AMD Radeon RX 6400 Tested on PCI-Express 3.0

[W1zzard]

AMD is using a fairly narrow PCIe x4 interface on their Radeon RX 6400 GPU. We're taking AMD's new budget offering for a spin in a PCI-Express 3.0 configuration to determine how big the performance hit will be when running on an older system.

Show full review

 

[Berfs1]

Why in Far Cry 5 the 1060 doesn't have a huge performance drop in 1440p vs 1080p, but all the others its massive?

 

[lightofhonor]

Thanks for doing this. It would be interesting to see how settings can affect this scaling since on 3.0 I can get 60fps on Halo Infinite at 1080p with low settings. I don't think anyone would actually play it at 7fps with your settings.

 

[ModEl4]

Why in Far Cry 5 the 1060 doesn't have a huge performance drop in 1440p vs 1080p, but all the others its massive?

6GB?

 

[Sabotaged_Enigma]

Doesn't make too much difference 'cause it's already weak enough... Just a full version of iGPU Radeon 680M...

 

[Denver]

Is this difference lower with low/medium quality?

 

[defaultluser]

Is this difference lower with low/medium quality?

Why does it matter?

I can get playable mix of settings on my HTPC's ancient GTX 960 at 1080p in most recent games , but that doesn't mean I want to buy a replacement with the same performance at PCIe 3!

This thing has a short lifetime as a viable 1080p low card (because they cut VRAM bandwidth and capacity in-half versus the 6600 , AND pcie lane bandwidth to a QUARTER!)

I mean, look at that drop-off:

It' only going to get worse with each passing-year

 

[docnorth]

I had hopes that the performance hit would be lower because the card is weaker than 6500 xt, but I was wrong. Maybe better with lower settings (obviously more realistic for this GPU). Thanks again for your work and time @W1zzard.

 

[GoldenX]

CPU <> GPU transfers are the death of this card. And I'm starting to suspect that's the main reason AMD removed the encoding engine.

Solution: MCM 2 of these to get an 8x card!

I had hopes that the performance hit would be lower because the card is weaker than 6500 xt, but I was wrong. Maybe better with lower settings (obviously more realistic for this GPU). Thanks again for your work and time @W1zzard.

Frametime is king.
The PCIe bottleneck is a fixed delay, since both the 6500 and 6400 share the same maximum bandwidth, but since the 6400 is weaker, it also has to spend more time in every rendering step. Longer frametime, lower FPS.

 

[mama]

Not sure how a product with less potential than the most mediocre product from the previous generation gets a release for the current generation. I thought the 6500XT was tight!

 

[mechtech]

cool stuff, but where is pcie v 2.0 and 1.1??

 

[trsttte]

cool stuff, but where is pcie v 2.0 and 1.1??

Check the article for the 6500xt, given the results for the 3.0 are about the same, 2.0 should follow suit

AMD Radeon RX 6500 XT PCI-Express Scaling

The AMD Radeon RX 6500 XT comes with only a narrow PCI-Express x4 interface. In this article, we took a closer look at how performance is affected when running at PCI-Express 3.0; also included is a full set of data for the academically interesting setting of PCI-Express 2.0.

www.techpowerup.com

 

[Tetras]

Why does it matter?

I can get playable mix of settings on my HTPC's ancient GTX 960 at 1080p in most recent games , but that doesn't mean I want to buy a replacement with the same performance at PCIe 3!

This thing has a short lifetime as a viable 1080p low card (because they cut VRAM bandwidth and capacity in-half versus the 6600 , AND pcie lane bandwidth to a QUARTER!)

I mean, look at that drop-off:

It' only going to get worse with each passing-year

It is kind of crazy we're talking about compromises to make a card that costs $200 play modern games comfortably @ 1080p (the equivalent of a phone screen nowadays).

 

[RedBear]

Thanks for the tests, this should hopefully put to rest to the bizarre claim that this GPU might not be significantly limited in PCIe Gen 3.

 

[chrcoluk]

Is there a meaningful cost saving by reducing the lanes on the card? This seems to be a way to get people to buy gen 4 platforms?

Something I did observe in my own testing though is that 8 lanes gen 3 is faster than 16 lanes gen 2. I think there is efficiency improvements in newer gen aside from the bandwidth?

I think my GT 1030 has 8 lanes which was I think the lowest Pascal card released, so why did they drop to 4 lanes instead of 8? Seems very aggressive.

Thanks for the tests, this should hopefully put to rest to the bizarre claim that this GPU might not be significantly limited in PCIe Gen 3.

Was the case with 16 lane products, I feel like in my previous post this decision may have got made to shift hardware for gen 4 products.

 

[RedBear]

Something I did observe in my own testing though is that 8 lanes gen 3 is faster than 16 lanes gen 2. I think there is efficiency improvements in newer gen aside from the bandwidth?

There is, Gen 2 (and Gen 1) used 8b/10b line encoding, Gen 3 switched to 128b/130b, there's less overhead.

I think my GT 1030 has 8 lanes which was I think the lowest Pascal card released, so why did they drop to 4 lanes instead of 8? Seems very aggressive.

GT 1030 has 4 lanes, but it was launched for $79. Crazy pricing aside it's actually similar in concept to these (Navi 24) GPUs, it reuses the GP108 chip that Nvidia used in many iterations of its MX series.

Was the case with 16 lane products, I feel like in my previous post this decision may have got made to shift hardware for gen 4 products.

I think it doesn't make a lot of sense for AMD to do this on purpose, it basically kills one realistic usage for this low profile GPU: giving a relatively "cheap" upgrade to old prebuilts that can't accept anything larger and don't have standard 6pin PSU connectors. Also, AMD essentially offers no cheap PCIe Gen 4 CPU that you could match with this thing, Ryzen 3100 is seemingly out of stock, so it's like they're asking people to buy a cheap Alder Lake part to match this GPU, it's counterproductive for them.

 

[chrcoluk]

Maybe some AIB 1030's had lanes increased? my 1030 in a 16x slot uses 8 lanes under load. However it never goes above 10% even with 4 lanes (on gen 3), to me releasing a product with 4 lane limitations that needs most of that bandwidth seems nonsensical as you cutting off a big chunk of the market, but because the supply and demand is so bonkers right now, I think they decided to use it to sell more chipsets and cpu's or given your info on the lack of low end gen 4 parts just didnt care.

Thanks for the info on the overhead stuff, explains my testing data.

I wonder if TPU can do a poll to see the spread across its userbase on gen 3/4/5 board/cpu ownership.

 

[RedBear]

Can you share the model of your GT 1030? I was pretty much sure that every GT 1030 used four lanes, maybe some had a x8 PCIe bridge for some reason? This model from PNY claims to work at x8, for instance.

 

[trsttte]

Can you share the model of your GT 1030? I was pretty much sure that every GT 1030 used four lanes, maybe some had a x8 PCIe bridge for some reason? This model from PNY claims to work at x8, for instance.

Weird claim when the product picture seems to only show 4 lanes connected

 

[MarsM4N]

If you exclude the few exeptional games, the difference is actually pretty neglectable.

I also find the games selection pretty off. Most are practically unplayable (under 60fps at 1080p). Kinda missed the target audience who buy these cards, playing for the most part lower demanding multi player games like CS:GO, Valorant, League of Legends, Apex Legends, Dota 2, Overwatch, Dead by Daylight, PubG, Fortnite, TF2, Payday2, War Thunder, Rocket League, etc.

Sources: Steam Game Stats & Twitch Streams

I understand that benchmarking in MP environments is a tricky one, but it should be adressed somehow to get some meaningful tests.

 

[AusWolf]

A very interesting review!

What makes it especially interesting to me is that I've found other reviews where God of War doesn't care about the PCI-e bandwidth, but here it does.

Also, Metro Exodus seems to be running fine here, but it (or at least its benchmark) is totally unplayable on my 6400 in PCI-e 3.0. I wonder why this is.

 

[Mussels]

Hey w1zz, since you speculate that the performance drop is due to running out of VRAM and paging to system RAM, did you feel up to doing a quick test with that - find the game with the biggest performance loss, and see how system RAM Speed affects it?

As a total pain in the ass idea to stick in your mind, would a user with slower RAM be hurt even more? (Is this why some reviewers find less of a change, because they have top tier memory?)

 

[AusWolf]

Hey w1zz, since you speculate that the performance drop is due to running out of VRAM and paging to system RAM, did you feel up to doing a quick test with that - find the game with the biggest performance loss, and see how system RAM Speed affects it?

As a total pain in the ass idea to stick in your mind, would a user with slower RAM be hurt even more? (Is this why some reviewers find less of a change, because they have top tier memory?)

I can do that. What game eats the most VRAM nowadays?

 

[Jism]

These cards where actually designed for Ryzen 3 series and thus going with PCI-E 4.0 and not 3. It's basicly a GPU chip sticked onto a desktop card. Its good for OEM's that build systems based on PCI-e 4.0 but not consumers. A RX580 would even be better in regards of performance.

 

[Frick]

If you exclude the few exeptional games, the difference is actually pretty neglectable.

I also find the games selection pretty off. Most are practically unplayable (under 60fps at 1080p). Kinda missed the target audience who buy these cards, playing for the most part lower demanding multi player games like CS:GO, Valorant, League of Legends, Apex Legends, Dota 2, Overwatch, Dead by Daylight, PubG, Fortnite, TF2, Payday2, War Thunder, Rocket League, etc.

Sources: Steam Game Stats & Twitch Streams

I understand that benchmarking in MP environments is a tricky one, but it should be adressed somehow to get some meaningful tests.

There's also a lot of people buying low end GPUs for the purpose of playing modern games besides the titles you listed, and for them keeping 60fps at all times may not be that important. Remember these reviews uses max settings, I'd imagine many of them could get around 45fps with some tinkering and still look good and that is fine. Maybe even 30fps is ok. And they probably have older, lower end systems. The H110 and H310 chipsets only supported PCIe 2.0 which would absolutely cripple this card.