PCI-e 3.0 x8 may not have enough bandwidth for RX 5500 XT

[IceShroom]

As title suggest, RX 5500 XT 4GB may be bottlenecked by PCI-e 3.0.
Here is a article in (german) : https://www.pcgameshardware.de/Rade...pecials/PCI-Express-3-vs-PCI-E-4-GPU-1339415/

 

[Hyderz]

interesting...

 

[Sybaris_Caesar]

Aw. Since 5500 XT is only a x8 card instead x16 card. I just did a cursory glance on its reviews so this was a surprise. Guess it's an all around bad buy compared to GTX 1600 series.

Lol. What was AMD thinking? That people were gonna pair these xards with X570 systems?

 

[londiste]

Looking at the tested games, all are run with max details. Is this a VRAM-limited situation? That would very clearly explain the big differences because card would need to use system RAM and more bandwidth definitely benefits there.

I wonder if the results are same for 8GB cards?

 

[INSTG8R]

Yeah budget 4GB card further hobbled by x8 interface it can apparently saturate, which I have to say is a first. W1Z has regularly tested PCI bandwidth and its effects which for years ha# been barely #how any real losses.

Looking at the tested games, all are run with max details. Is this a VRAM-limited situation? That would very clearly explain the big differences because card would need to use system RAM and more bandwidth definitely benefits there.

I wonder if the results are same for 8GB cards?

I saw some charts earlier it was not that much higher but the lines we’re definitely not having the same issue

 

[Hyderz]

battlefield v has a massive increase

 

[Deleted member 67555]

It really does seem like it's being held back by the memory... What an awful thing to do.

 

[londiste]

Yeah budget 4GB card further hobbled by x8 interface it can apparently saturate, which I have to say is a first. W1Z has regularly tested PCI bandwidth and its effects which for years ha# been barely #how any real losses.

I saw some charts earlier it was not that much higher but the lines we’re definitely not having the same issue

Oh, you are right. I read the graphs wrong. For some reason I read 8GB cards as 5700
8GB versions do gain from PCIe 4.0 but not to the same degree as 4GB variants.

 

[potato580+]

tgis card targeting to competed with 1650super, sadly amd failed this time, 5500 xt is such a waste, it even cant compete with old gtx1060/rx480, and the price seems pretti strange at my country, $220 for an newcard which have same performance as old $60 rx480, no thx,unless it set on $150 i would probably pay for it, just for testing

 

[IceShroom]

tgis card targeting to competed with 1650super, sadly amd failed this time, 5500 xt is such a waste, it even cant compete with old gtx1060/rx480, and the price seems pretti strange at my country, $220 for an newcard which have same performance as old $60 rx480, no thx,unless it set on $150 i would probably pay for it, just for testing

What was the price for GTX1060/RX480 when it was new??

 

[potato580+]

What was the price for GTX1060/RX480 when it was new??

they were rarely on new stock these day, but based on my country pricetag, you can just grab rx590 for an $200, or even saving few buck by picking 1650 super, i strongly recomend 1650super this time, since it was more power efficent yet good performance

 

[IceShroom]

they were rarely on new stock these day, but based on my country pricetag, you can just grab rx590 for an $200, or even saving few buck by picking 1650 super, i strongly recomend 1650super this time, since it was more power efficent yet good performance
View attachment 139841

Well that chart is old and for RX 5500 not XT. Here is new chart and here GTX 1650 Super is slower than RX 5500 XT.
@W1zzard can you look into this PCI-e 4.0 x8 vs PCI-e 3.0 x8 for RX 5500XT on AMD X570??

 

[INSTG8R]

Oh, you are right. I read the graphs wrong. For some reason I read 8GB cards as 5700
8GB versions do gain from PCIe 4.0 but not to the same degree as 4GB variants.

Yeah there's been quite alot of surprise with the 4GB card. They literally hobbled it in almost every way possible.

 

[Vayra86]

Those are some serious gaps. Damn

 

[Vya Domus]

It's definitely something else other than the PCI-e bandwidth, there is literally no other card that shows this behavior not even the 5700XT.

 

[ratirt]

Yeah budget 4GB card further hobbled by x8 interface it can apparently saturate, which I have to say is a first. W1Z has regularly tested PCI bandwidth and its effects which for years ha# been barely #how any real losses.


I saw some charts earlier it was not that much higher but the lines we’re definitely not having the same issue View attachment 139840

What is the "Y" Axis represent? delay or FPS or something else like bandwidth? It s not clear to me and the graph doesn't seem to have any map-key and I'm not sure if I read id correctly.

 

[londiste]

It's definitely something else other than the PCI-e bandwidth, there is literally no other card that shows this behavior not even the 5700XT.

5700XT has twice the PCIe lanes and twice the memory. At the same time article addresses the memory part where PCIe 3.0 vs 4.0 results are pretty close with 8GB of memory. Low-end cards are often enough with PCIe x8 but 5500 is the first one with 4.0 capability so there is nothing to compare this to.

When VRAM runs out, the data will get put to system RAM and accessed over PCIe, doubling that bandwidth is expected to make a huge difference and it does seem to do exactly that.

What is the "Y" Axis represent? delay or FPS or something else like bandwidth? It s not clear to me and the graph doesn't seem to have any map-key and I'm not sure if I read id correctly.

Frametimes in milliseconds. Less is better and FPS is 1000/Frametime in ms - for example at 25ms frametime, 1000/25 = 40 FPS.

 

[INSTG8R]

What is the "Y" Axis represent? delay or FPS or something else like bandwidth? It s not clear to me and the graph doesn't seem to have any map-key and I'm not sure if I read id correctly.

Yeah sorry I can’t even give you the source I got it from a discord group when we all got suspicious someone posted that graph. Zooming in the Y axis appears ro be FPS but even then the green data point doesn’t make sense?

 

[Vayra86]

Yeah sorry I can’t even give you the source I got it from a discord group when we all got suspicious someone posted that graph. Zooming in the Y axis appears ro be FPS but even then the green data point doesn’t make sense?

Frametime lower is better; that is why the green one (lowest spec) is so spiky, it struggles visibly. The other setups lose performance, but are not 'stuttery'.

VRAM being low AND the PCIe bandwidth combined just creates a horrible cocktail... Though it does also seem like a driver/efficiency problem, it should be much more consistent under 3.0 given the data that is transported.

 

[arbiter]

Frametime lower is better; that is why the green one (lowest spec) is so spiky, it struggles visibly. The other setups lose performance, but are not 'stuttery'.

VRAM being low AND the PCIe bandwidth combined just creates a horrible cocktail...

Yea it would seem to be the memory is getting maxed out which then its calling upon system memory to act as vram. If that is the case that would explain why the frame times are so bad since its maxing bus it has out.

 

[Vayra86]

Yea it would seem to be the memory is getting maxed out which then its calling upon system memory to act as vram. If that is the case that would explain why the frame times are so bad since its maxing bus it has out.

Driver fix: lower FPS... Right now it just keeps running into that wall like a blundering idiot.

Yeah like I said I just used the first source available showing it tested in different PCI speeds it makes much more sense now. Doesn’t change the fact AMD really hobbled this card to the point of almost being pointles.

Yeah if this is how they intended to sell us PCIe 4.0 .... fail

 

[INSTG8R]

Frametime lower is better; that is why the green one (lowest spec) is so spiky, it struggles visibly. The other setups lose performance, but are not 'stuttery'.

VRAM being low AND the PCIe bandwidth combined just creates a horrible cocktail...

View attachment 139852

Yeah like I said I just used the first source available showing it tested in different PCI speeds it makes much more sense now. Doesn’t change the fact AMD really hobbled this card to the point of almost being pointles.

 

[londiste]

Wolfenstein Youngblood at Mein Leben! settings (which is not maxed if I remember correctly) consumes between 5-6 GB of VRAM. On a 4GB card that extra GB or two that does not fit into VRAM must be accessed over PCIe bus.

 

[Flaky]

The main takeaway from the article seems to be "don't set detail level so stupidly high that it exceeds card's vram". The only result that seems anomalous is one from Far Cry.

RX 460/550/560 also did have x8 link max, and it seems to be the new norm for mid-end cards from AMD.
One of the reasons is GPU link in their APUs - it is exactly x8. As they will be put together in many laptops, extra 8 lanes would be often wasted anyway.

 

[Vayra86]

The main takeaway from the article seems to be "don't set detail level so stupidly high that it exceeds card's vram". The only result that seems anomalous is one from Far Cry.

RX 460/550/560 also did have x8 link max, and it seems to be the new norm for mid-end cards from AMD.
One of the reasons is GPU link in their APUs - it is exactly x8. As they will be put together in many laptops, extra 8 lanes would be often wasted anyway.

Its been said many times, but VRAM demands have increased faster than core grunt demand. Processing power in consoles is a slow crawl, but their VRAM also exploded to 6GB and will be pushed further soon.

Its a simple 1+1=2.... even with playable settings you can exceed card VRAM.

Inb4 the number cruncher telling us its not true and games usually have enough with 3/4GB...