Question GDDR5 Dedicated VRAM versus DDR5 System RAM ?

[jnjnilson6]

I am perfectly aware that GDDR5 dedicated vRAM (first to have come with the Radeon HD 4870 in 2008) is greatly different from DDR5 system memory which has come out shortly onto the market.

I know my question may not be answered fully and entirely correctly, however, I would like to pose it nevertheless.

If we had a video card without dedicated memory which were to use 4 GB DDR5 memory from the system RAM and the exact same card, however, equipped with 4 GB GDDR5 dedicated memory of its own, which would outperform the other? And if anyone can provide more specific information, I would be glad to note it out.

This is a question which is hardly answerable, however, it is a wonderful expression of the vivid changes in the computer world and am sure has made exuberant more than a single mind.

All answers are greatly appreciated. Thank you!

 

[SkyNetRising]

If we had a video card without dedicated memory which were to use 4 GB DDR5 memory from the system RAM and the exact same card, however, equipped with 4 GB GDDR5 dedicated memory of its own, which would outperform the other?

Dedicated vram is faster than system ram.
This has always been so.

And system ram as vram is available on integrated GPUs only.
So - your proposed comparison is not really possible.

 

[jnjnilson6]

Dedicated vram is faster than system ram.
This has always been so.

And system ram as vram is available on integrated GPUs only.
So - your proposed comparison is not really possible.

Yes; that is true. The synonymity of the two GPUs was entirely hypothetical (and impracticable within the real world).

Thank you very much for the swift response and illumination!

 

[thestryker]

The basic differences between the two is that GDDR sacrifices latency for quad pumped clockspeed.

I believe the highest effective bandwidth GDDR5 (keep in mind this was derivative of DDR3) got to was 9 gbps per pin (32 pins for 32bit so around 36GB/s worth of bandwidth per chip).

DDR5 on the other hand only comes in 64bit memory modules so hypothetically speaking if we had DDR5-9000 it would have twice the bandwidth per module as that top GDDR5 would per chip. DDR5 also has lower latency so if all things were equal it would be better.

Where this falls apart is if you're trying to use the DDR5 on the system bus for three reasons:

1. The additional latency required to go from the DRAM to the GPU
2. Desktop parts are limited to 128bit bus
3. You'd now also be limited by the upper end of PCIe bandwidth

So even though the DDR5 latency is lower than GDDR5 the travel distance would completely negate this and then some.

Example of why the bus limitation matters: GTX 970 has a 256bit bus with GDDR5 running 7gbps which means it has an effective bandwidth of around 224GB/s. Even that hypothetical DDR5-9000 would cap out around 144GB/s.

PCIe 5.0 x16 carries ~64GB/s maximum bandwidth so as you can see this also hobbles any chance system memory could have.

Now for an IGP you can ignore the extra latency and PCIe limitation, but the system bus limitation isn't going anywhere, and compared to current cards the memory is much too slow. To put it in comparison to a current card that has a 128 bit bus the RTX 4060 has 17 gbps GDDR6 memory which means it has ~272GB/s of memory bandwidth. Unless AMD/Intel move to higher bus widths for consumer parts this will always hold back anything without dedicated memory.

 

[Eximo]

You can look at cards like the GT1030 DDR4 and GT1030 GDDR5 for a somewhat close comparison.

I wouldn't be surprised if DDR5 based budget GPUs don't start showing up relatively soon.

Going to be a mess though. So many companies mislabel GDDR5 as DDR5 that even Partpicker and Newegg already have filters for them that match up almost exclusively to GDDR5 cards

 

[Szymon Kubisiak]

You can't really compare DDR with GDDR on a GPU, as said above. But there is a board showing how GDDR performs on a CPU:

AMD 4700S Review: Defective PlayStation 5 Chips Resurrected

Break out the broken chips

www.tomshardware.com

 

[amcantech]

The RAM is physically closer to the compute unit (i.e. GPU) via GDDR in addition it is most likely to be binned at a higher speed and confirmed to be verified by the GPU manufactuturer

 

[Avro Arrow]

VRAM always outperforms system RAM. This isn't only because it's faster, it's also because there's a lot of latency involved in a video card using system RAM. System RAM is already slower, it has to be constantly allocated (because the system is always using it for other things) and it takes up some of the PCI-Express bandwidth that is otherwise used to communicate with the CPU. None of this is required for VRAM that is already on the card itself. This is why it causes stuttering and improper texture rendering.