hardware encoding uses GPU, 'software' encoding means no hardware acceleration, meaning all the work is done by the CPU.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
AMD Confirms Zen 3's Performance Is Monstrous and Speculation Thread
- Thread starter erek
- Start date
And? I'm getting the message, you don't have anything else to back up your claims. So far we have four reviews backing up my numbers, a single bench showing 3700x and up being tied with 8086k/8700k and that's it.
Come on everyone knows that people don't have high end PCs to do production work, they have them to play games which by games I mean one specific emulator of course. /s
erek
[H]F Junkie
- Joined
- Dec 19, 2005
- Messages
- 10,875
erek
[H]F Junkie
- Joined
- Dec 19, 2005
- Messages
- 10,875
I linked benchmarks that backed up that Intel is faster in games and blows away AMD in emulators, and of course the response is "BUT ISN'T THERE MORE?"
What more do you want? It's already been objectively proven.
Compilation flags that introduce buggy behavior which is why they're not included in release builds.
It isn't just one emulator. Intel smokes AMD in RPCS3. Look, I understand that you're fact denying fanboys, but you can't just write this off. Intel is better for gaming.
You linked benches that show what I said, up to 10% faster in every games except emulators. Nobody is refuting that intel is faster in emulation, and if thats your use case, buy intel.
You didn't link anything that shows 20-30% in games as well as emulation, which is what you claimed, and claimed it more than once.
You also linked to a Reddit thread about building a system for emulation, that said pretty much nothing.
I'm asking you for sources on your claim that intel is 20-30% faster in games. I concede that in dolphin vs a 10900 Zen 2 is down by 20%. Can you show me the rest of the games you're claiming?
That isn't what I said. I said "Gaming is pretty much the only reason people care about high end PCs." Get it right next time if you're capable of that (which I doubt).
What I said is true. The high end CPU market is driven almost entirely by gaming. Workstations tend to be higher core count/lower clock speed machines.
So I'm glad we've confirmed that I'm right and you're wrong. Thanks!
Intel is better in games, it's better in video game emulators, and given the low core count in the "next gen" consoles, Intel already has enough cores for that generation of ports. The best case scenario for AMD is to catch up, which, given Intel's utter failure over the past few years, is utterly pathetic.
After seeing the obnoxious, bad faith posts with "/s" in them and the obvious fanboyism, it's not worth my time to do a simple Google search for you. There are dozens of games where Intel has 30%+ performance advantages. Frankly, I don't care whether you accept the truth or not. Doesn't matter to me.
NightReaver
2[H]4U
- Joined
- Apr 20, 2017
- Messages
- 3,791
Lol, it feels like a troll post. Some barely noticeable differences in a handful of game benchmarks (@ 1080p) and emulating (which is very niche).
Followed up with "I'm right, you're wrong" and "deal with it". Good stuff
.Um I'm not denying Intel is better for gaming. However, not knowing what you're talking about and saying idiotic things like compilation flags introduce buggy behavior which you haven't even tested yourself is ridiculous.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
I actually don't think this is true. I haven't seen sales/use analyses recently, but at least a few years ago, say 5 years ago, most upper-tier CPU sales were to businesses. Now there are usually top-tier CPUs that are marketed toward gamers and not always offered by OEMs, and sure, those SKUs are going to favor gamers, but high-end workstations get tons of volume.
Also I don't think it's fair to compare Zen 2 to Zen 3 performance, especially in something so oddly specific like emulation. We don't know how it will do across the board but in general, it sounds like Zen 3 will best current-gen Intel counterparts, or at least, trade blows. Core count isn't where it's at for the most part, even today with the upcoming next-gen consoles. Cores are small and cheap and add a degree of future-proofing. Gaming, at least for now, is single-thread performance-bottlenecked.
Not that you really need high end components for emulation, unless you're doing weird upscaling/modding/tweaking. That's not mainstream gaming, either.
They're very obviously trolling at this point, each post gets more and more ridiculous and outlandish.Lol, it feels like a troll post. Some barely noticeable differences in a handful of game benchmarks (@ 1080p) and emulating (which is very niche).
Followed up with "I'm right, you're wrong" and "deal with it". Good stuff
Red Falcon
[H]ard DCOTM December 2023
- Joined
- May 7, 2007
- Messages
- 12,445
You actually need a really beefy computer for RPCS3 and Dolphin with LLE and so on, and like you said, the graphics card requirements even get pretty nuts if you run at a really high internal rendering resolution. A Geforce 1080 isn't even enough for Dolphin with 4k internal rendering.
RPGWiZaRD
[H]ard|Gawd
- Joined
- Jan 24, 2009
- Messages
- 1,217
Emulators are good benchmarks in the sense in most emulators you can totally remove GPU performance impact in a way more than say PC 1080p gaming performance or 720p even. I did a CPU overclocking impact review a long time ago on PCSX2 which were at the times AMD was heavily beating Intel (AMD64 X2 / Opteron vs Pentium 4) and discovered the amount performance boost you get pretty much were in-line with the overclocked CPU amount, it's a difficult scenario to find such a benchmark that involves games outside emulators. Nowadays there's typically internal rendering resolution scaling built in and various graphical effects which can increase the GPU load so much the GPU again becomes the bottleneck.
Anyway the potential bad thing about comparing Intel vs AMD in emulators again is probably the compilation flags and various instructions can provide a big difference, sometimes it can be the case one or the other might get a benefit more so than the other part. Also they are typically not utilizing multiple cores very well, typically one or two cores maybe, seldom more than that mostly due the added complexity so they represent better the IPC differences.
In PC gaming at 1080p for example the 10900K seems to have roughly 10% lead over 3900X. I think this upcoming AMD launch we might potentially turn that advantage around in favor for AMD even if leaks are accurate.
Anyway the potential bad thing about comparing Intel vs AMD in emulators again is probably the compilation flags and various instructions can provide a big difference, sometimes it can be the case one or the other might get a benefit more so than the other part. Also they are typically not utilizing multiple cores very well, typically one or two cores maybe, seldom more than that mostly due the added complexity so they represent better the IPC differences.
In PC gaming at 1080p for example the 10900K seems to have roughly 10% lead over 3900X. I think this upcoming AMD launch we might potentially turn that advantage around in favor for AMD even if leaks are accurate.
NightReaver
2[H]4U
- Joined
- Apr 20, 2017
- Messages
- 3,791
If only I was elite enough to play 1080p on a $500+ cpu.
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
RPCS3 is definitely one of those target things to get working when I finally upgrade to my Zen 3 8-core.
But even today on my 4790k, the current builds are so buggy that I have no problem waiting another year for that kind of performance update.
Like PCXS2 was around 2005 we're still another five-ten years away from most games playing successfully/various NECESSARY performance tweaks being MOSTLY bug-free. The ten years of GPU upgrade will also help with increasing the rendering resolution.
But even today on my 4790k, the current builds are so buggy that I have no problem waiting another year for that kind of performance update.
Like PCXS2 was around 2005 we're still another five-ten years away from most games playing successfully/various NECESSARY performance tweaks being MOSTLY bug-free. The ten years of GPU upgrade will also help with increasing the rendering resolution.
We are leaving the era of single thread performance and have been for some time even for gaming. Multithreading is the only thing that is allowing performance to increase moving forward.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
Yeah, I mean, people have been saying that for more than a dozen years now. I'm still waiting for it.
Actually, I think with NVME and caching we're going to see bandwidth play a bigger and bigger role going forward.
I don't know if it's my new SSD or if MSFT fixed FS2020, but the load times are super fast compared to before.
We've been past single core performance for while now. As an example a single core vs a 4 core at a lower clock speed. In 99.9% of the games we have out today would do better on the quad over the single or even a 2 core over a single will be faster.
What people do debate is the optimum core count. But are we in the era of multithreading? Yes. 8 is the current limit. Less than that and the FPS will go down.
Last edited:
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
Since dual-core was introduced, technically we've been past single-threaded performance. But multi-threaded gaming performance doesn't benefit nearly as much over those initial gains compared to real multi-threaded applications. The first two-core CPUs let you offload background tasks and OS stuff, which upped performance by about 10 percent. It wasn't until Windows 7 that multi-core support really started to benefit some games, and I'm sure there are specific games that are built to scale better, but you're still mosly looking at one master core passing off minor operations to other cores.
If multiple cores made such a huge difference, we wouldn't have had to wait for Zen 3 for AMD to shine against Intel.
So now we're gonna change the goal posts? Is dual quicker then single, Yes or no? Yes it is. Is quad quicker than dual? Yes it is. Having a discussion on benefit of gain is a completely different topic.
Who is really waiting for Zen 3 to outperform Intel? Most of us like hardware and for those of us who know the actual deficiency in the architecture are just waiting to see if AMD fixes it. Basically it's good PC Hardware drama. But right now AMD is killing Intel specifically in the area where you believe it isn't necessary and that's SMT workloads. Why? Because schedulers are more multi-core aware than ever before and that's because that's where the gains can be made. Gaming consoles went SMT years before PC ever did. Why? Because that's where the gains were.
And now with console architecture parity with PCs multi-threaded games are going to be more and more common, and it’s honestly about time.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
No, it's not, it's that you can offload the OS and other functions onto other cores. The core that's handling the main game is still just as fast as it ever was. The gains don't stack, it's not like 2 cores = 110 percent, 4 cores = 130 percent and so on.
I hope so too but we've also had multi-core consoles for several generations. Xbox One and PS4 were both introduced in 2013, and they weren't the first.
Aside from the fact that gaming engines became multithreaded years ago. I'm just gonna highlight what you said and leave it there.
If you think singly thread is king try disabling all by one core and running Windows that way for a while.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
I never said that. That's like saying if single-thread performance is irrelevant we should cap our CPUs at 1Ghz on all cores.
My last stab at this and then I'm done.
Let's look at it from some equations. This is totally simplified but it should work for this:
(1971-2005) Single Core IPC * Clockspeed = Performance
(2005 - NOW) Single Core IPC * # of cores * Clockspeed = Performance
Now Unreal Engine gained multi-threaded support since 2009. This means that ALL games built on Unreal Engine 3 have the possibility (<--- notice this definition) to issue threads (commands) simultaneously on more than 1 core because from schematic to rendering was broken down to work on more than one thread.
So this means that the Unreal Engine since 2009 is going to use that latter equation. So using that latter equation if you adjust IPC upwards then of course it's going to affect performance. However, if we adjust the # of cores that can affect performance too (in perfect scaling, everything is multithreaded land). Do you understand?
So when you say "IPC / Single Core Performance matters more" you need to look at that latter equation and realize that we have more than 2 values we need to plug in.
Since 2005 when we talk about SMT workloads we now discuss the optimum # of cores to achieve near perfect scaling (which varies) so this is where we are today. Just because this latter part varies does not at ANY TIME allow us to say single core performance matters more (unless there was some magical CPU designer person who increased IPC 4 fold).
Intel succeeds in gaming because they have enough cores (8) and their cache performance is better which affects IPC. Gaming is very IPC and cache latency sensitive. So AMD is trying to improve cache latency (Zen 1 ->2 ->3) in order to increase performance on applications/games that are latency sensitive. If Intel's highest core count was 4, it would get obliterated in most if not all gaming benchmarks. Why? See equation above.
Last edited:
RanceJustice
Supreme [H]ardness
- Joined
- Jun 9, 2003
- Messages
- 6,623
As someone who is an "early access" user (I contribute a couple bucks to the open source projects and in return get new stuff to test a week in advance more or less) to many of the most demanding emulators - Yuzu (one of the two "main" Switch emulators), Citra (3DS Emulator), Cemu (WiiU ; not initially open source, will be opened later like Dolphin was), RPCS3 (PS3), etc - there may be some advantages to the highest end, highest (over)clocked turbo Intel chips in some cases, but especially as the emulators mature this is usually not such a big deal.
Things can change very quickly in high end emulator development (ie Literally a day or two ago, Yuzu went from black-screening the visuals on in-game cutscenes to playing them flawlessly on most titles thanks to the implementation of NVDEC codec emulation suppor, for instance). Things that used to not work very well on AMD, or Nvidia, or OpenGL, or Vulkan, or Linux, or Windows etc... can be fixed and that element comes to parity or better down the line. Thus, even for emulators I don't think there's often such a huge , persistent gulf between Intel and AMD that its worth buying one or the other specifically for that reason. In many cases even when there is a difference, its often explained in clock rate + IPC raw performance differences and if Zen3 is another step forward for both of those, its likely to have a big differentiating factor and may be equal or better than Intel's offerings in the current/next generation. One would hope the days of specific Intel flags in a compiler, other software or whatnot is rapidly diminishing and any performance difference will be "real" differences regardless of platform.
Even in the most demanding emulation cases, when it comes to playability AMD and Intel are often close to each other outside of a few edge cases (ie even the RPCS3 element which showed a 20-30% difference in certain titles was minimized with a later update to something closer to 10-15% last I checked). Especially with Zen3 on the horizon, I wouldn't move to buy Intel even for demanding emulation usage in the vast majority of cases. Outside of those, AMD is often equal or better throughout other gaming usage, not to mention other types of software. This of course separate from ethical or philosophical reasons to prefer AMD. I really am enjoying emulation quite a bit these days. During the Zen and Zen+ days, I definitely would have conceded that the cases for gaming in this regard favored Intel by if not a significant but at least a noteworthy margin, whereas against Zen2 things are much closer; I expect even with very modest performance gains Zen3 will close the gap or even eclipse Intel in all save for a handful of cases. It need not be superior in every single one to be overall a better choice, even for those who enjoy emulation.
Things can change very quickly in high end emulator development (ie Literally a day or two ago, Yuzu went from black-screening the visuals on in-game cutscenes to playing them flawlessly on most titles thanks to the implementation of NVDEC codec emulation suppor, for instance). Things that used to not work very well on AMD, or Nvidia, or OpenGL, or Vulkan, or Linux, or Windows etc... can be fixed and that element comes to parity or better down the line. Thus, even for emulators I don't think there's often such a huge , persistent gulf between Intel and AMD that its worth buying one or the other specifically for that reason. In many cases even when there is a difference, its often explained in clock rate + IPC raw performance differences and if Zen3 is another step forward for both of those, its likely to have a big differentiating factor and may be equal or better than Intel's offerings in the current/next generation. One would hope the days of specific Intel flags in a compiler, other software or whatnot is rapidly diminishing and any performance difference will be "real" differences regardless of platform.
Even in the most demanding emulation cases, when it comes to playability AMD and Intel are often close to each other outside of a few edge cases (ie even the RPCS3 element which showed a 20-30% difference in certain titles was minimized with a later update to something closer to 10-15% last I checked). Especially with Zen3 on the horizon, I wouldn't move to buy Intel even for demanding emulation usage in the vast majority of cases. Outside of those, AMD is often equal or better throughout other gaming usage, not to mention other types of software. This of course separate from ethical or philosophical reasons to prefer AMD. I really am enjoying emulation quite a bit these days. During the Zen and Zen+ days, I definitely would have conceded that the cases for gaming in this regard favored Intel by if not a significant but at least a noteworthy margin, whereas against Zen2 things are much closer; I expect even with very modest performance gains Zen3 will close the gap or even eclipse Intel in all save for a handful of cases. It need not be superior in every single one to be overall a better choice, even for those who enjoy emulation.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
That is not at all how core count scales performance, even in a simple example. It doesn't just overflow into the next logical processor, a single game can use an entire core by itself and still be CPU limited, even if all the other processes are offloaded onto other cores, with other cores not under load.
Let's say you have 5 cores running at 10GHz and an application that runs three main sets of threads. You've got a main game thread that requires a whole core at 10GHz to run smoothly. Let's call them 10 compute units. The other two sets of threads only require 5 compute units each, and the OS requires 2 compute units, and you also have 3 other programs in the background using 1 compute each for 3.
With that CPU, one core is fully saturated at 10 with the main core, and the four other cores can split the remaining processes between themselves without maxing out.
Now let's assume the game needs 12 compute units in its main set of threads, and the other two sets of threads require 6 computes each. Because a single core can only produce 10 computes at full tilt, even though none of the other processes are able to saturate the other cores, the main thread is still bottlenecked.
You can see this in many current CPU reviews. Gaming in general is CPU-bound by a single-core, even if other cores are tasked and under-utilized.
Here's an example that compares CPUs with similar clock speeds and similar architectures with different core counts: https://www.tweaktown.com/reviews/9529/amd-ryzen-3600xt-3800xt-and-3900xt-zen-2/index.html You can see that there are plenty of tests where they are within a few percentage points of each other because they're constrained by a single core's performance, not the multiple cores available.
In well-multi-threaded applications, the extra cores make a huge difference. But in many other cases, the processing power of a single core determines the overall performance for the test.
Nenu
[H]ardened
- Joined
- Apr 28, 2007
- Messages
- 20,315
As above, this is terribly limited thinking.
Care to expand on it?
ManofGod
[H]F Junkie
- Joined
- Oct 4, 2007
- Messages
- 12,863
Yet, he stated IPC first before multicore or clockspeed and therefore, not sure where it is limiting nor constrained, at least based upon the formula he is using.
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
PEMDAS?
ManofGod
[H]F Junkie
- Joined
- Oct 4, 2007
- Messages
- 12,863
Axman
VP of Extreme Liberty
- Joined
- Jul 13, 2005
- Messages
- 17,307
Quick, lets derail this thread we no one is looking!
No, those are the basic rules for the math. If there's some other kind of arithmetic I'm happy to hear it; I never understood reverse Polish notation.
PEMDAS = IPC * Cores * Clocks
Doesn't really matter what comes first.
Nenu
[H]ardened
- Joined
- Apr 28, 2007
- Messages
- 20,315
The formula he gave stands as stated.
Number of cores/threads does not have a direct correlation with performance beyond 4 to 8 cores, less with many games.
It cannot be used as a blanket rule of thumb.