How is Intel Beating AMD Zen 3 Ryzen in Gaming?

[Cobain]

you seem to be new to the game.

You are the one that is referencing people that can't setup their machines. -50% less performance. like in what world. At first you talk about stock, then you talk about overclock. You are all over the place.
TPU average is only 3% difference. Most review outlets have slightly higher 5950x, 5900x averages.
It only comes down to price (with motherboard included).

so far you haven't linked anything that back up your claims.

jfyi no review outlet is reviewing at non-stock bc overclocks are not guaranteed.

I quoted a max tuned AMD vs Intel comparasion. It was considered "inválid" because GamersNexus review with Intel at 3600mhz only (wich is wayyyyy low value compared to the speeds Intel can get) Said otherwise

Talk about being Over the place. Só lets recap, AMD vs Intel tests are only valid when you use all favourable AMD tuning even if it means spending 300€ on RAM, while on Intel you tune it Mildly

Check the vídeo I posted before

 

[NesteaZen]

ok let me help you. all with testing methodology outlined.

AMD Ryzen 9 5950X Review

Here's our first look at Ryzen 5000, starting with a full review of the Ryzen 9 5950X flagship. This new chip packs 16 cores with SMT support...

www.techspot.com

Ryzen 9 5950X and 5900X Review: AMD Unleashes Zen 3 Against Intel's Last Performance Bastions

AMD continues its onslaught on what was once Intel's undisputed turf.

www.extremetech.com

AMD Ryzen 9 5950X and 5900X Review: Zen 3 Breaks the 5 GHz Barrier

AMD's Magnum Opus

www.tomshardware.com

AMD Ryzen 9 5900X and 5950X review

We review ZEN3, the new Ryzen 9 5900X, and 5950X. Released by AMD as a new architecture that will once again attack Intel, this round with a heavy focus on your gaming performance. Overall, this pro...

www.guru3d.com

AMD Ryzen 9 5950X and 5900X Review: Zen 3 Breaks the 5 GHz Barrier

AMD's Magnum Opus

www.tomshardware.com

AMD Ryzen 5 5600X, Ryzen 7 5800X, Ryzen 9 5900X en 5950X

AMD doet vandaag zijn grootste introductie van dit jaar: de Ryzen 5000-processors voor desktops, op basis van de gloednieuwe Zen 3-architectuur.

tweakers.net

@Cobain

There is nothing wrong. I followed his thread on another fórum where Many users also repórted and shown that tuned Intel beats tuned Ryzen 5000 in games, BE it 240hz scenario low settings or 4k. It is not wrong.

link me/us

SIMRACE pl contradicted himself with his zen2 vs coffee lake comment.

----------------------
the spiritual successor to pclab.pl

Test procesora AMD Ryzen 9 5950X - Niezwykle szybkie 16 rdzeni | PurePC.pl

Test procesora AMD Ryzen 9 5950X - Niezwykle szybkie 16 rdzeni (strona 43) Test wydajności procesora AMD Ryzen 9 5950X (Zen 3 Vermeer) w grach komputerowych i programach. Który procesor jest szybszy?

www.purepc.pl

I don't see anything to get wet over with intel. It really comes down to price.
AMD has €80-150 motherboards that can handle the 5950x (mostly not OC), intel not really.

--------------------------
since you're so RAM focused. same RAM for both systems
ttps://www.sweclockers.com/test/30638-amd-ryzen-9-5950x-och-ryzen-9-5900x-vermeer/5

AMD Ryzen 9 5950X och Ryzen 9 5900X "Vermeer" - Test - Speltest: Battlefield V

Ryzen 5000 och arkitekturen Zen 3 är här, med stora löften om dominerande prestanda på processormarknaden.

www.sweclockers.com

 

[Dredi]

I quoted a max tuned AMD vs Intel comparasion. It was considered "inválid" because GamersNexus review with Intel at 3600mhz only (wich is wayyyyy low value compared to the speeds Intel can get) Said otherwise

Talk about being Over the place. Só lets recap, AMD vs Intel tests are only valid when you use all favourable AMD tuning even if it means spending 300€ on RAM, while on Intel you tune it Mildly

Check the vídeo I posted before

What? Gamers Nexus has both AMD and intel running with 3200 MHz dual rank RAM. Where did you get the idea of them using different RAM settings for different platforms?

The only way intel outperforms AMD is if you overclock it to absurd (=unstable) levels and use a lot of time and money to get well performing ram for it. For AMD 3200 MHz dual rank is all you need.

If you'd like to see stock results, Anand has a great article about it, with 13 games tested: https://www.anandtech.com/show/1621...ive-review-5950x-5900x-5800x-and-5700x-tested
The performance difference is especially absurd when you take GPU bottlenecks out of the picture.

As for equally tuned systems, as in the same amount of time used to tune both an intel and AMD system of the same price point, the results are not good for intel either. Just look at the following video:

 

[blu3dragon]

@W1zzard Coming back to these results in prep for the 12th gen reviews...
Any chance you could re-generate this chart and add an intel 11th & 12th gen cpu, maybe as a future article?
I think it will be interesting to see where things stand now.
Don't need all the other charts, and it doesn't specifically need to be with a 2080Ti.

 

[phanbuey]

As for equally tuned systems, as in the same amount of time used to tune both an intel and AMD system of the same price point, the results are not good for intel either. Just look at the following video:

Those results are great for intel - that's an 14nm 8700K OC matching a stock 7nm 5600x with a 3090. Not bad at all.

Also that is the current GPU limit cap - basically if you're GPU limited at all, there is very little difference between tuned Intel and tuned AMD with a 3090, lesser video cards even less.

Anyone running an OC on 8700K (4 year old chip), 9900K, or 10 series isn't going to be pressed to upgrade until RDNA 3 and even then only if they're going for a 7800XT+.

 

[seth1911]

AMD wouldnt release it to Consumer but the real King for IGP (1% slower than 5600G IGP) and CPU overclock would be the 5300G

In the seight from AMD, why should we release a Consumer 5300G for 150 $ if that stupid Users like dumb monkeys buy a 5600G 280$ anywas?

 

[RandallFlagg]

Those results are great for intel - that's an 14nm 8700K OC matching a stock 7nm 5600x with a 3090. Not bad at all.

Also that is the current GPU limit cap - basically if you're GPU limited at all, there is very little difference between tuned Intel and tuned AMD with a 3090, lesser video cards even less.

Anyone running an OC on 8700K (4 year old chip), 9900K, or 10 series isn't going to be pressed to upgrade until RDNA 3 and even then only if they're going for a 7800XT+.

I think gen 8 and anything short of a 9900K is becoming limiting on higher end new GPUs. It will be interesting to see how new GPUs late next year affect the last of the Skylake derivatives.

You need a $1400 GPU to show them up right now, but in normal / past times that would be a $650 GPU, which is really why they're still viable. Zen 2 (3000 series) and below are ofc in worse shape on that front.

I think the next gen GPUs late next year will push anything less than Zen 3 and RKL to the bottom of the charts.

The SOTR benchmark thread here is a really good place to find information on how CPU limited you may be.

And with a 6800XT, the 3800X is CPU limited to 161 FPS :

And a 10850K, CPU limited to 181 FPS with a 3080 :

 

[phanbuey]

I think gen 8 and anything short of a 9900K is becoming limiting on higher end new GPUs. It will be interesting to see how new GPUs late next year affect the last of the Skylake derivatives.

You need a $1400 GPU to show them up right now, but in normal / past times that would be a $650 GPU, which is really why they're still viable. Zen 2 (3000 series) and below are ofc in worse shape on that front.

I think the next gen GPUs late next year will push anything less than Zen 3 and RKL to the bottom of the charts.

The SOTR benchmark thread here is a really good place to find information on how CPU limited you may be.

And with a 6800XT, the 3800X is CPU limited to 161 FPS :
View attachment 223437
And a 10850K, CPU limited to 181 FPS with a 3080 :

View attachment 223439

I am pretty familiar with the SOTR bench, couple things:
1) that bench is poorly optimized for processors on the DEMO build - and the newer builds take much better advantage of processor performance:
2) untuned (i.e. not memory overclocked Zen 3) score almost the same despite being 25%-30% faster than Skylakes on this build
3) 6800XT will absolutely smoke a 3080 in average fps at 1080P due to a bit of nvidia driver overhead, which will impact CPU score a bit (zen 3 does better here)
here is my 10850K w/ 3080 at 1080P Lowest on the real build of the game:

Am I bottlenecking? Yes - I am bottlenecking anywhere to 215FPS (highest) to 238FPS (lowest) average at 1080P for this game...

the bottleneck is not really ipc but memory bandwith.

 

[RandallFlagg]

I am pretty familiar with the SOTR bench, couple things:
1) that bench is poorly optimized for processors on the DEMO build - and the newer builds take much better advantage of processor performance:
2) untuned (i.e. not memory overclocked Zen 3) score almost the same despite being 25%-30% faster than Skylakes on this build
3) 6800XT will absolutely smoke a 3080 in average fps at 1080P due to a bit of nvidia driver overhead, which will impact CPU score a bit (zen 3 does better here)
here is my 10850K w/ 3080 at 1080P Lowest on the real build of the game:
View attachment 223443

Am I bottlenecking? Yes - I am bottlenecking anywhere to 215FPS (highest) to 238FPS (lowest) average at 1080P for this game...

the bottleneck is not really ipc but memory bandwith.

I intentionally only looked for benchs that followed the rules - 1080P, highest settings, DLSS and RT off. The hardest part of using that thread to find useful information is that people don't follow the rules.

And yes the CPU numbers can vary significantly and definitely favor OC / high mem speeds on Skylake derivatives which have always been the tweakers favorites and with good reason. Still, for those looking for knowledge as opposed to talking points, the information in that thread is quite valuable IMO.

 

[phanbuey]

I intentionally only looked for benchs that followed the rules - 1080P, highest settings, DLSS and RT off. The hardest part of using that thread to find useful information is that people don't follow the rules.

And yes the CPU numbers can vary significantly and definitely favor OC / high mem speeds on Skylake derivatives which have always been the tweakers favorites and with good reason. Still, for those looking for knowledge as opposed to talking points, the information in that thread is quite valuable IMO.

If you look at that bench you will see there isn’t a large difference between processors at stock memory settings.

 

[RandallFlagg]

If you look at that bench you will see there isn’t a large difference between processors at stock memory settings.

Ya I know. I'm not sure why people buy K and X processors and run them at stock, kinda like people who buy boats and never go boating and whatnot. Buy a Hellcat and drive the speed limit. w/e

And I know where you are going with this, yes the 10850K can get 20% better than at stock pretty easy with minimal effort (literally 220 CPU score vs 180ish stock) while the 3800X cannot - provided you're willing to spend an extra $100 on an AIO and $50 on faster RAM.

Still, a stock 10850K with power limits intact and no fancy cooling or RAM speed will not keep up with a 3080. This means 90% of 10850K's in existence can't keep up with the 3080, which means gen 10's time being able to push a GPU to its limit is probably going to end with the next gen of GPUs in 2022. This is not a slam on intel because frankly Zen 2 is in worse shape on a 3080 or 3090, and I saw a benchmark where it appeared a 2600X is about a perfect match to a 2060 Super. Just that Skylake's time is about to pass, too. The GPUs are just getting too fast.

 

[phanbuey]

Ya I know. I'm not sure why people buy K and X processors and run them at stock, kinda like people who buy boats and never go boating and whatnot. Buy a Hellcat and drive the speed limit. w/e

And I know where you are going with this, yes the 10850K can get 20% better than at stock pretty easy with minimal effort (literally 220 CPU score vs 180ish stock) while the 3800X cannot - provided you're willing to spend an extra $100 on an AIO and $50 on faster RAM.

Still, a stock 10850K with power limits intact and no fancy cooling or RAM speed will not keep up with a 3080. This means 90% of 10850K's in existence can't keep up with the 3080, which means gen 10's time being able to push a GPU to its limit is probably going to end with the next gen of GPUs in 2022. This is not a slam on intel because frankly Zen 2 is in worse shape on a 3080 or 3090, and I saw a benchmark where it appeared a 2600X is about a perfect match to a 2060 Super. Just that Skylake's time is about to pass, too. The GPUs are just getting too fast.

But SOTR does not really show this. What SOTR shows is exactly the opposite, that game build, and ram speed are more important than IPC/ processor power - to an extent. The 10850K has much lower IPC, less cache and fewer cores than a 5900X - here is SOTR. Even if I took my overclock off and ran at 4.8Ghz I would beat that 5900x at stock.

Stock 5900X w/ 3080

The 10850K w/ a 3080

I understand what you're trying to say, but by this logic that 5900x also can't keep up with a 3080 (it can).

Different game build on 10850K:

I honestly think DDR5 and AMD's 3D cache are going to make a bigger difference than more IPC/Cores. Zen 3 and even Skylake are bandwidth starved -- if you feed them enough bandwidth & lower latency they scale beyond 150-200FPS for bottlenecking even the highest end cards at low resolutions. Rocket lake, for example, has a higher IPC than Skylake by about 20%, and basically ties (and even loses) some games to Skylake -- that to me shows that the current memory platform as a whole is a bottleneck. That extra processor power is just going to waste.

 

[Vayra86]

But SOTR does not really show this. What SOTR shows is exactly the opposite, that game build, and ram speed are more important than IPC/ processor power - to an extent. The 10850K has much lower IPC, less cache and fewer cores than a 5900X - here is SOTR. Even if I took my overclock off and ran at 4.8Ghz I would beat that 5900x at stock.

Stock 5900X w/ 3080
View attachment 223462


The 10850K w/ a 3080
View attachment 223463

I understand what you're trying to say, but by this logic that 5900x also can't keep up with a 3080 (it can).

Different game build on 10850K:
View attachment 223466

I honestly think DDR5 and AMD's 3D cache are going to make a bigger difference than more IPC/Cores. Zen 3 and even Skylake are bandwidth starved -- if you feed them enough bandwidth & lower latency they scale beyond 150-200FPS for bottlenecking even the highest end cards at low resolutions. Rocket lake, for example, has a higher IPC than Skylake by about 20%, and basically ties (and even loses) some games to Skylake -- that to me shows that the current memory platform as a whole is a bottleneck. That extra processor power is just going to waste.

Exactly.

CPU Performance in games is down to optimization and how the API / abstraction is handled. We have literally a whole slew of DirectX versions worth of proof. Every time the burden is on developers to make use of what's possible and do it the right way. If they don't, it costs CPU cycles. GPU is simple, you just crunch pixel info as fast as you get it. CPU is a timing thing and it directly relates to how code is used.

For gaming, raw CPU performance was NEVER a big issue. In DirectX 11, the problem wasn't primarly the lack of CPU grunt, but the lack of threading and handling high counts of draw calls, struggling with cache coherency, scheduling, etc.. Closer to metal APIs fix this - the CPUs are still the same. Similarly, a different API can in that way also improve GPU performance on the same CPU. Why? Because there is less waiting to draw frames. The GPU and CPU are the same, and yet you get more pixels crunched, because code is executed faster early in the pipeline.

The top-down view on CPU performance in gaming based off an FPS number and an ingame counter was always flawed, and it always will be. Processing is a pipeline and the weakest link is really never in hardware - you meet specs or you don't - but in coding things proper. And yes, you can use more hardware to mitigate the problem of shitty code, but you're not fixing anything. You're just throwing your money at a problem someone else should have fixed proper - and in many games, where the CPU efficiency is abysmal, complaints lead to improvements in due time. Especially now that the APIs do support threading. Developers can barely get away with games choking on one thread now.


Really, for PC gaming hardware, there is one overarching trend that echoes through the past few decades: we move to a new level of mainstream performance, and lagging way behind that OR jumping way above it is just a big pile of overexpensive waste of time. Gains suffer heavily from lack of optimization for those performance levels, and with the prevalence of console ports, that has only gotten worse. New console generations determine what becomes feasible on the PC, most of the time. That even goes for the adoption of RT right now. Not Nvidia nor the PC gaming space can or will carry that on its own. VR: same rocky start for gaming, chicken-egg problems and it still remains a niche. And that technology isn't new either, has been attempted before. Even natively developed DX12 or Vulkan games are only now starting to pop up - because consoles have now adopted it, so there is an economical advantage.

Its good to recognize these facts because they place the practical use of all your upgrading in a rather different perspective. Trailing under the top-end of performance is generally the best way to go, and lagging a gen or two behind is very cost effective. Match the mainstream as long and as closely as possible, and you have the least problems, the best value for money and a good/great experience, simply because the market demands it.

 

[AusWolf]

Exactly.

CPU Performance in games is down to optimization and how the API / abstraction is handled. We have literally a whole slew of DirectX versions worth of proof. Every time the burden is on developers to make use of what's possible and do it the right way. If they don't, it costs CPU cycles. GPU is simple, you just crunch pixel info. CPU is a timing thing and it directly relates to how code is used.

For gaming, raw CPU performance was NEVER a big issue. In DirectX 11, the problem wasn't primarly the lack of CPU grunt, but the lack of threading and handling high counts of draw calls. Closer to metal APIs fix this - the CPUs are still the same. Similarly, a faster API can in that way also improve GPU performance on the same CPU. Why? Because there is less waiting to draw frames. The GPU and CPU are the same, and yet you get more pixels crunched.

The top-down view on CPU performance in gaming based off an FPS number and an ingame counter was always flawed, and it always will be. Processing is a pipeline and the weakest link is really never in hardware - you meet specs or you don't - but in coding things proper. And yes, you can use more hardware to mitigate the problem of shitty code, but you're not fixing anything. You're just throwing your money at a problem someone else should have fixed proper - and in many games, where the CPU efficiency is abysmal, complaints lead to improvements in due time. Especially now that the APIs do support threading. Developers can barely get away with games choking on one thread now.

One might argue that game tests are irrelevant because they employ different code to use hardware differently. One might argue that X API is irrelevant compared to another one because of the same reason. One might argue that testing X game is irrelevant because it's not optimised well. All of these are valid points from a strictly scientific point of view, but you can't describe real user experience in a scientific way.

Also, there is no golden way to test anything, as every hardware and software is different. You can establish trends at best, but there will always be some deviation. For that reason, I say every hardware is good hardware if it serves your needs, and every test is a good test if it relates to your specific use case. Arguing about which test method is irrelevant and why is irrelevant.

I do agree about games not being CPU-dependent, though (testing at 720p minimum with an RTX 3090 really is... irrelevant).

 

[Vayra86]

every test is a good test if it relates to your specific use case. Arguing about which test method is irrelevant and why is irrelevant.

I do agree about games not being CPU-dependent, though (testing at 720p minimum with an RTX 3090 really is... irrelevant).

This is key and far too often, overlooked. And it begins with a proper specification of your use case, quite a few of those are rather dishonest (for their own selves) - and it often seats on a lack of insight/knowledge of how PCs and games work. OTOH we're making progress as a gaming community too - think of the revelations on frame pacing etc. So if the experience is truly lacking, we'll move in to fix it because we can collectively recognize it. This is what makes up the mainstream progress.

Good examples are everywhere. New consoles now support higher refresh rates. Have multiple quality/performance modes. Are somewhat tweakable. And PCs have also moved towards the console space. We're blending, another trend that has been going on for years but is now also technically feasible.

 

[AusWolf]

This is key and far too often, overlooked. And it begins with a proper specification of your use case, quite a few of those are rather dishonest (for their own selves) - and it often seats on a lack of insight/knowledge of how PCs and games work.

Agreed.

It is commonly overlooked by reviewers as well. They tend to say things like "X CPU is crap because it gets beaten by Y CPU by a margin of 5%". Then they show graphs of averages across a million games, when in fact, nobody notices 5% in real life and nobody plays "Average Game" as there is no such thing.

They also say things like "X CPU uses a lot more power, therefore it runs hotter" when actually, the two statements share no logical connection - primarily, larger heat density (more heat in a smaller area) and heat dissipation issues lead to hotter chips. I had to learn this the hard way, not being able to cool a R5 3600 at stock with the same case and cooler that my i7-11700 is happy with even with a modified 125 W PL1.

In our informationally saturated times, one must read between the lines of marketing BS and reviewer's (false) conclusions and find information crumbs that are relevant to one's own use case, and establish a picture from there. X CPU isn't good because it beats Y CPU, but because it fits your use case. Whether it's a well-optimised game or a shitty console port, doesn't matter. The fact that your PC runs it, and you enjoy playing it does.