AGP: 1996-2004 (8 years) PCI-E: 2004-present (11 years) When

>>51415942
We can't even max out pcie, why would we need to advance

>>51415949

So I have a reason to upgrade my mobo

Well we've gone through PCI-E 1, 2 and 3.
And nothing we have saturates PCI-E 2, let alone 3.

>>51415975
not on the consumer level maybe, but we regularly bump up against those numbers in data center applications and with compute based cards

>>51415965
>amd everyone

>>51416033
Yeah well, nVidia is working on something called NVLink for server applications. 2TB/s bandwidth or some shit.

For consumers, there's PCI-Express 4.0, which uses optical interconnects and is dozens of times faster.

fuck pci-e

when are we getting upgraded sata? thats been at the limit for a long time

>>51415942
PCI: 1992-present
Instead of ditching PCIE, why not just have x32? The only non GPU x16 cards I've seen in the wild are for networking and Intel's Phi. I currently have a PCI card in my computer, btw. There's no reason to ditch something that still works fine. That's Apple logic.

>>51416085
I guess they figure most people who need multiple large and fast hard dicks will use SAS.

>>51416085
sata is deprecated, NVMe is the new hotness

>>51416085
Sata-X/m2

>>51416181
>Sata-X
Whoops, sorry I meant SATA Express

>>51415942
>tfw the last GPU I ever bought was still AGP.

When PCIe 3.0 x16 (16GB/s) becomes obsolete which may be never.

>>51416051
Sam go to sleep

File: htxpcie.png (75KB, 975x438px) Image search: [Google]

75KB, 975x438px

HTX?

en.wikipedia.org/wiki/HyperTransport

Lurk

>>51416199

I think it's basically dead at this point though

>>51415965
PCIe has versions. With every version, it's double the bandwidth. PCIe 2.0 is twice as fast as PCIe 1.0, 3.0 is 2x as fast as 2.0, and 4.0 will be 2x as fast as 3.0. We are advancing, just with the same standard.

>>51415975
it's nicer to have faster pcie since jewtel limits our pcie lanes horribly

>>51417338
plus pcie is back and forwards compatible
well except for 4.0 which I think is backwards compatible with older cards, but 4.0 cards won't work in 3.0 slots

http://www.nvidia.com/object/nvlink.html

THANK YOU BASED NVIDIA

>>51417639
But does it work?

>>51417812
It is NVIDIA, so no.

>ITT we shill for nvlank
good job op, hope they are paying you well to "grab" home customers

>>51417312
Is it just me or do the overhead savings get completely negated by the low maximum packet size?

>>51416085
Uh, now.
What do you think m.2 is?
More important than that, ACPI needs to be updated for SSDs. That shit is OLD.

>>51417639
Nvlink is enterprise only for now

>>51418822
nope

find me a desktop motherboard that uses m.2

sata needs a new revision as backwards compatible compatible ports always succeed

>>51415942

>PCI-E 1.0
>PCI-E 1.1
>PCI-E 2.0
>PCI-E 2.1
>PCI-E 3.0
>PCI-E 3.1
>PCI-E 4.0

Because we don't need to.

>>51418081
Slow down there AMD, you'll get your turn.

>>51416230
This. Fucking GPUs still dont even max out pcie 2

>>51415942
PCI-E =/= PCI-Express

>>51415942
Why do we need to? We update the standard every few years and each iteration is twice as fast. Why reinvent the wheel?

>>51419897
Every not-cheap-ass-shit Intel motherboard from the last two generations

>>51418822
>what is NVMe

>>51415942
PCI-E still has 4.0 incoming and 32x and 64x link capabilities, the vast majority use 16x to x1
There is no reason to as there is plenty of headroom.

>>51415942
>I like having proprietary connectors replace industry standards becuz I wanna see kool stuff happen

GoyLINK 1.0a connector which changes slot connection every two years when?

>>51416230
Even then we will just switch to PCI Express 4.0 by then. You can actually run a Fury X in a PCI Express 2.0 x8 connection (4GB/s) with literally no bottleneck whatsoever. I can't imagine what in the name of fuck would benefit from 8GB/s transfer speeds let alone 16GB/s.

>>51421359
How the fuck does a fury x even use 4GB of bandwidth anyway?

>>51421769
It doesn't. GPUs need to transfer textures and data from games as quickly as possible. Maybe like 1GB to the Vram for example. This transfer only takes 1/4 of a second on a pci-express gen 2 x8 connection under ideal conditions.

However that transfer speed is not possible because HDD are limited to ~150MB/s, SSDs to ~500MB/s. This is why a RAM disks are the future. They can transfer data over 10GB/s (ddr3 @ 1600Mhz) if configured properly.

So at best GPUs are using 1-2GB of bandwidth on the pci-express connection. Storage drives are the bottleneck here not the pci-express connections themselves.

With all this said you can literally run a fury x on a pci-express gen 2 connection x4 and there would be no bottleneck even if used with a 500MB/s SSD.

>>51422072
>However that transfer speed is not possible because HDD are limited to ~150MB/s, SSDs to ~500MB/s. This is why a RAM disks are the future. They can transfer data over 10GB/s (ddr3 @ 1600Mhz) if configured properly.

Fucking retard. RAM disks aren't the future simply for the fact that it's volatile memory, and this volatility is what makes it possible to be so fast.

In other words, that's why we temporarily store shit in RAM when we need to and why loading times are a thing.

>>51422151
>Fucking retard. RAM disks aren't the future simply for the fact that it's volatile memory, and this volatility is what makes it possible to be so fast.
You can store ram disk images on an SSD or HDD to load up later or when you need to shutdown your PC.

>In other words, that's why we temporarily store shit in RAM when we need to and why loading times are a thing.
Loading times don't have to be a thing, they can be eliminated or at least reduced a fuckton.

Look we wouldn't be having this conversation if RAM manufacturers would stop being greedy jews and let use buy 64GB ram configs for less than $200.

File: Screenshot_2015-11-19-12-46-44.png (131KB, 540x960px) Image search: [Google]

131KB, 540x960px

>>51422264
So close, yet so far. *sighs*

>>51422376
Maybe prices could drop as demand for ramdisks increases. $500+ for 64GB of RAM is ridiculous, companies really are jewing us out.

>>51417621
Iirc PCI 4 and 3 will work but anything lower won't.

>>51418784
Yes.

File: 20151119_180004[1].jpg (3MB, 3264x1836px) Image search: [Google]

3MB, 3264x1836px

Fuck PCI-E.

>>51415949
PASCAL
A
S
C
A
L

>>51422510
>>/trash/

>>51422517
>>>/trash/
>implying people who like retro stuff should throw away their old stuff
no u

>>51422376
>tripfag
>asterisk action
>>>/Gaia/

>>51416085
If I remember correctly, SATA's hit a limit to how much its bandwidth can increase without cost penalties. It's not efficient to add more bandwidth and scale SATA3 into a hypothetical SATA4
http://www.tested.com/tech/pcs/456464-why-ssds-are-transitioning-sata-pcie-next-gen-form-factor/

>>51419897
>find me a desktop motherboard that uses m.2
I have one, and it's an ASRock AMD board. From 2014.