Bottleneck? AMD Radeon RX 6700 and AMD Ryzen 5 3400G

[cake_1]

After getting my PC built, the technician told me I'd suffer a bottleneck with my GPU and CPU. As the title says, I have an AMD Radeon RX 6700 10GB and a AMD Ryzen 5 3400G with Radeon Vega Graphics 3.70 GHz.
So far I've only tested the performance hit on Skyrim with ENBs and high graphic settings, I get 30 FPS in certain areas so I'm guessing that's because of the bottleneck?
All in all I was just wondering if I should try more demanding and modern games or not. I don't wanna set my PC on fire...
I appreciate any advice!

 

[Durvelle27]

It definitely won’t cause a fire but that CPU is absolutely dog water and a severe bottleneck.
Luckily you can easily upgrade the CPU with no issues and have many options to choose from. I’d recommend a 5600 or 5600X which would pair nicely with that GPU.

 

[tussinman]

Yeah that's a pretty good bottleneck. The 3400G if I remember correctly is slightly slower than a 9th gen i3

Probably the only thing saving it from being a gigantic issue is your monitor is only capped at 60hz so it's not like you're doing high FPS gaming. Bottlenecks like this can still exist at 60hz but is way more noticeable or annoying at higher FPS (EX: Can't get more than 70FPS on your favorite games but your monitor can run 144hz).

For your scenario (6700 with 3400G at 60hz) what would probably be the most noticeable is dips more so average FPS

 

[Count von Schwalbe]

Yep. The 3400G is the Zen+ uArch, pretty low on the scale in 2023. It is also a quad-core, which isn't much for modern AAA games. Your 6700 could stretch its legs a lot more if you had a Zen 3 CPU in there.

R5 5500 at minimum, 5600(X) or even better 5700X would be good. 5800X3D if you have the money - those things are awesome and you could keep the system for many years to come.

 

[Durvelle27]

Yeah that's a pretty good bottleneck. The 3400G if I remember correctly is slightly slower than a 9th gen i3

Probably the only thing saving it from being a gigantic issue is your monitor is only capped at 60hz so it's not like you're doing high FPS gaming. Bottlenecks like this can still exist at 60hz but is way more noticeable or annoying at higher FPS (EX: Can't get more than 70FPS on your favorite games but your monitor can run 144hz).

For your scenario (6700 with 3400G at 60hz) what would probably be the most noticeable is dips more so average FPS

It’s normally a misconception of bottlenecks being a problem just at FPS 60 or above. A lot of people seem to forget the lows which actually make the difference between playable and unplayable. You can have 100FPS but if you get dips to 30 or less your experience will be horrid and the stutter irritating. And with a CPU like the 3400G it’s definitely going to have crap lows which IPC plays a huge part of.

 

[eidairaman1]

After getting my PC built, the technician told me I'd suffer a bottleneck with my GPU and CPU. As the title says, I have an AMD Radeon RX 6700 10GB and a AMD Ryzen 5 3400G with Radeon Vega Graphics 3.70 GHz.
So far I've only tested the performance hit on Skyrim with ENBs and high graphic settings, I get 30 FPS in certain areas so I'm guessing that's because of the bottleneck?
All in all I was just wondering if I should try more demanding and modern games or not. I don't wanna set my PC on fire...
I appreciate any advice!

Get hwinfo64 so we can see all that you have as of hardware, the powersupply you will have to take a picture of.

 

[tussinman]

It’s normally a misconception of bottlenecks being a problem just at FPS 60 or above. A lot of people seem to forget the lows which actually make the difference between playable and unplayable. You can have 100FPS but if you get dips to 30 or less your experience will be horrid and the stutter irritating. And with a CPU like the 3400G it’s definitely going to have crap lows which IPC plays a huge part of.

Depends on the type of games.

I just recently retired a 3770k system with slow 1333mhz ram (overall probably 15 to 20 percent slower then his setup) and on all the modern AAA titles I played (ff7 remake, red dead 2, hitman 3, Forza horizon 5, God of war) I never noticed any irritating stutter or drops. The in game benchmarks never showed bad 1% lows either

Don't get me wrong I'm not defending the 3400g but on low hz monitors it's going to be a little harder to notice drops especially on single player heavy games (poorly optimized games and online heavy ones sure but those arent easy to run even on expensive hardware)

 

[Lew Zealand]

It totally depends on game type, I used an i7-4790 (close enough to the 3400G) with a 6600XT for a bit and while it was CPU bound in many games, that doesn't mean that they dropped below 60fps much. However when playing games that have that CPU demand (Horizon ZD fights with lots of bots and NPCs), it would dip to the 40s. That's not stuttering, that's a reduction in fps. It was still smooth though those are marginal fps for my playing taste.

The 6700 will be bottlenecked by the 3400G in more recent CPU demanding games but not in all. Try your games out and use the AMD Adrenaline overlay to view your FPS, %GPU use and %CPU use. If your %GPU use goes below 95 or 90% when you see those 30fps numbers, then you have a bottleneck. And if those 30fps times bug you, then consider an R5 5500 or better CPU upgrade. If they don't bug you much, then just enjoy!

 

[NSX05]

After getting my PC built, the technician told me I'd suffer a bottleneck with my GPU and CPU. As the title says, I have an AMD Radeon RX 6700 10GB and a AMD Ryzen 5 3400G with Radeon Vega Graphics 3.70 GHz.
So far I've only tested the performance hit on Skyrim with ENBs and high graphic settings, I get 30 FPS in certain areas so I'm guessing that's because of the bottleneck?
All in all I was just wondering if I should try more demanding and modern games or not. I don't wanna set my PC on fire...
I appreciate any advice!

You won't have any fires.

What games do you play, and what board do you have?

 

[mechtech]

After getting my PC built, the technician told me I'd suffer a bottleneck with my GPU and CPU. As the title says, I have an AMD Radeon RX 6700 10GB and a AMD Ryzen 5 3400G with Radeon Vega Graphics 3.70 GHz.
So far I've only tested the performance hit on Skyrim with ENBs and high graphic settings, I get 30 FPS in certain areas so I'm guessing that's because of the bottleneck?
All in all I was just wondering if I should try more demanding and modern games or not. I don't wanna set my PC on fire...
I appreciate any advice!

Depends on resolution and frame rate. If your monitor is 1080p and 60HZ and you optimize your video settings should be no issue. Sometimes there is quite a few graphics settings that are diminishing returns imho. They really nerf fps for trivial increases in eye candy. Just do a search and should be lots of recommendations for ‘optimized’ fps vs eye candy settings. CPU usually becomes less of an issue at higher resolution like 4k.

 

[cake_1]

You won't have any fires.

What games do you play, and what board do you have?

if board means motherboards then I have an ASUS TUF Gaming B450 PLUS II AM4. So far I've only tried Skyrim with ENBs and high graphic settings. Also The Sims 4 but runs smoothly.

Depends on resolution and frame rate. If your monitor is 1080p and 60HZ and you optimize your video settings should be no issue. Sometimes there is quite a few graphics settings that are diminishing returns imho. They really nerf fps for trivial increases in eye candy. Just do a search and should be lots of recommendations for ‘optimized’ fps vs eye candy settings. CPU usually becomes less of an issue at higher resolution like 4k.

I see. I do have a 60hz monitor so should I change the graphic setting from the game a bit? Or you mean other type of settings

 

[Dr. Dro]

if board means motherboards then I have an ASUS TUF Gaming B450 PLUS II AM4. So far I've only tried Skyrim with ENBs and high graphic settings. Also The Sims 4 but runs smoothly.


I see. I do have a 60hz monitor so should I change the graphic setting from the game a bit? Or you mean other type of settings

When you refer to Skyrim, are you running the old version of Skyrim (released in 2011), or the newer Special Edition? In any case, you will definitely benefit from a CPU upgrade across pretty much any and all games.

Spoiler: technical babble

If it's the latter, the primary issue lies in the Creation Engine itself. It's basically a sworn archenemy of Radeon graphics as it makes extensive use of driver command lists with deferred contexts, which are AMD's achilles heel in DirectX 11 games. You often hear of the infamous frame rate issues that Fallout 4 exhibits in the Boston outskirts and in and around the Corvega plant, this occurs due to an overload in the graphics driver caused by excessive amount of draw calls. The same phenomenon occurs in modded installations of Skyrim SE and in Fallout 76. A well-written engine wouldn't overload the graphics driver, but this is not a well-written engine.

Nvidia cards can batch and defer draw calls across multiple processors through the use of a command list, so the graphics driver is able to scale this workload across multiple CPU cores. The result is that it makes CPU usage balloon up and heavily strains the processor, but as long as you have CPU resources left the frame rate will scale accordingly (or even be hurt if the game ends up fighting for resources - this is often seen on low end 4-thread CPUs), while on AMD, the graphics driver uses the GPU's own command processor in an immediate context, which means that the driver will serially introduce commands using a single CPU thread into the GPU, causing a massive bottleneck (or a potential relief in the same thread-bound scenario I've described earlier).

Note that my information on this might not be fully up to date, for RDNA 2 and 3 GPUs AMD has recently rewritten the DirectX 11 stack to run on their new PAL abstraction layer and I do not know if they have finally introduced command lists support. GPU-Z can tell you if it is supported by going on Advanced tab and then selecting DirectX 11, I would actually like to know if this is supported in RDNA 3 as I am a massive fan of Bethesda RPGs and this is an actual consideration for me. What I do know, however, is that RX Vega cards and GCN-based APUs do not support and do not benefit from the new PAL abstraction layer.

 

[QuietBob]

for RDNA 2 and 3 GPUs AMD has recently rewritten the DirectX 11 stack to run on their new PAL abstraction layer and I do not know if they have finally introduced command lists support. GPU-Z can tell you if it is supported by going on Advanced tab and then selecting DirectX 11

According to GPU-Z, this feature isn't currently implemented on RDNA3:

But we see a similar bottleneck in games utilizing the Creation engine on current NVidia cards as well. Even the RTX4090 combined with a 13900K suffers huge fps drops in some scenes, as evidenced in this thread.

 

[Dr. Dro]

According to GPU-Z, this feature isn't currently implemented on RDNA3:

View attachment 296490

I see, that is a shame. I've linked this same article to someone else while explaining the whole deal with scheduling earlier today, but you should read it too. It's good technical info to understand how draw call handling works:

Game Development

Access tools, tutorials libraries, and code samples from Intel to optimize your games.

www.intel.com

A very common misconception is that AMD's graphics driver is more optimized and thus places a lower burden on the CPU, but that isn't what occurs. It's the approach that Nvidia chose that favors PCs with high spec CPUs with more room for improvement (i.e. widening their performance potential at the cost of higher CPU load) over ensuring a smoother experience at the low end. Many games that seem to perform inexplicably poorly in an Nvidia GPU with a low end CPU are actually experiencing an unusual situation where the graphics driver is fighting for resources with the game itself, while Radeon's conservative threading keeps the frame rate more consistent in that scenario. Recently this has also affected some games with unusually high CPU usage due to bugs and/or DRM, with similar behavior where the Radeon cards punch way above their usual weight.

Last that I've had a chance to play with this, the Radeon driver with the Radeon VII didn't use more than 6 CPU threads even in the most parallelized tests, but Nvidia's would load all 36 threads on the same processor.

Unfortunately it's not a setting you can choose and is intertwined with the driver code itself. Nvidia explains the different driver instancing models on this article:

D3D11 Deferred Contexts Sample

docs.nvidia.com

I'm not a graphics programmer but I've spent a very long time poring over the documentation on this specific issue trying to understand it better, as I've always been very frustrated with Creation's performance on Radeon. I've come to the conclusion the engine is just very unfriendly to the way AMD elected to write their driver, which isn't bad, just different.

Oh well, as our Godd and Savior would say, "see that mountain? you can climb it, it just works, sixteen times the detail. thank you for listening to my ted talk and remember to buy Skyrim".

BTW checked out that thread, note that they're using a i7-9700K: it's essentially the one modernish processor you didn't want to use with a 4090 if you wanted to upkeep performance in this scenario. Having only 8 threads will really have it by the balls here, you could knock it down to the 4070 and they'd be getting the exact same fps with that CPU, I'm sure of it.

Creation also has a weird affinity for the Ryzen's 3D VCache for some reason. 7950X3D just demolishes even the 13900KS in this game.

 

[QuietBob]

A very common misconception is that AMD's graphics driver is more optimized and thus places a lower burden on the CPU, but that isn't what occurs. It's the approach that Nvidia chose that favors PCs with high spec CPUs with more room for improvement (i.e. widening their performance potential at the cost of higher CPU load)

Thanks for the links, I'm currently reading these. In theory, Nvidia's approach to DX11 MT rendering should have been more future proof, but it seems the benefits may not always translate to actual games.

I just ran the 3DMark API Overhead test, which employs the same method as the one in the Intel article you posted. The DX11 MT run uses all available CPU threads. In this particular test RDNA3 doesn't appear to take advantage of multiple threads utilized in deferred context rendering:



Even though this is an unsupported test from 2015, it'd be interesting to see how contemporary Nvidia hardware does here when not limited by the CPU. Could you run this test on your system and post the results?

BTW checked out that thread, note that they're using a i7-9700K

There is a user there with a 13900K and an RTX4090 showing 42% GPU utilization in a problematic area in Fallout 4, and another one with a 12900K showing 40% usage with the same card.

 

[mechtech]

if board means motherboards then I have an ASUS TUF Gaming B450 PLUS II AM4. So far I've only tried Skyrim with ENBs and high graphic settings. Also The Sims 4 but runs smoothly.


I see. I do have a 60hz monitor so should I change the graphic setting from the game a bit? Or you mean other type of settings

Well. Ideally make sure your monitor is set to default resolution and refresh rate. Probably 1920x1080 at 60 Hz.

As for graphics settings it’s personal preference. I personally disable hdr/bloom, depth of field and some others I don’t really like or really hinder performance. If it runs to your liking just sit back and enjoy or you will get lost down the upgrade rabbit hole.

 

[Dr. Dro]

Thanks for the links, I'm currently reading these. In theory, Nvidia's approach to DX11 MT rendering should have been more future proof, but it seems the benefits may not always translate to actual games.

I just ran the 3DMark API Overhead test, which employs the same method as the one in the Intel article you posted. The DX11 MT run uses all available CPU threads. In this particular test RDNA3 doesn't appear to take advantage of multiple threads utilized in deferred context rendering:

View attachment 296514

Even though this is an unsupported test from 2015, it'd be interesting to see how contemporary Nvidia hardware does here when not limited by the CPU. Could you run this test on your system and post the results?


There is a user there with a 13900K and an RTX4090 showing 42% GPU utilization in a problematic area in Fallout 4, and another one with a 12900K showing 40% usage with the same card.

I'll definitely run it again but just this old score with a GTX 580(!) has your 7900 XTX beat in DX11MT with a slower CPU too, and I have a hunch that is why the 3DMark guys discontinued this test, there's just no possible way any game will run better on a 580 than on your card

I scored 0 in API Overhead Test

AMD Ryzen 9 3900XT, NVIDIA GeForce GTX 580 x 1, 65536 MB, 64-bit Windows 10}

www.3dmark.com

I see about the guys using a 13900K, was very sleepy... I'm installing F4 for another playthrough but it takes a few days with the mod setup I run. It's some 80 GB of stuff I have to sift and sort through

@QuietBob I have the update for you, ran it on my PC

I scored 0 in API Overhead Test

Intel Core i9-13900KS Processor, NVIDIA GeForce RTX 3090 x 1, 32768 MB, 64-bit Windows 10}

www.3dmark.com

 

[QuietBob]

a GTX 580(!) has your 7900 XTX beat in DX11MT with a slower CPU too

According to the whitepaper, this synthetic test represents a hypothetical "best-case" scenario. Unlike typical 3DMark benchmarks, it cannot be used to compare GPU architectures, or even GPU performance in general. Its sole purpose is to highlight the difference in API overhead on a specific hardware configuration. As such, results from systems utilizing different graphics cards and processors are not comparable.

I have the update for you, ran it on my PC

Interesting results! So, in DX11 the RTX3090 could theoretically achieve 62% higher performance in CPU bound scenarios, when paired with the fastest 32-threaded desktop processor. And on condition that deferred context rendering was implemented properly in the game engine.

Creation also has a weird affinity for the Ryzen's 3D VCache for some reason. 7950X3D just demolishes even the 13900KS in this game.

I wish I could test the engine's dependence on the Vcache myself, but just realized I don't have Fallout 4. I was sure I grabbed it in some sale

 

[Dr. Dro]

That is true, but it also shows how lopsided the API overhead situation has been all along, and that doesn't quite translate in real world performance as you have seen. It can cause problems or huge benefits for both vendors within that range of whether something benefits from it or not. I think the true reason why 3DMark devs weren't willing to call this a competitive benchmark because it really depends so much on the system configuration that even two practically identical PCs can show wildly different results with changes that could be called generally minimal such as memory timings or interconnect bandwidth, and at the same time the drivers from each vendor do things their own way as well.

Regarding Fallout 4, it's not a new thing that it loves cache, I found out about it a very long time ago, that's actually the reason why I've purchased that 18-core Xeon for my old X99 machine. Despite low clocks, the 45 MB L3 was unheard of back in 2017, and Fallout 4 delighted itself on it vs. my previous 4770K build. Much better performance. That's also kind of a reason why these decommissioned Haswell and Broadwell Xeons became so popular, even though they finally began to show their age (as they have a lot of cache but IPC is terrible, and software just came about requiring a bit of both).