AMD Ryzen 1700 Emulation Testing...

[Josh128]

Guys its way past my bedtime so Ill make it quick-- just upgraded my old i5 2500K to a Ryzen 1700, 8G of 2667 DDR4, Gigabyte GAMING MB. I want this thread to be a showcase and discussion place, good or not so good or whatever, of the new Ryzen chip for informative purposes. Ill be posting various benchmarks and game tests I have done and am willing to take requests for certain tests as I have time.

I'll start with my longtime favorite CPU test of emulating the Atari "Seattle" 3DFX system in MAME. This particular systems games were unplayable in MAME until the Intel E8400 + Win64 combo came to pass circa 2008. Prior to the Core 2 chips it would bring any system to its knees, my old Athlon 64 3000+ used to run it at about 10 fps or less, roughly 30% emulation speed. The E8400 hovered around 100%, my previous i5 2500 around 205%. This game is multithreaded (2 threads) in MAME.

This first batch of tests shows no video emulation % for a 60 second run and an in game snapshot of emulation %. The 60 second run % is always much higher because it contains little 3D graphics, mostly FMV, but I included it because it is how most MAME games are benchmarked. First set of images is stock speed, second set is OC to 3600. Note that at stock speeds, turbo kicks in which is why CPU-Z shows 3100+ MHz as it constantly bounces around with varying loads and temps. When OC'ed, the chip is locked at 3.6 and does not change. Without further ado...

[bigbadboaz]

So, are you personally happy with the numbers on that first batch of tests?

Very excited about Ryzen. Gaming seems to be a bit of a weakpoint out of the gate, but I think they've delivered on the promise overall. Looking more towards the 5 series myself as one of the 6c/12T chips might be the sweetspot.

I'm just hoping once they get through the launch phase they'll work on official support for Windows 7. Just before launch they told a German magazine they would be offering full support unlike Intel with their line. The statement was quickly retracted but I continue to hope there was a reason for the slip. Legacy support would be another strong selling point as they attempt to regain share in the market..

[Josh128]

I was pleasantly surprised that even with stock clocks, it is outscoring my old i5 2500 which ran at 3.3/3.5 GHz on this
particular test. This test is only "barely" multithreaded, as disabling the multithreaded option in my experience only decreases the percentages scored by ~20% or so.

*EDIT-- The disabling of the "multithreaded" option in MAME 123b indeed causes little effect and the option has been removed from the config file of subsequent MAME builds -- BUT that has nothing to do with individual game drivers themselves-- the Atari Seattle MAME driver is indeed multithreaded and disabling cores or setting affinity to use only 1 thread causes massive performance losses in this title. The "multithread" option in earlier MAME builds only attempted to multithread some basic functions of the emulator itself, not game drivers.

So, for those who like to say MAME is reliant on only single core IPC, thats a bunch of bollocks. MAMEs own documentation says otherwise for many game drivers...

The 1700 also does incredibly well on the CPU-Z benchmarks, topping the reference chart even. I have not had much time to test anything else though, but I will add results as I can and will take requests from anyone interested.

Cheers.

[donluca]

Which version of MAME are you using? I'm pretty sure that multithreading is gone in 0.183 as it caused more issues than anything and gave only a marginal increase in performance.

I want to test my 4.7Ghz overclocked G3258 to see how's the performance on that game compared to the Ryzen.

[Josh128]

*EDIT* Its MAME 0.123b 64bit . I know this particular game used multithreading for 1 core to emulate the R5000 and the other to emulate the 3DFX chip. I remember Aaron Giles (the programmer for this driver) blogging about it and it did indeed help performance on this particular driver when I tried it back in the day. 64 Bit OS also helped quite a bit over the 32bit version.

I never tried to go higher than 3.6, I didnt even have to touch the voltage to reach it. I could probably get 4.0 with some voltage tweaks, but Im all about longevity, I dont want to stress the chip more than necessary. Ill post my 2500k results later for reference.

I did try disabling SMT, and at first it looked like it gave an extra 5% or so, but more testing showed it was probably a wash. I do have the ability to shut off cores in the BIOS, I believe I can also isolate cores to a single CCX (2+0 vs 1+1). I'll also have to look into the multithreading issue with newer MAME versions to see how it compares. A lot of variables really, but I love testing this kind of stuff.

I plan on adding the Dolphin Benchmark as well as in game if possible, and also things like CPU-Z, CrystalMark and perhaps some 3D Mark tests, though Im only using an R9 285 (which I bought used and Im not so sure its not doing some wonky stuff, need to test further).

I bought the chip mainly because my mobo for my 2500 was doing some really wonky stuff and you cant really get those new anymore, so I figured an octocore would be nice! I dont really game much other than some emulation here and there , but I do a good bit of photo and video editing/encoding, so this looked like a killer chip for my purposes. Other than having hell (and giving up) on running it on Windows 7 , ( I got it working, just had driver issues), it seems pretty bad ass so far. More in the coming days....

[Guspaz]

Mark my words, Scorpio is going to be Ryzen, and the decision to wait for Zen instead of just bumping up the clocks a bit is going to be why Scorpio so enormously faster than the PS4 Pro.

[lettuce]

You might have been better of waiting for the 1600 or 1600X as there single core clock speed should be higher

Mark my words, Scorpio is going to be Ryzen, and the decision to wait for Zen instead of just bumping up the clocks a bit is going to be why Scorpio so enormously faster than the PS4 Pro.

I sure hope so.

Phil Spencer says their taking the Scorpio seriously and this will be completely disregarded if the Scorpio isn't Ryzen

[bobrocks95]

Mark my words, Scorpio is going to be Ryzen, and the decision to wait for Zen instead of just bumping up the clocks a bit is going to be why Scorpio so enormously faster than the PS4 Pro.

I hadn't thought about that, but I think you're right. The PS4 Pro is severely limited by its CPU at this point.

[bigbadboaz]

Scorpio = Ryzen

I really don't think so.. timeline is all wrong and the claimed power isn't so much more than PS4 Pro to represent that kind of jump. The way they would have to price such a beast at this point in time would also be prohibitive. It's much more likely to be an even bigger GPU tacked onto the old Jaguar package than the one in the Pro.

I'm as eager to see a Ryzen APU in the consoles as anybody, but expect it/would rather see it in the true next gen, with the more significant jump that implies. Ryzen done right really deserves to be paired with a much better GPU than what they've got now. This is likely to be what we see for PS5, and it really doesn't make much sense for MS to be doing anything significantly different. Give it another two or three years.

[Guspaz]

Why don't the timelines work out? Zen taped out in mid 2015 and sampled in late 2015, while Scorpio is expected to ship in late 2017. That seems like a reasonable timeline to me.

Admittedly, Microsoft has said nothing about CPU performance (only that the GPU performance was roughly 4x), and there do seem to be hints that the console remains CPU bottlenecked.

[Josh128]

Despite fairly powerful graphics hardware, I really believe the CPU is whats holding back 60fps 1080p titles on PS4 pro--
as most everybody knows, 4K stresses the GPU much more than the CPU, whereas 1080p is more CPU bound these days.

A 2.4GHz Jaguar octacore is OK, but I think its the limiting factor in getting 60fps titles at 1080p. The GPU is PS4Pro should be more than enough, yet you still see very few titles that can pull it off (yes, there are a few popping up recently, but still).

Ryzen makes perfect sense-- even at the same clock speeds it is more than double the power of Jaguar for many benchmarks, yet it is an extremely power efficient architecture. AMD could command a decent price for it, much less than what Intel would want, and MS would have a beast of a console CPU. Win/win for both companies. 1080p/60 would be no sweat for this thing. Microsoft would have a commanding hardware lead over PS4 and be able to tout that lead and nobody could deny it.

Im still a bit skeptical though, because never in the history of console releases has a new console CPU been designed on the same process node as its contemporary current gen CPU. They are always at least 1 process node behind due to lower prices, better yields, etc. If Scorpio would launch this year, chances are Intel and AMD would still not have released 10nm yet, so that would be a first.

[Josh128]

My Stock Ryzen 1700 3.0GHz base/ 3.7GHz Turbo CPU-Z Benchmark. Beats a 3.6 Intel 6850K rather handily in both single and multi-core benchmarks here.

[Josh128]

Ran the Dolphin benchmark. Results below.

[bigbadboaz]

Why don't the timelines work out? Zen taped out in mid 2015 and sampled in late 2015, while Scorpio is expected to ship in late 2017.

It's certainly possible but I think it's too quick to be likely.

AMD are still in the launch stage of the main CPU product - remember they still have to release the 3 and 5 lines, only the 7s are out - and there is no word at all yet on APUs even for the PC landscape. For them to have a completed console-specific design just about ready to enter production right now.. just doesn't seem right.

From the Xbox side, two things. One, Spencer has been very clear that their intention is for Scorpio to act just like PS4 Pro - that is, to run all the XO titles but better. For the forseeable future, they do NOT want to lock base XO owners out of the library. A whole new CPU just doesn't make sense either from a cost or design standpoint when you're not trying to make a major break from previous hardware. Two, he's said that pricing will be "premium" but not out of the normal console realm. I expect it to match Pro at $400 or push against $500 at the absolute max. If you look at how new Ryzen is and how hard AMD would have to be pushing to get that APU out RIGHT NOW, it seems using it would push costs way too high.

Lastly, as I touched on before, it just seems like a better fit to wait until Ryzen can be paired with a GPU that is a real leap over what's available now and is a better match with it. It's going to be such a leap over the Jaguar cluster that using it with whatever will be in Scorpio would almost reverse the current bottleneck - that is, too much CPU and too little GPU. I don't think it makes sense yet for an overall design. And by waiting for a more suitable GPU (which coincidentally will allow the current console cycle to play out along a typical timeline) they will also save a lot of $$$ on the Ryzen cluster itself, as the CPU will be well past any launch issues/scarcity, and maybe even on a newer process.

Hey, we'll see what's up at E3. But it just seems that Ryzen now is overkill, but an absolute no-brainer perfect fit for PS5/Xbox Whatever, 2019/20ish.

[gray117]

...from a fabricating point of view it'd be silly not to at least be a derivative of zen. It'd be a confusing state of affairs if Guspaz wasn't right, I'd be shocked if some part of the zen roadmap was not dedicated to (continuing to) deliver the kind of resources required to maintain profitability in such a marketplace such as consoles or similar units.

Microsoft are also wanting to reverse the (fairly insignificant to most users but still evident) mantle of having the better performing system. Whether there's a sensible balance in 'bottlenecking' the alignment of resources is perhaps not as significant if there are other other benefits based around power/heat/yield/construction material cost.

Whether customers will care now we're into the lifespan of these products/services... who knows...

It's nice to see amd compete at this kind of level of performance (+price)...

[Josh128]

Agree with most points on Scorpio, possible but improbable Ryzen implementation. As I said before, it would be the first time in console history to have current state-of-the-art CPU guts in a newly released console. IF it were a Ryzen APU, it could also very well have a VEGA GPU, as that what the upcoming Ryzen APUs will utilize, but Vega still hasnt been released, which makes the scenario all the less likely.

Anyways, more benchies. FYI Im using an R9 285 GPU.

[Josh128]

Timespy... CPU test ran much smoother than most 3D Mark CPU tests do for me.

[tomwhite2004]

Any chance you could run more results from mame (non multi core games) or actual single threaded applications (like Higan) as that is primarily is what most emulators will rely on? Quite interested to see how it stacks up against my humble i3-6320.

"4K stresses the GPU much more than the CPU, whereas 1080p is more CPU bound these days"

Can someone explain to me why that is in laymen terms. I recently saw a digital foundry video where they said they were seeing a cpu bottleneck when using the new gtx 1080 ti on 1080p settings with an i7-7700k. Understand why 4k needs more gpu but don't see how going down to 1080p causes issues.

edit: Just ran the Dolphin benchmark for reference.

[gray117]

Can someone explain to me why that is in laymen terms. I recently saw a digital foundry video where they said they were seeing a cpu bottleneck when using the new gtx 1080 ti on 1080p settings with an i7-7700k. Understand why 4k needs more gpu but don't see how going down to 1080p causes issues.

Well it's a bit of a case by case basis... unless you're dealing in extremes... and emulation is a special case again where for the most part we're only worried about (single thread) cpu performance due to the implementation of most emulators.

The basic idea is that depending on what needs to be computed (most of the time cpu) before render happens (most of the time gpu) can lead to a situation where the gpu can carry out it's tasks faster (best suited to lots of small jobs all at once in parallel) than the cpu can reach the render point in it's list of tasks (lots of jobs, but sequentially organised/prioritised - like a long list - especially if the task is single core only, not multithreaded, i.e. only one long list).

However, exactly how this can be implemented, and what the tasks are, and what can require more processing, can vary massively on both the gpu and cpu side - especially when it comes to keeping all of this in sync. Generally this is why the difficulty/compatibility of parallel or multi-threaded support is purposefully somewhat ignored by most emulators whose main objective and initial challenge is just in modelling the hardware being emulated, and why commonly you'll see high demands on a single core cpu process to emulate the hardware, run game, and most times render the frame* (*I think this happens in mame for example - before sending that frame to gpu for scaling/display/shading via d3d). Other emulators might offload a (portion of) the frame rendering (thread) to a compatible d3d/opengl gpu, but you'll often find a software (cpu) driver option too.


4k vs 1080p ... it's a case of relative challenges. Again take it with a bit of salt depending on the case in question. In 4k broadly tasks remain the same for cpu, but the challenge here is most of the time the simple jump in resolution is currently a massive rendering overhead - although you're only doubling the h+v resolution it's effectively 4 1080 screens in one. Now, all things are not equal. Tasks vary in length and there are cheats/savings/costs and unavoidable overheads that may vary (or not) depending on resolution. There are hard limits too - those 4k frame buffer(s) (many layers in most contemporary 3d games) all eat into the amount of v-ram that would otherwise be available for models/textures/shadow maps etc - what once fit in memory with a 1080p frame buffer may not with a 4k buffer in there.

[tomwhite2004]

Yeah, I get why emulators generally tend to be single core / threaded (thanks for taking the time anyway) its why I bought an i3-6320, outside of the various i7's it offered the best benchmark scores for the money until the new Kaby processors came along. But.....

4k vs 1080p ... it's a case of relative challenges. Again take it with a bit of salt depending on the case in question. In 4k broadly tasks remain the same for cpu, but the challenge here is most of the time the simple jump in resolution is currently a massive rendering overhead - although you're only doubling the h+v resolution it's effectively 4 1080 screens in one. Now, all things are not equal. Tasks vary in length and there are cheats/savings/costs and unavoidable overheads that may vary (or not) depending on resolution. There are hard limits too - those 4k frame buffer(s) (many layers in most contemporary 3d games) all eat into the amount of v-ram that would otherwise be available for models/textures/shadow maps etc - what once fit in memory with a 1080p frame buffer may not with a 4k buffer in there.

See I understand that 4k is a huge pixel bump from 1080p and all the extra requirements that puts on vram etc. But what is being said is that a 1080p resolution creates a cpu bottleneck with the same gpu. I could understand this bottleneck with 4k content due to all the extra demands but not the other way round.

To quote the digital foundry article when running 1080p games on a gtx 1080 ti.
"In short, many of our test titles lose a big chunk of their performance very suddenly, at any given point. Typically, the more the CPU has to calculate - for example, intense physics in Crysis 3 or a detail-packed vista in Far Cry Primal - the more we see the processor becoming the limiting factor, resulting in a sudden lurch downwards with in-game frame-rates."

Why is this especially the case for 1080p content and not mentioned when running at 4k, how can the cpu become the limiting factor when running the same content at a lower resolution?

But sorry to side-track. To get on topic I think Intel are still the way to go for a dedicated emulation build. Take this Ryzen 1700, at least in the UK its the same price as an i7-7700k and that would stomp all over it for the same cost.

I read one of the PCSX2 devs say that a score of 2000 was needed on the below chart to run most games comfortably, and AMD are still quite a ways off it still.
https://www.cpubenchmark.net/singleThread.html

[bigbadboaz]

...from a fabricating point of view it'd be silly not to at least be a derivative of zen.

Maybe in a more typical console era, but today we have PS4 Pro just released and going to be in-production for at least a few more years. AMD are going to be fabbing super-Jaguar APUs for quite a bit of time. This type of hardware would certainly have been offered to Microsoft in their planning phase, and will be easily more cost-effective.

Anyway, it's an interesting discussion as long as the system is cloaked. Not much more than a month before we know it all.

[Guspaz]

Why is this especially the case for 1080p content and not mentioned when running at 4k, how can the cpu become the limiting factor when running the same content at a lower resolution?

Because the CPU power required is constant while the GPU power is reduced. Imagine this hypothetical and oversimplified scenario:

4K
CPU: 4 performance units
GPU: 8 performance units
Result: GPU bottleneck

1080p
CPU: 4 performance units
GPU: 2 performance units
Result: CPU bottleneck

[Josh128]

Why is this especially the case for 1080p content and not mentioned when running at 4k, how can the cpu become the limiting factor when running the same content at a lower resolution?

But sorry to side-track. To get on topic I think Intel are still the way to go for a dedicated emulation build. Take this Ryzen 1700, at least in the UK its the same price as an i7-7700k and that would stomp all over it for the same cost.

I read one of the PCSX2 devs say that a score of 2000 was needed on the below chart to run most games comfortably, and AMD are still quite a ways off it still.
https://www.cpubenchmark.net/singleThread.html

Ryzen 1700 and i7-7700k are not the same price when you consider that the Intel part does not come with a cooler and fan, which a decent one will set you back about $30. Unless you are upgrading from a recent gen Intel part (why even bother?) you will most definitely be buying a cooler, so go ahead and tack $25-35 bucks to the price of the 7700.

The 7700k is definitely faster for most emulation, but for all the emulation I have done, an i5 2500k runs every decently optimized MAME driver and most other emulators I have thrown at it pretty well. A R7 1700 is faster than the i5 2500k in single threaded stock form, and if you OC it to 3.6 or more you have a good bit of headroom over that. In fact, you'd pretty much have to tell me what games you NEED a 7700k to run at full speed that a 3.6 Ryzen wont. Programs and games are moving quickly in the multi-threaded direction and that's not going to stop anytime soon. For general PC use, both are fine, but the Ryzen is much more powerful for video editing / encoding / transcoding and other thread heavy tasks, bone stock. For right now though, if the sole intent is the fastest % in emulators, the 7700k is def what you want.

Most applications/ games/ emulators have been optimized for Intel arch over the past 8+ years or so due to AMD chips being so far behind. Optimizations for Ryzen arch will indeed make a difference as the AOTS Ryzen patch in the news today shows about a 30% gain in fps on average. This is a clear indicator that AMD is not blowing smoke when they cite optimizations are needed for the new arch. This follows for many apps, including emulators. A Ryzen 1700 is definitely a more future proof choice overall than a quad core i7 7700k because of this and the additional 4 core advantage.

I'll try some more MAME benches if anyone has a list of games they are interested in seeing.

http://www.guru3d.com/news-story/aots-g ... marks.html

[Udderdude]

Ryzen 1700 and 7700K are not the same price when you consider that the Intel part does not come with a cooler and fan, which a decent one will set you back about $30. Unless you are upgrading from a recent gen Intel part (why even bother?) you will most definitely be buying a cooler, so go ahead and tack $25-35 bucks to the price of the 7700K.

You also need to pay more for an overclocking-capable motherboard if you go with Intel. Whereas AMD has no such restrictions.

In terms of raw performance, 7700K has 30% better single core, but the 1700 has 60% better multi-core.

You can bet that tradeoff isn't going to be in favor of the 7700K for very long.

http://cpu.userbenchmark.com/Compare/In ... 3647vs3915

[Ed Oscuro]

There is the cheaper B370 range of motherboards without overclocking (automatic or otherwise). As a whole the range of AMD motherboards seems to be consistently cheaper than Intel.

It is worth noting that Ryzen seems to have fewer PCI-E lanes than Intel, which might matter in some configurations (obviously multi-card, which I'm not interested in, but also maybe if you're running new PCI drives and so on). I'd like to wait for Zen Refresh to see if single-threaded perf can go up, but I suspect Intel stays ahead here.

For what it's worth my old Sandy Bridge-E seems to give good performance in any Dolphin title, but I'd like to get max single threaded performance for PCSX2 and any future emulators (it seems that this would be good for RPCS3 or whatever it's called, too). I would really like not to encourage Intel's TIM practices by buying a 7700k.

[tomwhite2004]

Because the CPU power required is constant while the GPU power is reduced. Imagine this hypothetical and oversimplified scenario

That makes sense, thanks. I always assumed that even though a gpu was running a lower resolution it would still have access to or be able to use those extra performance units.

A R7 1700 is faster than the i5 2500k in single threaded stock form, and if you OC it to 3.6 or more you have a good bit of headroom over that. In fact, you'd pretty much have to tell me what games you NEED a 7700k to run at full speed that a 3.6 Ryzen wont. Programs and games are moving quickly in the multi-threaded direction and that's not going to stop anytime soon. For general PC use, both are fine, but the Ryzen is much more powerful for video editing / encoding / transcoding and other thread heavy tasks, bone stock. For right now though, if the sole intent is the fastest % in emulators, the 7700k is def what you want.

An i5-2500k is what 7 years old now? There have been huge increases in performance since then so I don't think its the best comparison to make anymore as no one is going to use them with a new emulation build.

As far as emulation goes I would say you need an 7700k to run Cemu at near full speed (breath of the wild seems very taxing), especially as the emulator is still in such an un-optimised state all round, would be really interested in how Ryzen compares as that is one of the few emulators that can take advantage of multiple cores. Same with RPCS3 but I don't have any experience with that one.

Would also be interested to see what results you got with say Mushihimesama Futari in Mame with an unlocked framerate on both of your processors as the CV1k driver is more taxing than most stuff on there.

Please don't think I'm shitting on the Ryzen range, no doubt multicore is the future for gaming and agree that they seem like a great cpu for the examples you gave like encoding etc. Will be very interesting to see how Intel responds now that they actually have some competition.......

[Udderdude]

it seems that this would be good for RPCS3 or whatever it's called, too

Ryzen is going to shrek any Intel HT-enabled quad core at PS3 emulation. Cell has 7 cores on it's own, not counting the controller.

[tomwhite2004]

Ryzen is going to shrek any Intel HT-enabled quad core at PS3 emulation. Cell has 7 cores on it's own, not counting the controller.

However the Intel quad cores will be a thing of the past by the time PS3 emulation has comparative compatibility and stability to say Dolphin, think that is many years away despite recent good progress.

[Udderdude]

We can only hope. I'm definitely going Ryzen/Vega for my next build in 2018. Given enough money to team blue and team green. It seems pointless to stay on 4 core/8 thread forever.

[bigbadboaz]

Curiosity, why are you already locked into Vega? nVidia has been pretty handily ahead of AMD on the GPU side for.. seems like forever now, and until we see actual Vega results there isn't much to indicate this is going to change. Recent case in point, the Polaris GPUs are good, but probably not as good as hoped, and certainly not a better architecture than the competing nVidia parts.