PCIe 4.0 Will Still Deliver 75W of Slot Power

Subject: Graphics Cards, Motherboards | August 29, 2016 - 01:20 AM |
Tagged: pcie, PCI SIG

Last week, various outlets were reporting (incorrectly) that PCIe 4.0 would provide “at least 300W” through the slot. This would have been roughly equal to the power draw that a PCIe 3.0 GPU could provide with an extra six-pin and an extra eight-pin power connector, but do so all through the slot.

View Full Size

Later, the PCI-SIG contacted Tom's Hardware (and likely others) to say that this is not the case. The slot will still only provide 75W of power; any other power will still need to come from external connectors. The main advantage of the standard will be extra bandwidth, about double that of PCIe 3.0, not easing cable management or making it easier to design a graphics card (by making it harder to design a motherboard).

Video News


August 29, 2016 | 03:01 AM - Posted by Master Chen (not verified)

FFFFFFFFUUUUUUU...well, it was great while it lasted, I guess...

75W in 2020? Yep, into the trash it instantly goes...my dream of ever using anything on a mid-high/high-tiered level without any need to connect it to a PSU at all, that is. F*CK THIS SHHHHEEEEEEEEEETTT-T-T-T-T-T-T. >:((

August 29, 2016 | 09:30 AM - Posted by BlackDove (not verified)

Why would you need more than 75W from a PCI-E slot in a system so shitty that the PSU doesnt have a spare cable?

Or is one cable really that obtrusive?

August 29, 2016 | 10:51 AM - Posted by arbiter

Less you buy a sub 300watt psu which even then still has least 1 likely 2 6pin power. All modern psu's have them.

August 30, 2016 | 09:17 AM - Posted by Master Chen (not verified)

Nice try, but no cigar. I can bet that any of my currently owned (and regularly used) six builds, out of which three are ITX-based configurations, not even mentioning my GODLIKE i7 2600K main station, all can easily and absolutely effortlessly run circles around that Pentium/Athlon garbage you're sitting on there.

August 30, 2016 | 02:21 AM - Posted by ol1bit

Yes it was! Uncluttered MB/GPU setups...Sigh

August 30, 2016 | 09:23 AM - Posted by Master Chen (not verified)

A HBM2-based, XDMA-enabled, third party-made AND with the best custom cooler of all that are available, 150~220W mid-high/high tier Radeon...and all of that only through PCI-e...sigh...I guess a man can only dream...

August 29, 2016 | 04:08 AM - Posted by Anonymous (not verified)

The reference RX 480 is still non-compliant for the PCIE 4.0 spec. Haha.

August 29, 2016 | 09:14 AM - Posted by Anonymous (not verified)

Yes but the RX 480 has the about the same FP compute as a GTX 1070 and the RX 480 is clocked lower. 2 RX 480s at around $500 have the around the same compute as 1 Titan X(Pascal) at $1200! So there is the matter of $$$$ to Nvidia for any FP compute workloads. The RX 480 has a great amount of compute for around $250. RX 480 owners also have Vulkan/DX12 multi-GPU adaptor to look forward to as well as async-compute and better low latency gaming/VR gaming as the gaming titles become more optimized for Vulkan/DX12. So the RX 480 owners will get more for their dollars, while Nvidia owners will get gimped drivers and have to cough up more $$$$$ for the latest Nvidia KIT.

The big question is will Nvidia allow their GTX 1060 SKUs to utilize any Vulkan/DX12 multi-adaptor or will Nvidia gimp that also!

August 29, 2016 | 09:31 AM - Posted by BlackDove (not verified)

By FP do you mean SP FLOPS?

August 29, 2016 | 10:47 AM - Posted by arbiter

Compute isn't what most people look at when buy a card so saying it has as much means little as raw fps in a game. As you say vulkan/dx12 well depends on cpu if that turns in to an issue of holding the gpu back as 1 test has shown least in doom opengl/vulkan has shown. Pair a 480 with a lower end cpu it suffers a big performance hit where as a gtx106 yes did take an fps hit but wasn't no where near hit rx 480 seen.

August 29, 2016 | 02:49 PM - Posted by Anonymous (not verified)

Lots of people do non gaming rendering where FPS does not matter, and AMD's RX 480 has plenty of compute to accelerate Ray Tracing interaction calculations. So AMD's product will have more FP for that and cost less. Dual CPUs can be scaled fine for non gaming rendering with most graphics software able to utilize more than on GPU to get the rendering work done. These types of non gaming rendering workloads with Ray Tracing interaction calculated on the GPU's FPUs benefit from having the extra FP/Flops that AMD's GPU provide at the best flops/Dollar metric. So for the price of a GTX Titan X(Pascal) at $1200, a user could get twice the SP FP/Flops by get 4 RX 480s at around $1000(at $200 less than the Titan X/Pascal) for twice the SP/FP performance. AMD has also open sourced their FireRays software help with the task of accelerating Ray Tracing calculations on AMD's GCN based GPU SKUs.

Not all people even care about FPS, they care about the compute and the ability to accelerate Ray Tracing and other computations on the GPU! Also Vulkan/DX12 and any optimized Vulkan/DX12 gaming titles will be making more use of that extra compute that AMD provides, especially for VR gaming and low latency asynchronous compute done for VR gaming. AMD offers plenty of row compute and who cares about the little extra power usage, because that extra power usage comes with much higher compute/dollar value for those that need GPUs for more than gaming usage.

Nvidia sure like its high margin accelerator business for the HPC/Server markets, but AMD will be back in those markets also with Greenland/Vega on some very powerful APUs for the HPC/Server/Workstation markets. Let’s see Nvidia try and compete with AMD’s monster APUs on an interposer SKUs. AMD offer’s more compute on its Consumer SKUs and does not gimp its consumer SKUs to serve only the gaming market. Vega does not have to get here this year as AMD’s RX 480’s can be scaled at low costs simply by purchasing 2, or More, RX 480s for those that want a little more compute. Also there is nothing stopping AMD form making a larger Polaris SKU and who knows what AMD could do between now and when Vega arrives.

Enjoy your Nvidia Flagship Fapping with empty pockets! I’ll take the RX 480 for my rendering needs, with the affordable ability to get extra RX 480s and scale up to meet my rendering needs without breaking the bank!

August 29, 2016 | 02:50 PM - Posted by Anonymous (not verified)

Edit: Dual CPUs can be scaled
to: Dual GPUs can be scaled

August 29, 2016 | 04:29 AM - Posted by JohnGR

I think this is much better than having 300W of extra power going through the motherboard, when you can happily get that power from the pcie cables directly from the PSU.

August 29, 2016 | 11:31 AM - Posted by BlackDove (not verified)

Exactly. Why have all the extra load on small traces within a PCB when you can have it in actual cables?

August 29, 2016 | 08:44 PM - Posted by Anonymous (not verified)

This will also allow people with PCIe 3 motherboards buy PCIe 4 cards. Imagine having to build a whole new system just to get the GPU upped.

August 29, 2016 | 05:06 AM - Posted by Dee Gonsiewski (not verified)

I understood that when I read it, but if the cards still have the 6 and 8 pin sockets they will be compatible. I really doubt we will see this in consumer boards. It looks more like an optional feature specifically aimed at server applications with new connectors and stuff..

August 29, 2016 | 05:07 AM - Posted by Dee Gonsiewski (not verified)

Yeah, but that adds costs one way or another. This will inherently be more expensive than current solutions. I agree, they have already been doing similar with some boards by the addition of a Molex for extra power if needed.

August 29, 2016 | 05:33 AM - Posted by this_is_eric

It was an impossible premise to begin with. Motherboards are not going to have 1200 watts coursing through their veins to power quad slots.

August 29, 2016 | 07:07 AM - Posted by N. Grimm (not verified)

When I heard 300w to each PCIe connector through the motherboard, I thought to myself, who needs RGB, when your motherboard will have its own toaster red glow.

August 29, 2016 | 08:08 AM - Posted by razor512

High wattage over a PCB requires a different design in many cases. For example,thick traces with no solder masking so that additional solder can be added to them during them (this could be difficult if you are reflow soldering instead of wave soldering.

Wave soldering limits how densely you can pack components.

Even then, you will still end up with losses as compared to simply taking a power cable and connecting it directly from the PSU, to the video card.

August 29, 2016 | 10:49 AM - Posted by arbiter

Yea it would introduce whole new issues if they bumped power to 300 watts. As you said the thicker traces to support it or even more of them to spread it out, but also the electromagnetic interference of all that power going through the board would be another issue. Noticed a few boards use a separator to separate the small area of the board that holds the audio chip to limit interference.

August 29, 2016 | 09:33 AM - Posted by BlackDove (not verified)

People are acting like connecting one power cable from the PSU to the GPU is such a big deal.

August 29, 2016 | 10:40 AM - Posted by Anonymous (not verified)

Clowns copy / pasting each other's BS. Welcome to the internet "journalism". Why would you call people, verify any info before "making" an "article"? No, let's copy/paste unverified crap for instant content! Clowns.

August 29, 2016 | 11:32 AM - Posted by BlackDove (not verified)

I laughed out loud. Its so true.

August 29, 2016 | 03:04 PM - Posted by TeamShadowXGP

75w is not bad but they could of at least bump it up to 100w
at least it's not crazy High imagen 300w going to the Motherboard then that means you'll need another Connector to power up even more your motherboard i'm pretty sure they thought about other components on the Motherboard that needs power like CPU and other Devices or USB 3.1 "that can do 100w of power"

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

CAPTCHA
This question is for testing whether you are a human visitor and to prevent automated spam submissions.