PCI Express 2.0 Nears Completion

[malware]

PCI Express, version 1.0 which arrived in 2003, is about to enter the next generation with version 2.0. PCI Express 2.0 brings a bevy of changes, starting with a speed boost, according to the PCI Special Interest Group (PCI-SIG) that governs the technology. The PCI-SIG has released a penultimate edition of the basic PCI Express 2.0 specification to its members in version for comments, and the final version is due after a 60-day comment period. The central feature of the base version of PCI Express 2.0 is a speed boost. It doubles each serial line's data transfer rate from 2.5 gigabits per second to 5Gpbs. But future enhancements also are in the works. One will support high-end graphics cards that slurp 225 or 300 watts of power. Finally, a longer-term effort, code-named Geneseo, will let coprocessor cards such as graphics or encryption accelerators be tightly connected to central processors.

View at TechPowerUp Main Site

 

[Pinchy]

Great...now when i upgrade to PCIe 1.0 (if that ever happens ) there will be still something i dont have

 

[Alec§taar]

Great...now when i upgrade to PCIe 1.0 (if that ever happens ) there will be still something i dont have

For nearly 15 years on for myself on using PC's, this has always been the case.

I have watched PC's grow from using busses like ISA, (never tried this -> MCA by IBM), to VLB (Vesa Local Bus), to PCI, & now to PCI-Express 1.0 & eventually this PCI-e 2.0...

* Naturally though, each time things did get better/faster!

(Still, if you wanted that better? Well, of course, You had to dig up some coins/deadpresidents first for it...)

APK

P.S.=> It never changes, & yet it always changes for the better, technology-wise... & I have bought newer/better ones over that period every 4 years or so! Heh, I remember telling my pal "I'll never need another PC - this 486 does the job!"... apk

 

[lemonadesoda]

ISA to EISA (double the bandwidth = 2x)

ISA to PCI (double the bandwidth = 2x)

PCI33 to PCI66 (double the bandwidth = 2x)

PCI66 to PCIX (double the bandwidth = 2x)

PCI to PCIe (same = 1.0x)

Oohps!!! PCI to PCIe was a format change. It is also SIGNIFICATNLY cheaper to manufacture. But performance wise there is no gain IN MOST real-life situtations. Some people have commented on worse latency (e.g. Creative Audigy) although I don't know the true facts behind that one.

(PCIe has a theoretical gain by allowing BOTH read and write without losing bandwidth in either direction. Whereas PCI must share the bandwidth. But for Joe Consumer, this point is moot, since what cards does Joe use that would gain from such a feature? GPUs... NO, Sound... NO, Internet... NO. The only gain is on bandwidth constrained LAN servers. But for them, PCIX is still a better solution).

I welcome PCIe 2.0. Its time to apply that 2x rule!

Rather unfortunate for add-in card OEMs. They'll want to implement 2.0 asap, but they will also need to be 1.0 compatible.

PCIe 2.0 is interesting from a graphics card standpoint. Since NO CARD TODAY is saturating the PCIe x16 connection, then a PCIe2 x8 (same bandwidth as todays x16) would be more than sufficient. This is a much small form factor (only have the slot length and therefore easier to implement in SFF). The circuitry of the controller chips is also much simpler, since many fewer lanes are required on a M/B for the graphics port. This also makes SLI/crossfire cheaper to implement.

****

My wish list.

I would like to see PCIe v2.1X. This would allow a greater "distance" between the controller chip and the device. (Appropriate screened cables, of course). Just in the same way that a SATA connector can go a meter, I would like to be able to take a PCIe line and take it somewhere OTHER than just the internal slots, e.g. an external port for external hardware, etc. or be able to mount a device somewhere else in the case.

 

[XooM]

eh, x16 is easy enough to saturate with an x1950XTX or a 7950GX2. Ever wonder why we need those compositing cables or SLI bridges? It's because there isn't enough room on the PCI-E bus to communicate the required material for frame composition over the x16 slot, though no such problem arises on lower-end cards, which is why lower-end cards ie x1600s or 7600s can SLI/crossfire without dongles. So yes, the PCI-E bus is entirely saturatable.

Also, no card *today* is saturating it. what about next-gen? 6800ultras may not have saturated AGP slots (damn close i suspect) but a 7800GTX would completely wash one out.

also, in high-end gfx applications, ie CGI or GPGPU tasks, PCI-E bandwidth is extremely important. Movie CGI comfortably saturates a full x16 slot, hence certain workstation boards having dual chipsets for dual x16 SLI before nvidia's SLI32x chipsets.

I call shens on the worse latency stuff until somebody reliable can come up with some numbers on it.

There are plenty of performance gains; you're just not a server

 

[15th Warlock]

One more reason for waiting to upgrade my good ol' trusty AGP box.

Expect one year from now: GeForce 8850GTX2X and Radeon X2950XTX, only available for mobos with PCIe 2.0@32X slots LOL!

P.S.: I heard Intel is the driving force behind this Geneseo thing, they want to use this platform to compete with AMD+Ati's HT3 platform. This will effectively turn GPUs into FP and Physics coprocessors, kinda like the 387 math coprocessor days, funny how old ideas tend to resurrect after so many years (SLI, Crossfire, and Multi-GPU cards being trends set years ago by 3dfx, to be abandoned and resurrected later, this being one clear example of what i mean

 

[Judas]

One more reason for waiting to upgrade my good ol' trusty AGP box.

Expect one year from now: GeForce 8850GTX2X and Radeon X2950XTX, only available for mobos with PCIe 2.0@32X slots LOL!

P.S.: I heard Intel is the driving force behind this Geneseo thing, they want to use this platform to compete with AMD+Ati's HT3 platform. This will effectively turn GPUs into FP and Physics coprocessors, kinda like the 387 math coprocessor days, funny how old ideas tend to resurrect after so many years (SLI, Crossfire, and Multi-GPU cards being trends set years ago by 3dfx, to be abandoned and resurrected later, this being one clear example of what i mean

Heh yes most likely dont think your too far off

 

[i_am_mustang_man]

i always assumed the sli bridge was two fold
-keep things proprietary
-less complex and not as dependent on motherboard

i don't think even a 7950gx2 saturates x16

 

[spectre440]

jesus, they haven't even begun to exhaust the bandwidth that PCI-E offers...
why release 2.0? i mean... what for? because 2.0 is better than 1.0? pfftt....

its exactly the same type of premature release of a new/improved interface that charachterized the release of the "first" PCI-E...

AGP wasnt anywhere near maxxed out, seeing as how the highest end cards of the day (6800 ultra's and the likes) didnt even begin to max out the bandwidth... but nnnnnoooooo.... they have to realse this! cause its NEW and it'll make them shitloads more money...

 

[Azn Tr14dZ]

i always assumed the sli bridge was two fold
-keep things proprietary
-less complex and not as dependent on motherboard

i don't think even a 7950gx2 saturates x16

7950GX2 takes up X6-X8 I believe, about half of it's full potential. And spectre440, that's the reason why they're releasing it, because it's better. Yeah, I think it's gay too...

 

[spectre440]

but what's the point of it being better (which it is), and offering more bandwidth (which it does) if the technology to ACTUALLY USE that bandwidth and those capabilities isnt even around yet? and the previous interface isnt even NEAR being maxxed out?

 

[Azn Tr14dZ]

Exactly...

Is it gonna be the same physically?

 

[i_am_mustang_man]

i'm hoping it's like agp8x and agp 4x - same form factor, just some cards "work better" on the newer one, but they will work just fine on the other ones (like 1-2% diff)

 

[15th Warlock]

Take the AGP version of the 7900GT, the Gainward Bliss 7800GS 512+, a card based on the G71, with the same 24 pipelines G71 core found in the PCIe 7900GT and 512MBs of RAM, it works just fine in the AGP bus without loosing the performance of its PCIe sibling. Why was Gainward the only manufacturer with big enough cohones to release cards with the G71 core for the AGP bus is beyond me, all of these cards sold out within weeks of their release.

Yes, I know only a handful of this cards actually hit the retail, and the ppl who bought them had to pay almost an arm and a leg to get them (about $500) just for the premium of having the bragging rights of owning the fastest AGP card around, but what better example of the fact that even today, AGP is as good as PCIe for high end cards, at least for gaming.

Don't get me wrong, I'm not saying that actually having more bandwidth at your diposal is a bad thing, I never believed in the "640kb of RAM should be enough for everybody" philosophy, it is just that forcing ppl to buy a wholy new platform just for the right of owning the latest high video card models like Intel, Ati and nVidia did when PCIe was released, doesn't look like a ethical bussiness practice to me.

But that's the way it has always been in the IT industry, with the big players forcing newer technology down our collective throats every 6~12 months, making the concept of future proof seem ethereal, and making systems worth thousands of dollars, become virtually obsolete faster than any other industry in the history of humankind.

That's my two cents.

 

[Track]

but what's the point of it being better (which it is), and offering more bandwidth (which it does) if the technology to ACTUALLY USE that bandwidth and those capabilities isnt even around yet? and the previous interface isnt even NEAR being maxxed out?

Your totally right. The highest end card today dosent even use half of the bandwidth that PCI-E (1.0) can handle.

Its not going to be 32 lanes (x32), but higher bandwidth to each lane, so that a PCI-E 2.0 would have twice the bandwidth of a PCI-E 1.0 lane. But again, if the 16 lanes arent being used in PCI-E 1.0 then were going to see a 4 lane usage of PCI-E 2.0.
Maybe this is so that PCI-E (2.0) x4 could run high end cards.. it could save money to manufacture, but still not worth creating a whole new interface.

The only thing that i can think of, that would have driven them to create PCI-E 2.0 is that somehow there would be a performance boost if each lane were to be able to handle more bandwidth so that it would be consitrated in 4 lanes meaning that the card didnt have to answer to 8 lanes every time, but only 4 and even though those 4 would have the same bandwidth, it would save time getting the information to the lanes and then forwards.. but im not keen on this subject.

 

[i_am_mustang_man]

Your totally right. The highest end card today dosent even use half of the bandwidth that PCI-E (1.0) can handle.

Its not going to be 32 lanes (x32), but higher bandwidth to each lane, so that a PCI-E 2.0 would have twice the bandwidth of a PCI-E 1.0 lane. But again, if the 16 lanes arent being used in PCI-E 1.0 then were going to see a 4 lane usage of PCI-E 2.0.
Maybe this is so that PCI-E (2.0) x4 could run high end cards.. it could save money to manufacture, but still not worth creating a whole new interface.

The only thing that i can think of, that would have driven them to create PCI-E 2.0 is that somehow there would be a performance boost if each lane were to be able to handle more bandwidth so that it would be consitrated in 4 lanes meaning that the card didnt have to answer to 8 lanes every time, but only 4 and even though those 4 would have the same bandwidth, it would save time getting the information to the lanes and then forwards.. but im not keen on this subject.

you made me think about 3xGPU for physics and what not

maybe they need to split it into 4 lanes? and this is to ensure everything still works dandily (?) at 8x, instead of 4x

 

[Track]

Take the AGP version of the 7900GT, the Gainward Bliss 7800GS 512+, a card based on the G71, with the same 24 pipelines G71 core found in the PCIe 7900GT and 512MBs of RAM, it works just fine in the AGP bus without loosing the performance of its PCIe sibling. Why was Gainward the only manufacturer with big enough cohones to release cards with the G71 core for the AGP bus is beyond me, all of these cards sold out within weeks of their release.

Yes, I know only a handful of this cards actually hit the retail, and the ppl who bought them had to pay almost an arm and a leg to get them (about $500) just for the premium of having the bragging rights of owning the fastest AGP card around, but what better example of the fact that even today, AGP is as good as PCIe for high end cards, at least for gaming.

Don't get me wrong, I'm not saying that actually having more bandwidth at your diposal is a bad thing, I never believed in the "640kb of RAM should be enough for everybody" philosophy, it is just that forcing ppl to buy a wholy new platform just for the right of owning the latest high video card models like Intel, Ati and nVidia did when PCIe was released, doesn't look like a ethical bussiness practice to me.

But that's the way it has always been in the IT industry, with the big players forcing newer technology down our collective throats every 6~12 months, making the concept of future proof seem ethereal, and making systems worth thousands of dollars, become virtually obsolete faster than any other industry in the history of humankind.

That's my two cents.

This industry is saturated in companies and organizations, all creating new hardware all of the time. What nVidia and ATI do is just try and catch up, both with whats new and with eachother. Maybe if someone were to govern the making of new interfaces by different groups then we wouldnt be seeing PCI-E 2.0 when PCI-E 1.0 isnt even half used. You see, they make these things because they dont look at the big picture, they each just create new things without thinking of the rest of the industry or the efficiency of mass-producing what they have invented. Makes perfect sence, its just like society - no one takes responsibility and everyone tries to get a leg up over the other.

The thing with AGP and PCI-E is that well.. they are the SAME thing. AGP x8 = PCI-E x8. This is known because PCI-E x16 has exactly twice the bandwidth of AGP x8.
Even now, in 2007, cards havnt even gotten to use 8 lanes yet. Only thing that comes close is the 7950 GX2, wich is actually 2 cards. If they had released PCI-E 1.0 right now, it would have made a lot more sence, performance usage wise, i dont know abt manufacturing price.

 

[XooM]

Where are you people pulling your arbitrary bandwidth usage numbers from? Heres an article, an oldy but a goodie (as much as i may dislike THG):
http://www.tomshardware.com/2004/11/22/sli_is_coming/index.html
which proves my point. it may be last-gen hardware, but its still a useful comparison. Performance at x16 is, across the board, always better than x8 or x4 or x2 or x1, but in games the difference is usually very small (ie <1 FPS). However, in enterprise applications, which is the only market that really matters, with apps such as maya and 3dsmax, every doubling of bandwidth yeilds a linear increase in performance.

...and those are last-gen's cards, much less next's.

 

[i_am_mustang_man]

that is probably one of the most interesting reads i've ever seen

thanks

 

[qwo]

PCIE1.0 isnt enough for the power supply today

nvidia 8xgpu need two 6pin cables?

with pcie 2.0 maybe it doesnt need extra 6 pin cables

 

[Ketxxx]

PCI-E 2 is all marketing hype from what I've seen. I dont see it bringing any significant performance increases, just like the movefrom PCI-E 8x > 16x.

 

[jocksteeluk]

no doubt some new card will start being made for pci express 2 real soon making board with pci express 1 out of date

 

[Pinchy]

Yep, i reckon its all to make the consumer to spend more money

 

[wazzledoozle]

I doubt any single card can saturate a x16 slot. Maybe the upcoming G80 and R600, but with x16 bandwidth up and down, its hard to see that much bandwidth being used.

 

[Azn Tr14dZ]

I doubt any single card can saturate a x16 slot. Maybe the upcoming G80 and R600, but with x16 bandwidth up and down, its hard to see that much bandwidth being used.

The current cards right now (7950GX2) can only take abour X6 of the X16...Idk why they do this.