BASED AMD TO BRING OUT PCIE 4.0!!!

How is AMD offering newer standards suddenly making your 1600X obsolete? That's like complaining that DDR5 will soon be a reality, so your DDR4 is now worthless.

For fucks sake I'd bet money you wouldn't use pcie 4.0 to it's fullest extent anyway.

>what are NVMe SSD at 8gig/s

This has been known for a while.
Zen2 and Vega20 both use pcie4 will communicate with IF via the pcie phy as they use multipurpose io blocks now.

I'm hoping pcie4 will bring something more budget for nvme drives. Having four lanes is a waste in most cases and there needs to be a slower variant to replace sata3 as sata4 is dead. Hope they just use sata cables for 1 pcie lane or something.

Jow Forums apparently ate my post, unless I just had some weird cookie issue where my post isn't showing despite multiple refreshes.
Let me try again:

Previously PCI-E 4.0 was slated for Zen3.
This appearing in Zen2 would be a fairly big surprise. It would also likely hint at their new 7nm GPUs supporting the new standard.

>PCIe4
>When GPUs are nowhere near maxing out PCIe3
>When NVME drives are nowhere near maxing out PCIe3

it will support it, you just won't get all the features. This is known.

If you use more than a single device the bandwidth matters.

Literally the exact same fucking thing happened to your old PC.
>Sandy Bridge has PCIe 2.0
>Ivy Bridge adds PCIe 3.0
>Ivy Bridge chips work on old Sandy Bridge boards despite this

Uhhh guys what's the most advanced graphics card my Q9550 can run without stuttering?

IBM POWER9 already supported this.

how is this relevant

It was known like 2 years ago though

This is the final piece that will make the chiplets + IO chip concept work in Rome. IF at PCIe4 speeds? Yes please.

YO THIS HEEBS FINNA BOUTTA GET DABBED ON

Attached: 1529710649435.gif (554x400, 135K)

Placebo i imagine the drives are already so fast

When is Radeon group even going to do something again?
It's been way too quiet over there lately.

Attached: 173319975195.jpg (396x628, 113K)

you could have had PCIe 4 and CPU-GPU NVLink if you weren't a gaymer faggot

Im about to pull the trigger on one of the AMD based Thinkpads, seems like a good deal

So I guess the new zen CPUs won't be that much of an upgrade, so they bring something new on platform to give some kind of an incentive to upgrade?

I really hope it ain't so, I wanted zen 2 to rock.

Can't boot win7, no lpddr support, still inferior to 7nm zen2.

It would mainly be a Linux box and maybe dual boot Win10

That's the weird thing about AMD, they always seem to focus on improving the wrong things.
They'll give you PCIe 4.0, high bandwidth memory, 64 lanes on their CPU's, liquid cooled reference models with 3xPEG8... but when it comes to actual performance they'll still fall behind Nvidia's 1000 cores and 128-bit memory.

It's like they already want to be prepared for the day they'll release that behemoth of a GPU with 10 times the power of Vega 64. You know, just in case.

Attached: 336498663465.png (565x541, 157K)

See

I don't want to replace a laptop in a year so I'll wait for zen2 apus.
>dual boot
No thank you
>Win10
No thank you
I'll rather use a win7 virtual machine on linux host if I need windows

Infinity Fabric on Navi would be a devastating blow to Nvidia so not gonna happen anytime soon.

Attached: AMD slide.png (1588x979, 285K)

Remember when AMD ignored PCIe 3.0 and inturds were mentioning it every time even though neither GPUs nor drives got even remotely close to PCIe 2.0 speeds. Now when both GPUs and drives get bottlenecked by PCIe 3.0 suddenly nobody needs PCIe 4.0. Inturd logic.

such dissapoint

Attached: Untitled-1.jpg (1920x1200, 625K)

>still renders slower much slower than a dual core laptop running final cut pro

>GPUs get bottlenecked by PCIe 3

>ryzen is so based it even has a matrix slowdown mode
based amd

does PCIe4 promise better latency?

>separating the logic with an interstice would be beneficial in a GPU
Ehh, not sure about that.

Too bad it will not double the speed as 2.0 and 3.0 did. But I hope it adds some cool features.

>!!!

If the workload split happens all in hardware instead of needing to write drivers to do it it can work really well. For PC aplications other than games it will be awesome, for game consoles it will be flawless, but for DX11/12 it will struggle because fuck Microsoft.

The better potential for binning and improving yields would offset the performance and cost differences from using an interposer or more advanced active silicon connections.

PCIe latency is already a non-issue. It's so low you can only measure it if you're trying to.

What a profoundly retarded statement
and another

Its not only about speeds its about lanes if PCIE 4.0 has more bandwidth it can distribute that bandwidth to more devices currently unless you have x399 or z399 your going to run out of lanes trying to run more than one 16x 3.0 gpu and one 4x3.0 NVME due to all the other devices like SATA USB 3.1 etc also using PCIE lanes

Peak IOPS and latency will improve on a pcie 4 spec. Many high performing SSD's are 8x wired. These could now fit on 4x or get updated and receive extra performance for the same interface.

PCIe 4 is nice and all but not really necessary for most.
The only things that really need it are SSDs, and even then the only time you'd need it is if you read gigantic files.
We don't currently need PCIe 4 for GPUs. Maybe the next gen 7nm stuff will start to strain the limits.

>Its

>PCIe latency is already a non-issue. [Citation Needed]

>200ns is an issue for me
buying consumer hardware for anything that 200ns would make a difference in

>and latency
No, that's down to the NVME protocol. SSDs will be just as slow to respond.

Attached: 1379591566.png (656x258, 17K)

Great posts loser

no one should be satisfied until it drops to below 60ns.
We get bottlenecked on inference tasks in machine learning simply because it takes more time to transfer the data than it does to compute it.

>tfw so poor you can only afford to send 15 characters in your POST request

>We get bottlenecked on inference tasks in machine learning simply because it takes more time to transfer the data than it does to compute it.
That's true at all. But when it becomes true, PCIe latency won't be the bottleneck.

I'm curious, where did you get 60ns from?

>That's true at all
Meant not.

>I'm curious, where did you get 60ns from?

Not user but 60ns looks to be realistic if pcie design/implementations ever look at latency.

AMD's implementation is likely going to be another regression in latency

All technological improvement always sacrifices latency. Only later do past devices get actual improvements. Pretty much a law due to the speed of light, and the nonlinear increase in signalling paths versus bandwith

>that's not true at all
Yes it is
The exact paper escapes me but a researcher noted a drop in performance on inference when running on a GPU as opposed to running on a CPU simply because of the data transfer latency over PCIe

>60ns
The time it takes to load data into a register from L3 cache.
Incidentally DRAM takes ~100ns.

this is the end, intel is finished.

Attached: intell btfo.jpg (638x599, 123K)

Not if we bribe physics, goy

AMD could focus on making actual GPUs.
Ah ah ah. No.
Let's work on meme PCIe 4 instead

>The only things that really need it are SSDs, and even then the only time you'd need it is if you read gigantic files.
>What are BDMV in rar files

>GPUs are massively parallel and parallelizable
>stick 4 GPUs side-by-side connected in-package
>then connect the boards with XGMI, NVlink equivalent
>not beneficial
Nvidiot detected

gen2 PCIe is 5.0 GT/s
gen3 is 8.0 GT/s
gen4 is (I believe) 15 GT/s

But why was it a disappointment, user?

AMDs makes a ton of pipecleaner and test products. Some make it to market (4770), others don't (Tiran)

AMDrones harp on Intelfags on having no upgrade path...when Intelfags stick to their CPU for at least 4-5 years while AMDrones buy the newest AMD CPU every year, sometimes they buy the newest mobo as well

Take a look at the Gen 1 Ryzenfags, they all bought Ryzen 2 and will be buying Ryzen 3...The More You Buy, The More You Save -AMD

>implying intel fags did not do this
>implying zen 1 doesn't work on other zen mb.
>intel shills still mad about pcie lanes

Nvidia need developer NvLink for low latency and memory coherent for HPC.

AMD only could do APU on chips because latency PCIE even rumors suggest new EPYC use new interconnection for Vega 7nm

>implying intel fags did not do this
Except no one did.

Those who bought Nehalem didn't buy Sandy Bridge, they most likely upgraded to Haswell or Devil's Canyon. Those who bought Sandy Bridge upgraded to Devil's Canyon or Skylake.

I am not an idiot.
I bought gen 1 Ryzen and I do not plan on upgrading until Ryzen 5 - 7. You really do not have to upgrade your PC internals every year.

are you retarded? there are threads archived about people moving to the latest thing on Intel all the time, changing sockets.

It benchmarked like an old haswell cpu

the best SSD for sequential read is the 970 Pro which maxes out around 3.5 GB/s. Will the next generation of PCIe improve that or no?

I didn't think they were actually bottlenecked.

You can get PCIe 4.0 on Talos II already

>only 7
lmao amdrones ara pathetic

Much rather have one than the joke that was the 7700k

/v/ mostly games on ps4 and switch. If you try to make a pc gaming thread on /v/, it actually gets deleted because PC's aren't video games like the switch or ps4 are.

>tfw gen 1 and intend to upgrade to gen3
i need the performance and higher clocks mang, my 1700 clocks like shit

Here's why I'm not at all impressed with PCIe4:
a) PCIe3 is fast enough but there really isn't enough of it. The Ryzen CPUs have 20 lanes, Ryzen APUs have 12. Yet the AM4 ATX boards come with 3 x16 slots.. which can't actually be used for anything.
b) A 1080ti can't saturate PCIe3 x8. It's enough. The 2080ti does run marginally slower at x8, it's the first card to do so. There's still plenty of headroom up to x16.

>Finally motherboards can support 7 M.2 sockets!!!
What would PCIe4 have to do with that ability? PCIe M2s use x1 or x4. If we assume x4 then it's just a matter of having 28 PCIe3 lanes available. PCIe4 would let each individual M2 run faster at speeds that aren't possible with todays NAND anyway. A Ryzen CPU with PCIe4 and 20 lanes where 16 is assigned to the first x16 GPU slot would leave you with 4 lanes for just one M2 device - assuming you have no other PCIe devices.

The very limited amount of lanes on "consumer" CPUs is a bigger problem than the speed of those lanes.

shintels will increase pcie lanes to catch up to amd like the times ofdual cores and zen

Attached: 1512831367474.gif (250x250, 1.89M)

this is good for threadripper cuz it offers 60 lanes

>product optimized for garbage overheating dual core laptops doesn't scale well on to high thread count cpus
Sounds about right

A chiplet GPU will suffer absolute latency (and thus effective bandwidth will be much lower than theoretical) from splitting VRAM allocation.
Either the chips will need to crosstalk i.e. split VRAM or a central RAM pool will need to run through a master I/O controller.

In either case latency is added to memory operations and I do not have much faith in AMD's team pulling out some magic to make a multi-GPU card functionally transparent with no drawbacks.

Gen3 tops out at .985GB/s per lane, and 3.5GB/s from an x4 interface is about the best you're going to get because theoretical =/= reality

Latency is inevitable if you want to scale up efficiently for both CPUs and GPUs. Monolithic garbage is done for.

>latency is good if mama su says so

not him but its pretty much inevitable
12k resolution 60 fps but with 5 ms delay vs 2k resolution 300 fps but with 1 ms delay
speed of light is a bitch

YEAH LET'S MAKE 900mm2 MONOLITHIC DIES BECAUSE MUH LATENCY IS BAED BETTER MAKE COLOSSAL ABORTIONS INSTEAD

What are the advantages to 3.0?

m.2 mice fucking when

>Making this post when big af chips like TR and Epyc
Uhh

That's why PCIe 4.0 is great. You can split 16 PCIe 4.0 lanes into 32 PCIe 3.0 lanes without performance loss, for example.

>ayymdrone melt down because borg queen su was attacked
Priceless.

Yeah 8x PCIe 3.0 bottlenecks latest cards and we didn't even enter into GDDR6 stage yet. It's not like CPUs these days offer enough lanes to supply peripherals and still have 16 spare lanes for GPUs.

>AMD shills keep bragging about no motherboard change a year
>AMD has released two motherboard generations >any change to PCIE 4.0 and DDR5 would require a totally new motherboard
>everyone who got conned into buying Gen 1 and Gen 2 Ryzen has to fork over shekels for a new motherboard

A motherboard change a year keeps the AMDrones in fear!

SHILL HARDER RAJ AMD IS STILL SELLING LIKE HOTCAKES

WE NEED DIVERSITY AND WHITE GENOCIDE

Attached: 1541163543185.png (1800x850, 225K)