VOGONS


First post, by Almoststew1990

User metadata
Rank Oldbie
Rank
Oldbie

I've got an ITX board for a project and it has a PCI-E 2.0 16 slot, but only has the pins to transfer at 4x rates (as advertised in the boards specs).

Now my question is my general rule of thumb has always been that PCI-E bandwidth is very rarely the limiting factor for graphics - running a high end card on previous gen slots doesn't make a huge difference (1 or 2 frames), or even running on the latest PCI-E gen at x8 (which I think gives the same throughput now that I think about it...). But PCI-E 2.0 4x speed is 2GB/s.

However, my very basic AMD 7350 graphics card apparently has a bandwidth, according to GPUz, of 14GB/s?! And it's a crappy 64 bit GDDR3 900MHz memory clock card?! So my memory bandwidth (and "getting told what to do from the CPU" bandwidth?) is significantly throttled or am I missing something here?!

Ryzen 3700X | 16GB 3600MHz RAM | AMD 6800XT | 2Tb NVME SSD | Windows 10
AMD DX2-80 | 16MB RAM | STB LIghtspeed 128 | AWE32 CT3910
I have a vacancy for a main Windows 98 PC

Reply 1 of 8, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

And it's a crappy 64 bit GDDR3 900MHz memory clock card?!

GeForce 4 Ti 4600 had around ~10Gb/s with just 128-bit 325 Mhz DDR1 memory.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 2 of 8, by Oetker

User metadata
Rank Oldbie
Rank
Oldbie

That's how it works, a GPU's RAM is much faster than the bus it's on. However most data, such as textures and meshes, will be uploaded to the GPU at a strategic time (loading a map) with only specific dynamic stuff being sent over the bus during gameplay. Especially effects that depend on reading back from the GPU can cripple framerates.

Reply 3 of 8, by Almoststew1990

User metadata
Rank Oldbie
Rank
Oldbie

So if the 4600ti has a max bandwidth of 10GB/s and AGP x8 can only do 2GB, would a PCI-E version of the 4600ti get significantly higher frame rates?

Ryzen 3700X | 16GB 3600MHz RAM | AMD 6800XT | 2Tb NVME SSD | Windows 10
AMD DX2-80 | 16MB RAM | STB LIghtspeed 128 | AWE32 CT3910
I have a vacancy for a main Windows 98 PC

Reply 5 of 8, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie
Almoststew1990 wrote on 2020-02-29, 10:30:

So if the 4600ti has a max bandwidth of 10GB/s and AGP x8 can only do 2GB, would a PCI-E version of the 4600ti get significantly higher frame rates?

Just to point out the obvious : GeForce 4600 Ti does NOT support AGP 8x - it's AGP 4x only card.
What GPU gets from CPU, is only the raw data and instuctions what to do with it.
The work itself (based on provided data) is done 100% internally, unless GPU runs out of VRAM (or something will require additional data from RAM).
Either way : PCIe speed won't be the main issue with your card.

Think of it this way :
In CPU-GPU, time needed to tell GPU what to do (including time needed to get the stuff it needs), is usually a lot less than ammount of time that it will need to complete given task (in short : if CPU is fast enough in giving direction, GPU isn't fast enough to keep up with it).
^this is basic summary of how CPU and GPU operate/interact.

157143230295.png

Reply 6 of 8, by ragnar-gd

User metadata
Rank Newbie
Rank
Newbie

See this very informative article:

https://www.techpowerup.com/review/nvidia-gtx … xpress-scaling/

My posts on the subject of W98SE legacy are dedicated to Rudolph Loew, (+11/ 2019, *1952), as without his work my builds would not be possible.

Reply 7 of 8, by ragnar-gd

User metadata
Rank Newbie
Rank
Newbie

In a nutshell, on average cards:
PCIe 2.0/3.0/4.0 16x delivers full performance (2.0 may drop 1% on newest generation)
x8 will see a drop of 2% to 6%. You will most likely not notice that.
x4, most probably only 2.0 on your MoBo, will drop between 6% to 14%. You will notice that on games with already low fps

Framerates will drop on x4.
Because rendering (among other things) takes place on the GPU, higher detail-setting will not affect this.

My posts on the subject of W98SE legacy are dedicated to Rudolph Loew, (+11/ 2019, *1952), as without his work my builds would not be possible.

Reply 8 of 8, by ynari

User metadata
Rank Member
Rank
Member

You'll be fine. Ages ago I did some comparison of PCI-e 1.0 speeds. 16x PCI-e 1.0 was faster than 4x PCI-e 1.0 but not /that/ faster, using an HD6950. PCI-e 2.0 4x is the same as PCI-e 1.0 8x so I doubt you'll even notice.

On the other hand, PCI-e 1.0 1x is slow as dirt (much slower than AGP 2x), gaming on it is painful.