. They said that Ryzen 7 CPUs are typically beat by Intel i5/i7 quads (which are cheaper), meaning that Ryzen 7 CPUs offer worse performance per dollar if all you're doing is gaming. Which is exactly what is said about Intel's LGA 2011 CPUs for the most part. Once Ryzen 5 comes out, I imagine we'll see more recommendations for Ryzen CPUs in gaming-centric builds.

You're not getting the point. We see the CPU performance in synthetic benchmarks such as Cinebench and Handbrake. These synthetic benchmarks are made to drive the CPU cores independently and test the response. When testing a video game, you are essentially testing the entire system along with how the software and hardware interact. Ryzen 7 will eventually run software with similar performance as shown in the synthetic benchmarks; but it will take time for software developers to take advantage of the new architecture. When it comes to the CPUs, Ryzen 7 is basically on par with the i7-6900k in multiple threaded and single threaded applications. Ryzen 7 beats the i7-7700k in multiple threaded applications and loses in single threaded applications. The overwhelming verdict that Ryzen 7 is essentially a workstation CPU is ridiculous. Ryzen 7 performs well, relative to competition, on all tasks with the ability to run multi-threaded applications while maintaining a high clock speed for single-threaded applications. Match that with Ryzen 7's energy efficiency and you may have the best CPU to hit the market in years. But instead of the correct verdict, we are stuck on 1080P gaming with top of the line GPUs... this is not only a viciously stupid testing environment, it's not even a real scenario. It's neither a decent synthetic benchmark or real-world benchmark.

And you're missing the part where they already tested Ryzen with non-gaming benchmarks (both synthetic & real-world):
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-8.html: testing the 1800X for desktop/office applications, including compressing files with 7-Zip, encoding files with Handbrake, Python programming, etc.
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-9.html: workstation testing with AutoCAD, Creo, Solidworks, etc.
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-10.html: testing for scientific & engineering environments, including Euler3D, LAMMPS, Convolution, etc.
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1700x-review,4987-7.html: testing of the 1700X in some of the same desktop/office benchmarks they tested the 1800X with.

So, again, not sure what your complain is, since Tom's Hardware is obviously testing the Ryzen chips with more than just gaming benchmarks. They just also have articles that focus on gaming because a) a signficant portion of the membership here is interested in gaming, b) the particular testing done in this article (1080p resolution, top-line GPU paired with Intel & Ryzen CPUs) helps spotlight Ryzen's performance in the games in question, & c) even for those not necessarily technically inclined or not a "power" PC user, gaming benchmarks is one of those areas where you can not only get measurable results from your testing but also actively notice the resutls...unlike, say, that Adobe CC Photoshop Light/Heavy results, where even though the Ryzen 1800X & i7-7700K were noticeably faster than the other chips, every CPU still managed to finish it in under 1 or 2 seconds.

 

[bit_user]

Seeing these results, I wouldn't call it an AMD-specific optimization. AMD gains the most, but the Intel processors also see small improvements in FPS, and the 7600k even improves its frame-time variance. In an industry that loves to squeeze out every resource in optimizations, this might encourage developers to take a closer look.

I do wonder how much single-threaded code can be optimized for AMD. One thing I've read is that Ryzen has a half-width AVX pipeline, meaning that only half of an AVX instruction can be dispatched per-cycle. So, it's probably not like there's some magic instruction(s) that people can just start using in their code to unlock Ryzen's performance.

That leaves me wondering whether the optimizations were more along the lines of threading tweaks. Although, it's then curious that the i7-6900k wouldn't benefit more.

Final thought is that it might be cache-related. In that case, the optimizations should provide a smaller benefit when used with faster RAM.

 

[Fallen88]

If it is true that Ryzens are going into the next generation consoles this will ensure that it gets thoroughly optimized for.

 

[TJ Hooker]

If it is true that Ryzens are going into the next generation consoles this will ensure that it gets thoroughly optimized for.

Pretty sure people said the same thing about AMD's FX CPUs after the XB1 and PS4 were released with 8-core AMD CPUs inside.

 

[Petaflox]

The strength of Ryzen is the "prediction" technology witch make the cpu very good
in those program and benchmark with looped instruction, the way to play games is "unpredictable" and Ryzen fall to Intel cpu.

 

[Rolo473]

Ryzen is just going to be like the Rx 400 series. Slower then the competition at launch but overtime through driver updates it'll get faster and catch up. AMD is back baby!

 

[ARICH5]

very impressive.putting aside the tom's math and memory optimization id say this is a good step forward for AMD. a "first" optimization and a 5 fps leap in one game is quite good.

 

[cryoburner]

Older games may never receive patches, and the fact that Stardock (Oxide) has updated only its latest product is telling. However, the company hasn't indicated if we can expect an update for the original Ashes of the Singularity in the future, so it remains a possibility.

You shouldn't expect an update for the original Ashes of the Singularity, because it's a discontinued product that isn't even sold anymore. The original Ashes of the Singularity and Escalation were merged a couple months back, and owners of the original game were all upgraded to Escalation. See this...
https://steamcommunity.com/app/507490/discussions/0/133257324791253705/

So no, it's not necessarily telling that they haven't updated the original, since that version of the game has been discontinued, and escalation has taken its place.

...When testing a video game, you are essentially testing the entire system along with how the software and hardware interact. Ryzen 7 will eventually run software with similar performance as shown in the synthetic benchmarks; but it will take time for software developers to take advantage of the new architecture.
...But instead of the correct verdict, we are stuck on 1080P gaming with top of the line GPUs... this is not only a viciously stupid testing environment, it's not even a real scenario.

Gaming benchmarks like these are a perfectly sensible way to measure a CPU's performance in existing games. Sure, in future games, Ryzen might potentially have an edge over Intel's current CPUs, but there's no way to accurately measure what the performance in games released a year or two from now will be like. Anyone buying a CPU for gaming today is going to want to know how that CPU will perform in today's games, since very few of these existing titles will be optimized for the new hardware, and tests like these can provide that information.

I also don't see why you're complaining about this particular article. The article was highlighting how some game-specific optimizations made for Ryzen improved the CPU's performance significantly, allowing the 1800X to outperform the 7700k, which is generally considered to be the fastest processor in most existing games. The 6900k may have technically outperformed it, but that's a $1050 workstation processor that generally performs worse than a 7700k in most games, making it pretty impractical for a gaming build. And of course, if an 1800X can perform on par with the 7700k in this game, a 1700 overclocked to 1800X speeds should be able to as well, as might a 1600 or 1600x, at a much lower price point.

Also, I mostly disagree with your complaints about the testing methodology. Using a high-end GPU reduces the possibility of the graphics card limiting framerates. Using such a GPU at 1080p is also arguably a real-world scenario as well, for anyone gaming on a 144Hz screen. It can also be relevant for anyone using a mid-range GPU at reduced graphics settings, since lowering detail levels can likewise prevent the GPU from being the limiting factor in a game's performance. It is, of course, worth pointing out that on a 60Hz screen, or at higher resolutions and detail levels, the GPU's performance will likely limit framerates to levels below the CPU performance caps, but tests showing equal performance on all processors wouldn't provide much useful information for a CPU review beyond that.

 

[valzero]

Where are the DX11 numbers? We gamers select what give us the most FPS.

 

[valzero]

Where are the DX11 numbers for comparison? We all avoid DX12 like the plague because of low FPS.

 

[Todd Melanson]

#1) Games are the only way to benchmark hardware that gamers should REALLY be concerned about unless you're streaming as well. Artificial benchmarks mean nothing in the real world.

#2) Most competent testers test at 1080p because less than 2% of gamers play at a resolution higher than 1080p...and that includes those with hardware able to run at higher resolutions. Having able hardware doesn't mean there's necessarily compelling content to consume at higher resolutions.

#3) As with any new platform, Ryzen is still a fetus as a market segment. But clearly it is going to gain wide adoption quickly, and developers are going to be incentivized to optimize their offerings for it, just as happened for the FX8350, only this is going to happen MUCH more quickly. AMD has hit it out of the park with Ryzen and the falling prices of Intel chips is proof.

#4) Lastly, when Vega GPUs start hitting the market and AMD starts realizing their long term goal (began over a decade ago) of a unified platform design architecture for cpus and gpus where your entire PC can share resources... Processing and RAM... and then we could see the REALLY exciting stuff in the next five years.

 

[Josh_194]

Please stop testing CPUs by benchmarking video games. Bottlenecking the GPU doesn't eliminate enough variables to analyze how well the CPU performs. This is bad test engineering that showed performance issues stemming from software/hardware interaction. The test community falsely claimed Ryzen was "bad" for gaming, and gave millions of consumers bad data points.

If you're saying they should test non-gaming application performance, there's plenty of that in the actual Ryzen 7 CPU reviews.
If you're saying benchmarking video games is not a good way to measure video game performance... what? Or if there's something specific about the way that they're benchmarking that you don't like, you'll have to elaborate.

Most people didn't say that Ryzen was bad for gaming. They said that Ryzen 7 CPUs are typically beat by Intel i5/i7 quads (which are cheaper), meaning that Ryzen 7 CPUs offer worse performance per dollar if all you're doing is gaming. Which is exactly what is said about Intel's LGA 2011 CPUs for the most part. Once Ryzen 5 comes out, I imagine we'll see more recommendations for Ryzen CPUs in gaming-centric builds.

You're not getting the point. We see the CPU performance in synthetic benchmarks such as Cinebench and Handbrake. These synthetic benchmarks are made to drive the CPU cores independently and test the response. When testing a video game, you are essentially testing the entire system along with how the software and hardware interact. Ryzen 7 will eventually run software with similar performance as shown in the synthetic benchmarks; but it will take time for software developers to take advantage of the new architecture. When it comes to the CPUs, Ryzen 7 is basically on par with the i7-6900k in multiple threaded and single threaded applications. Ryzen 7 beats the i7-7700k in multiple threaded applications and loses in single threaded applications. The overwhelming verdict that Ryzen 7 is essentially a workstation CPU is ridiculous. Ryzen 7 performs well, relative to competition, on all tasks with the ability to run multi-threaded applications while maintaining a high clock speed for single-threaded applications. Match that with Ryzen 7's energy efficiency and you may have the best CPU to hit the market in years. But instead of the correct verdict, we are stuck on 1080P gaming with top of the line GPUs... this is not only a viciously stupid testing environment, it's not even a real scenario. It's neither a decent synthetic benchmark or real-world benchmark.

And you're missing the part where they already tested Ryzen with non-gaming benchmarks (both synthetic & real-world):
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-8.html: testing the 1800X for desktop/office applications, including compressing files with 7-Zip, encoding files with Handbrake, Python programming, etc.
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-9.html: workstation testing with AutoCAD, Creo, Solidworks, etc.
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-10.html: testing for scientific & engineering environments, including Euler3D, LAMMPS, Convolution, etc.
-- http://www.tomshardware.com/reviews/amd-ryzen-7-1700x-review,4987-7.html: testing of the 1700X in some of the same desktop/office benchmarks they tested the 1800X with.

So, again, not sure what your complain is, since Tom's Hardware is obviously testing the Ryzen chips with more than just gaming benchmarks. They just also have articles that focus on gaming because a) a signficant portion of the membership here is interested in gaming, b) the particular testing done in this article (1080p resolution, top-line GPU paired with Intel & Ryzen CPUs) helps spotlight Ryzen's performance in the games in question, & c) even for those not necessarily technically inclined or not a "power" PC user, gaming benchmarks is one of those areas where you can not only get measurable results from your testing but also actively notice the resutls...unlike, say, that Adobe CC Photoshop Light/Heavy results, where even though the Ryzen 1800X & i7-7700K were noticeably faster than the other chips, every CPU still managed to finish it in under 1 or 2 seconds.

I dont have a problem with a lot of the benchmarks. I'm only saying that bottlenecking the GPU doesn't isolate CPU performance for the reasons I expounded on earlier. Point is, if you run a test on a variable that isn't isolated, your test is invalidated.

 

[anubis44]

"two developers hardly account for an entire industry that stands to benefit from Ryzen-specific optimizations, and only time will tell how well the ecosystem adapts to the new Zen architecture. Much of the success will likely hinge on how much market share AMD gains, as developers have far more incentive to optimize for widely-used platforms."

Yes, and guess how many Zen core AMD processors there's going to be with the console refreshes + all the PC sales. There's also already 50 million previous gen AMD 8 core CPUs in the current consoles + all the FX PC CPUs. There's going to be tidal wave of AMD Zen processors flooding into the market, and Intel cannot stop it.

 

[Josh_194]

"two developers hardly account for an entire industry that stands to benefit from Ryzen-specific optimizations, and only time will tell how well the ecosystem adapts to the new Zen architecture. Much of the success will likely hinge on how much market share AMD gains, as developers have far more incentive to optimize for widely-used platforms."

Yes, and guess how many Zen core AMD processors there's going to be with the console refreshes + all the PC sales. There's also already 50 million previous gen AMD 8 core CPUs in the current consoles + all the FX PC CPUs. There's going to be tidal wave of AMD Zen processors flooding into the market, and Intel cannot stop it.

Can't argue any of that. Nice and agreed.

 

[alextheblue]

If it is true that Ryzens are going into the next generation consoles this will ensure that it gets thoroughly optimized for.

Pretty sure people said the same thing about AMD's FX CPUs after the XB1 and PS4 were released with 8-core AMD CPUs inside.

Not exactly. Since both major consoles have lots of cores, I've seen people imply that games that land on PC and consoles (and share code) would use more threads, more effectively. In many cases this did happen, but even in that situation FX was stuck on Piledriver, which continued to hinder it. Anyway I didn't see anyone talking about actual optimization for "FX" or any other Bulldozer-derived architecture (PD, SR, EV). Why? The consoles don't USE those cores. They use the VERY different Cat-family "Jaguar" cores. Optimizing for Jaguar doesn't benefit FX or related dual-module chips.

If an actual Ryzen with "Zen"-based cores ends up in next-gen console(s), the situation is quite different. That scenario would lend itself to actual optimizations for that architecture. If and when this comes to fruition remains to be seen. Sony just released PS4 Pro recently, and they'd be stupid to release PS5 next year. I believe Scorpio's semi-custom AMD APU was likewise too far along in development to be running Zen, but I'd like to be surprised. I'm expecting a higher-clocked Jaguar/Puma.

On a side note I'd like to see an article in the coming months discussing how far board manufacturers have come with memory. It isn't that big of a deal with Ryzen CPUs, but when the APUs land it will be far more important. Well, outside of low-to-mid range laptops where OEMs just shove in some cheap RAM and you're lucky if you get dual channel operation in the first place.

 

[pjgowtham]

"The update is our first example of a game optimized for AMD's Ryzen CPUs, which should improve their performance."

Not true, dota 2 was the first. I hope you guys will consider adding this game to your Ryzen 5 review, it IS the most popular online game in the world.

I dont think dota 2 is a CPU intensive game but would still be good to see who's the winner/loser. Any CPU would do great in that game.
Ashes of the singularity is a CPU intensive title.

 

[alextheblue]

Where are the DX11 numbers for comparison? We all avoid DX12 like the plague because of low FPS.

Speak for yourself. I see better results with DX12.

 

[spdragoo]

I dont have a problem with a lot of the benchmarks. I'm only saying that bottlenecking the GPU doesn't isolate CPU performance for the reasons I expounded on earlier. Point is, if you run a test on a variable that isn't isolated, your test is invalidated.

I'm really trying hard to understand where you're coming from on this. Tom's testing platform reduces (if not eliminates) any potential storage bottleneck by exclusively using SSDs, & used 16GB of DDR4 RAM to reduce any potential RAM bottlenecks, as well making sure all testing systems were running the same OS with the same patches, as well as running the same testing software. Heck, they even used the same PSU in all of the systems to eliminate any potential power fluctuation issues (http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-4.html). Aside from any potential differences with motherboards -- which, if you're going to focus on that, is going to invalidate every benchmark test ever made or that ever will be made, since Intel boards won't run AMD chips & vice versa -- that leaves 2 potential areas for testing: the CPU, & the GPU. Even though the GTX 1080 isn't technically the fastest GPU out there (being beaten by the Titan X & 1080Ti), it's still one of the three 1st-tier GPUs, & more than powerful enough to handle any game currently out there at 1080p resolutions without straining itself...which means that, everything else being equal, the testing Tom's Hardware did show how Ryzen chips perform.

Just because you don't like the results doesn't invalidate them. And I really don't see how you can dispute their testing methodology, because they've set up their testing systems to have at least 90% commonality, with the only differences between the Intel & AMD systems being the motherboards & CPU -- & again, that's because you can't even run FX CPUs on the same motherboard as a Ryzen CPU, let alone a Skylake or Kaby Lake CPU from Intel. That's as close to eliminating extraneous "variables" in testing as you can get.

 

[Diji1]

@BONSCOTT

Beyond 1080p gaming with a GTX 1080 (which is becoming less of an issue), the Ryzen wins on performance, power, and price.

Well no it doesn't win at above 1080p gaming performance either so that is incorrect.

But it is still a good CPU for gaming to be clear but not a good choice for the same money for a person building a gaming only computer.

 

[elbert]

Now can we add 2 more tests to this benchmark? Ryzen with g.skills flarex 3466 and the i7 6900 with a 3466 RAM kit. Given the high price of the broadwell-e its within reason to buy highend RAM for Ryzen.

 

[colemar]

AMD stock price went immediately down 10% in the morning when the very first gaming benchmarks were published.
This time the price went up about 6%, but with a 12 hours delay, WTF?