• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

PCI-E question

Soldato
Joined
29 Feb 2012
Posts
2,875
Location
London
With AMD bringing support for PCI-e 4 next year how much improvement will be over 3?

With 3 giving 984.86 MB/s
And 4 giving 1969 MB/s

Worth holding out on building a new rig?
Or just crack on now....
 
I don't understand why AMD are in any hurry to get PCI-E 4.0 out as it is only the high end NVidia cards that can really benefit in mGPU setups.

To swamp PCI-E 3.0 with AMD cards would require 4 Vega 64 cards but then they don't support that anymore.
 
I don't understand why AMD are in any hurry to get PCI-E 4.0 out as it is only the high end NVidia cards that can really benefit in mGPU setups.

To swamp PCI-E 3.0 with AMD cards would require 4 Vega 64 cards but then they don't support that anymore.
Unless there is a cheeky driver update around the corner that enables quad crossfire for all games. :D Ohh hang on Im day dreaming again
 
Maybe AMD like progress. I'm not going to complain that they are pushing tech along.

I'm sure the real reasons behind it are more investor focussed, but if AMD can push the industry to more cores and faster interconnects and whatever else they are planning then more power to them. The fact that they are (so far) being reasonable on price is the icing on the cake.

I own Intel right now, but I want them to succeed because we all benefit. Going AMD may be in my future if they keep it up.
 
Might see some decrease in loading times in games - during actual gameplay itself PCI-e bus utilisation is typically fairly low - resource streaming (tiled resources, etc.) is slowly taking off which would put a bit more load on the system buses but a long way from even coming close to saturating a reasonably modern PC.
 
Very little to be gained in your average setup. If you have a godlike threadripper build with multiple GPUs and NVME drives then you’ll likely benefit, a bit.

However, who know what nVidia and AMD have up their sleeves for PCI-E 4.0. Next generation cards might utilise the extra bandwidth.
 
I don't understand why AMD are in any hurry to get PCI-E 4.0 out as it is only the high end NVidia cards that can really benefit in mGPU setups.

To swamp PCI-E 3.0 with AMD cards would require 4 Vega 64 cards but then they don't support that anymore.

Do you know anything about computing or do you just bury your head in graphics cards?

It is utterly ignorant of you to think AMD are "hurrying out" PCI-E 4.0 for just the benefit of graphics card users, even more so for those that just waste their time running benchmarks or playing games.

The Zen2 CPU design has PCI-E 4.0 baked in at the lowest level, to support more bandwidth via PCI-E devices and allow more devices to be connected to one system. TCO and PPW are two of the metrics that improve with higher bandwidth PCI-E as each and every system can do more in less space/with less power/at less cost. Consider how much the prospect of EPYC with 128 PCI-E 3.0 lanes changed servers, moving from the limited Intel's 44 lane layout, now consider the fact that PCI-E 4.0 makes that the equivalent of 256 PCI-E 3.0 lanes in bandwidth terms.

The ability to have several 100Gb/s links, along side a fully array of NVMe storage in a single socket server is what PCI-E 4.0 is about, or the ability to install (effectively) double the number of GPU's using in a ML/AI server.
 
I don't understand why AMD are in any hurry to get PCI-E 4.0 out as it is only the high end NVidia cards that can really benefit in mGPU setups.
You have to remember it's not always about GPUs, by Bringing PCI-E 4.0 motherboard support AMD will enable something of a trickle down bandwidth situation. I.E ATM a high end mainstream AMD board has two 3.0 x16 slots capable of x16/x8+x8 and a 2.0 x16 slot capable of x4, that means if you're running two GPUs then any NVMe drive you put in the third slot will be gimped. It also means that even if you only have One GPU you have to choose between gimping the SSD in the 2.0 slot or gimping the GPU by making the x16 slot x8. If it was two 4.0 slots and a 3.0 slot none of this would be an issue.
 
Do you know anything about computing or do you just bury your head in graphics cards?

It is utterly ignorant of you to think AMD are "hurrying out" PCI-E 4.0 for just the benefit of graphics card users, even more so for those that just waste their time running benchmarks or playing games.

The Zen2 CPU design has PCI-E 4.0 baked in at the lowest level, to support more bandwidth via PCI-E devices and allow more devices to be connected to one system. TCO and PPW are two of the metrics that improve with higher bandwidth PCI-E as each and every system can do more in less space/with less power/at less cost. Consider how much the prospect of EPYC with 128 PCI-E 3.0 lanes changed servers, moving from the limited Intel's 44 lane layout, now consider the fact that PCI-E 4.0 makes that the equivalent of 256 PCI-E 3.0 lanes in bandwidth terms.

The ability to have several 100Gb/s links, along side a fully array of NVMe storage in a single socket server is what PCI-E 4.0 is about, or the ability to install (effectively) double the number of GPU's using in a ML/AI server.

Are you capable of making a point without being rude?

To answer your question though I am aware of what you have written above.

I was also probably one of the first people on these forums to use a PCI-E solid state drive but gave up on them many years ago due to the fact that they use PCI-E lanes I would rather went to GPUs.

Below is a pic of one of my old builds using a PCI-E SSD

2EW1czi.jpg
 
Last edited:
You have to remember it's not always about GPUs, by Bringing PCI-E 4.0 motherboard support AMD will enable something of a trickle down bandwidth situation. I.E ATM a high end mainstream AMD board has two 3.0 x16 slots capable of x16/x8+x8 and a 2.0 x16 slot capable of x4, that means if you're running two GPUs then any NVMe drive you put in the third slot will be gimped. It also means that even if you only have One GPU you have to choose between gimping the SSD in the 2.0 slot or gimping the GPU by making the x16 slot x8. If it was two 4.0 slots and a 3.0 slot none of this would be an issue.

To be fair, multi GPU and and multi NVME AMD users are more likely to be using Threadripper, where it's not an issue at all. But yes, it will add flexibility to the mainstream platform.
 
To be fair, multi GPU and and multi NVME AMD users are more likely to be using Threadripper, where it's not an issue at all.
Perhaps, though if you don't need more than 8 cores then it's a bit of a waste to pay the extra for TR just to get another PCI-E 3.0 slot, however many of us bought our AMD systems before Threadripper was announced so it wasn't even an option anyway.

I'm actually in the odd position where I have an RTX2080 in the first x16 PCI-E 3.0 slot, a 400GB NVMe drive in the second PCI-E 3.0 x16 slot and a 10GbE NIC in the PCI-E 2.0 x16 slot. So if I want to do any benchmarking I have to pull the NIC and move the SSD to slot three so the GPU can run at x16 lol.
 
Are you capable of making a point without being rude?

To answer your question though I am aware of what you have written above.

I was also probably one of the first people on these forums to use a PCI-E solid state drive but gave up on them many years ago due to the fact that they use PCI-E lanes I would rather went to GPUs.

I'd hardly say using the words ignorant, and head buried rude. You say you are aware of the information, but then you were the one who asked why AMD are "rushing out" PCI-E 4.0 therefore making your entire original post pointless, since you are aware it is not solely to do with graphics cards.
 
Perhaps, though if you don't need more than 8 cores then it's a bit of a waste to pay the extra for TR just to get another PCI-E 3.0 slot, however many of us bought our AMD systems before Threadripper was announced so it wasn't even an option anyway.

I'm actually in the odd position where I have an RTX2080 in the first x16 PCI-E 3.0 slot, a 400GB NVMe drive in the second PCI-E 3.0 x16 slot and a 10GbE NIC in the PCI-E 2.0 x16 slot. So if I want to do any benchmarking I have to pull the NIC and move the SSD to slot three so the GPU can run at x16 lol.

Sounds like a very peculiar setup you have :p Expensive GPU, NVME, 10Gb network with I assume an expensive NAS, yet build isn't based on HEDT?

With NVME becoming more popular, PCI-E 4.0 will give the flexibility. But would still expect most people that require the PCI-E lanes will likely have a HEDT setup.
 
Back
Top Bottom