Maximum Video Card for PCI-E 1.0: Do X2 cards have an effect on bandwidth?

[cbn]

http://alienbabeltech.com/main/?p=2249&all=1

In this article it looks like 4870 512MB shows no preference with regard to PCI-E 1.0 or 2.0 but 4870x2 does.

Is there something about these X2 cards that hogs up more PCI-E bandwidth than an equivalent performance single card would?

I am asking because I am hoping that HD5870 will still be able to work with my PCI-E 1.0 motherboard efficiently. Speculation now is that it will have 1200 shaders @900 Mhz which would approximate a 4870x2...but as a single GPU(not internal crossfire)

 

[Sylvanas]

X2 cards require more bandwidth because there is twice the traffic over the one PCI-E bus than that of a single GPU. When crossfiring two separate cards, each GPU has it's own 16x link to communicate- it is a little more cramped with X2 cards. We have no idea how the next gen cards will turn out and if they will be bandwidth limited on PCI-E 1.0. IMO I would say bandwidth limitations are the least of your troubles if you want to be using a next gen card in a 1.0 board- because most 1.0 boards are a little dated you'd probably be CPU bottlenecked before anything else. I'm sure they'll probably work fine in 1.0 boards, just you won't be maximising the potential of the GPU.

 

[cbn]

http://www.dvhardware.net/article34936.html

For example, In this table here 4870x2 is listed @ 2.4 TFLOPs while HD5870 is listed @ 2.16 TFLOPs.....but the big difference is that one is internal Crossfire dual GPU and the other (HD5870) is single GPU.

 

[cbn]

Originally posted by: Sylvanas
X2 cards require more bandwidth because there is twice the traffic over the one PCI-E bus than that of a single GPU. When crossfiring two separate cards, each GPU has it's own 16x link to communicate- it is a little more cramped with X2 cards. We have no idea how the next gen cards will turn out and if they will be bandwidth limited on PCI-E 1.0. IMO I would say bandwidth limitations are the least of your troubles if you want to be using a next gen card in a 1.0 board- because most 1.0 boards are a little dated you'd probably be CPU bottlenecked before anything else. I'm sure they'll probably work fine in 1.0 boards, just you won't be maximising the potential of the GPU.

I hear you about the issue of CPU bottleneck (The best processor I can install is a Q6600 quad core).

I am just wondering if HD5870 would work more efficiently with a PCI-E 1.0 slot than HD4870x2....even though they are both roughly the same total speed.

 

[Sylvanas]

Originally posted by: Just learning

Originally posted by: Sylvanas
X2 cards require more bandwidth because there is twice the traffic over the one PCI-E bus than that of a single GPU. When crossfiring two separate cards, each GPU has it's own 16x link to communicate- it is a little more cramped with X2 cards. We have no idea how the next gen cards will turn out and if they will be bandwidth limited on PCI-E 1.0. IMO I would say bandwidth limitations are the least of your troubles if you want to be using a next gen card in a 1.0 board- because most 1.0 boards are a little dated you'd probably be CPU bottlenecked before anything else. I'm sure they'll probably work fine in 1.0 boards, just you won't be maximising the potential of the GPU.

I hear you about the issue of CPU bottleneck (The best processor I can install is a Q6600 quad core).

I am just wondering if HD5870 would work more efficiently with a PCI-E 1.0 slot than HD4870x2....even though they are both roughly the same total speed.

We don't know the answer to that yet, all we can do is speculate until launch day . Remember you can always increase the PCI-E frequency to extract that little bit more bandwidth should you end up with a card that may be limited in some instances- that's what I'd do. I wouldn't be surprised if next gen cards max out PCI-E 1.0 .

 

[error8]

Originally posted by: Sylvanas

Remember you can always increase the PCI-E frequency to extract that little bit more bandwidth should you end up with a card that may be limited in some instances- that's what I'd do.

Increasing PCI-E frequency is quite dangerous, since it's also linked with other components on your motherboard and you can get hard drive data corruption and some other goodies. It's better to just get another motherboard and that's all.

 

[Sylvanas]

Originally posted by: error8

Originally posted by: Sylvanas

Remember you can always increase the PCI-E frequency to extract that little bit more bandwidth should you end up with a card that may be limited in some instances- that's what I'd do.

Increasing PCI-E frequency is quite dangerous, since it's also linked with other components on your motherboard and you can get hard drive data corruption and some other goodies. It's better to just get another motherboard and that's all.

This is true, but when getting a new mobo is not an option and you are limited it can be a good stop gap, note I said that's what I'd do- can't speak for the OP . Generally, people have reported corruption of SATA drives over 125mhz PCI-E, that high is quite ridiculous, and 100-110 seem to be 'safer' I have been running at 115 for a few years now, no problem. It is still overclocking though so YMMV.

 

[Schmide]

I didn't see much of a difference, a couple frames here or there for the 4870x2. I think 5% was the maximum deviation and that was at 160-170fps.

If you're worried about a couple FPS average, your priorities are mixed up.

Edit (Ok Crysis did show a 20% change) Still that's Crysis. Enough said.

 

[apoppin]

Originally posted by: Sylvanas

Originally posted by: Just learning

Originally posted by: Sylvanas
X2 cards require more bandwidth because there is twice the traffic over the one PCI-E bus than that of a single GPU. When crossfiring two separate cards, each GPU has it's own 16x link to communicate- it is a little more cramped with X2 cards. We have no idea how the next gen cards will turn out and if they will be bandwidth limited on PCI-E 1.0. IMO I would say bandwidth limitations are the least of your troubles if you want to be using a next gen card in a 1.0 board- because most 1.0 boards are a little dated you'd probably be CPU bottlenecked before anything else. I'm sure they'll probably work fine in 1.0 boards, just you won't be maximising the potential of the GPU.

I hear you about the issue of CPU bottleneck (The best processor I can install is a Q6600 quad core).

I am just wondering if HD5870 would work more efficiently with a PCI-E 1.0 slot than HD4870x2....even though they are both roughly the same total speed.

We don't know the answer to that yet, all we can do is speculate until launch day . Remember you can always increase the PCI-E frequency to extract that little bit more bandwidth should you end up with a card that may be limited in some instances- that's what I'd do. I wouldn't be surprised if next gen cards max out PCI-E 1.0 .

Well, if you look at the review i wrote that linked to in the OP, even a GTX280 strains the PCIe 1.1 or 1.0 spec

if you are gonna spend $400 on an X2, you really ought to consider upgrading your MB also
- heck, i would say 4890 already maxes out the PCIe 1.1 spec; mine runs fine at 955/1020

 

[cbn]

In those tests (I linked) are they doing tri-fire with the P35 via a 4870x2 mixed in with a single 4870 512 MB (in the PCI-E x4 slot)?

If so we can definitely say 4870 is exceeding PCI-E (1.0) x4 by a pretty good amount due to the fact performance is being seriously hampered.

This begs the question how much bandwidth is 4870x2 using?

 

[apoppin]

Originally posted by: Just learning
In those tests (I linked) are they doing tri-fire with the P35 via a 4870x2 mixed in with a single 4870 512 MB (in the PCI-E x4 slot)?

If so we can definitely say 4870 is exceeding PCI-E (1.0) x4 by a pretty good amount due to the fact performance is being seriously hampered.

This begs the question how much bandwidth is 4870x2 using?

we tested CrossFireX-3 both ways

* Nvidia GeForce GTX280 (1 GB, nVidia reference clocks) by BFGTech
* ATi Radeon 4870 (512 MB, reference clocks) by Sapphire
* ATi Radeon 4870 (1GB, reference clocks) by ASUS
* ATi Radeon 4870X2 (2 GB, reference clocks) by VisionTek

X-2 is using more bandwidth than PCIE 1.0/1.1 spec provides; even GTX280 shows this in some situations

- that is the reason for the 2.0 spec; isn't there a 3.0 spec coming later on this year?
. . . i believe it will carry a bit rate of 8 gigatransfers per second compared to the current 5 per second of the 2.0 spec.

 

[JASTECH]

apopin, 8 gigatransfers p/s is getting close to the Holograms isn't it? I will be watching this thread as I have a EVGA 8800GTS 640 that my get replaced and trying to figure out what is the best way to go. I will be using an EVGA "Classified" and a EVGA x3 SLI for 2 separate systems. 1 or the x2 I don't know...Nvidia or AMD cards? Both made in Taiwan so what will get through the gates and win the Graphics Ball Run? This my friend we sit like a chick in the nest waiting for that worm!

Thanks, JASTECH

 

[apoppin]

Originally posted by: JASTECH
apopin, 8 gigatransfers p/s is getting close to the Holograms isn't it? I will be watching this thread as I have a EVGA 8800GTS 640 that my get replaced and trying to figure out what is the best way to go. I will be using an EVGA "Classified" and a EVGA x3 SLI for 2 separate systems. 1 or the x2 I don't know...Nvidia or AMD cards? Both made in Taiwan so what will get through the gates and win the Graphics Ball Run? This my friend we sit like a chick in the nest waiting for that worm!

Thanks, JASTECH

this is kinda old:
http://en.wikipedia.org/wiki/PCI_Express

this is the last news we heard about August time frame:
http://www.engadget.com/2008/0...s-3-0-details-surface/

You cannot go wrong with the latest GPUs from *either* vendor - they are perfect for 16x10 and 19x12 resolutions
- 4890 when it Overclocks is awesome and a great bang for buck; i am getting GTX275 and can give my impressions of it later on

you just have to answer the question
---PhysX or DX10.1 ??

if you don't really care - look for the best sale and the best warranty from a card by either vendor

 

[angry hampster]

Apoppin: How limited would I be trying to run crossfire 4870's on a P35-DS3L? The second PCI-E is 8x max.

 

[JASTECH]

apopin, .......PhysX or M$10? I use to like ATI because it was at least in North America, lol....I like Linux and Android with their Open Source Code to allow us to make what we want and or need instead of the Trojan OS of M$. Please let me know on the GTX275 as I need to replace a few systems that have water damage real bad and my Alpha Tech 2000 CFR UPS system.

Thanks, JASTECH

 

[apoppin]

Originally posted by: angry hampster
Apoppin: How limited would I be trying to run crossfire 4870's on a P35-DS3L? The second PCI-E is 8x max.

i don't know as i have not tested 8x+8x PCIe CF

However, from what i read you should be fine with 4870 CF in that configuration
- 4980 CF OC'd to the max, who knows? Probably still decent scaling.

Please let me know on the GTX275

i will have a review up next week on 4890OC vs GTX275 - both stock and overclocked
- using 4870, 4870-X2 and GTX280 [at stock clocks]

This weekend, i finally answer the question - Do you *really* need Vista64 over Vista32 for maxed out PC gaming
- the answer may surprise you . . .

. . . even though i am now running Vista 64 on my main PC now