AMD: Speculation, Rumors, and Discussion (Archive)

[CSI PC]

@Grall Tomshardware shows that PCIe cables will still be used to deliver power to the motherboard.

http://www.tomshardware.com/news/pcie-4.0-power-speed-express,32525.html

Thanks for the headsup showing how this is now implemented, so it means many cables to the motherboard rather than the cables to the GPU.
I suppose technically this removes some of the issue about distributing the power demand between mainboard PCIe and auxiliary, but beyond that not sure if that is much better solution from a physical routing perspective, I guess I need to see it populated.

If you look closely, there is a single 20-pin power connector like any other modern motherboard. There are also four 8-pin and two 6-pin power connectors. AMD marked the PCIe power connectors as follows:

P0 ABCD PWR
P1 ABCD PWR
P0 EFGH PWR
P1 EFGH PWR

These connectors are in addition to the processor 4+4 pin connector marked P0 and P1 CORE PWR. Even with a 64-core platform, that much power for the board and processors would severely limit AMD's chances of getting into data centers. We feel that AMD added the extra power for something else, like the PCIe edge connectors.

When we asked the PCI-SIG, we received the news that for the first time, PCIe will get a massive power increase at the connector. Solomon couldn't recall the exact ceiling because member companies have proposed several options. Solomon stated that the minimum would be 300W, but the ceiling “may be 400 or 500W."

Thanks.

 

[Grall]

How this affects the existing motherboard ATX power connector will be interesting, because it sounds like the PCIE connectors will no longer be powered through this, well would be a nightmare as this would also need re-designing and spec change if they remained using the ATX 24-pin connector.

We could use this opportunity to re-design ATX connector and drop all of the 3.3V and 5V supply lines, replace them with 12V, 16V or 24V supply voltage. That's legacy; what in a PC today runs on straight-up 5V supply voltage? Or even 3.3V. If you're a laptop, perhaps your HDD/optical drive motors/actuators, but I doubt there's a lot of actual electronics left today that runs at TTL signaling levels...

 

[Deleted member 13524]

I guess you never had issues then re-plugging anything into the motherboard when it is populated such as fan connectors/etc.
And as I said this now needs to be a 600W minimum connector on the motherboard and probably taking up same/just smaller room than the ATX24-pin.
Cheers

LOL I definitely have had issues with overpopulated cases. I have 2 R9 290X which need 4 PCIe power cables and 6 SATA devices that need another 12 cables, plus 8 fans inside so 8 cables more. I have 24 cables in that case plus the case wires for power, reset, front USB, etc.

I've had a lot of hassle by getting a case that lets the cables pass behind the board so it could get a decent airflow despite all those cables. Which makes me even more curious about how someone would prefer having to cross the PCIe power cables through the middle of the case instead of just converging them from the PSU to a single area in the same PCB.

And as I said this now needs to be a 600W minimum connector on the motherboard and probably taking up same/just smaller room than the ATX24-pin.
Cheers

No, it doesn't. Besides the standard 24pin connector, the motherboard can take a number of 6 and/or 8-pin connectors that will tell the system how much extra power in the 12V it can distribute, and the PCIe 4.0 graphics cards will tell the system how much power it need to function. If the graphics cards require more power than the amount available from the board, then the system won't boot, or stop at the UEFI with a warning, or won't do 3D mode or something like that.

That's legacy; what in a PC today runs on straight-up 5V supply voltage?

All USB and SATA devices.

Or even 3.3V.

Most smaller integrated circuits in a motherboard use 3.3V VDD, AFAIK (sound codecs, SATA controllers, BIOS ROMs, etc.).

 

[CSI PC]

No, it doesn't. Besides the standard 24pin connector, the motherboard can take a number of 6 and/or 8-pin connectors that will tell the system how much extra power in the 12V it can distribute, and the PCIe 4.0 graphics cards will tell the system how much power it need to function. If the graphics cards require more power than the amount available from the board, then the system won't boot, or stop at the UEFI with a warning, or won't do 3D mode or something like that.

Glad it is not just me with what seems clunky sized hands when dealing with populated motherboards

The standard 24-pin ATX connector is specified with only 2x12V pins rated at 6A each giving the connector a total of 144W (then you have the 5V and 3.3V), HCS raises this to I think 9A (216W) and HCS+ to 11A, but standard is what most IHV implement with upper ones HCS. - all numbers given are just for the 12V.
Anyway firstminion linked an article showing the PCIe power cables are just moved from the cards to the motherboard.

. If you look closely, there is a single 20-pin power connector like any other modern motherboard. There are also four 8-pin and two 6-pin power connectors.

http://www.tomshardware.com/news/pcie-4.0-power-speed-express,32525.html

If they did not do this they would have the issues I mentioned.
Cheers

 

[homerdog]

http://www.bit-tech.net/hardware/graphics/2016/08/23/deus-ex-mankind-divided-benchmarked/6

As expected the 480 4GB is way up at the top of the value chart, though there is quite a difference between it and the 8GB version even at 1080p. This game is a horribly unoptimized mess for both vendors so that might have something to do with it. Only the Titan PX can maintain 60fps at 1080p

The game does seem to favor AMD cards but since no AMD cards run the game well I don't see that as a big win for the red team.

 

[CarstenS]

As expected:
http://www.tomshardware.com/news/pcie-4.0-power-speed-express,32525.html
„8/24/2016 Update:
We learned from PCI-SIG today that the power increase for PCI Express 4.0 will come from secondary connectors and not from the slot directly.“

 

[AlNom]

Time for a second 24-pin design?

 

[Anarchist4000]

They need to come up with something new and interesting just because of all these powered peripherals coming. Even a simple USB card powering multiple USB3.1 hubs probably needs a connector.

 

[Silent_Buddha]

The standard of power have not been decided yet, thats why its vague. Diferent vendors have diferent opinions from 300 to 500.
----------

But in the end doesn't really matter, 300W is enough to supply a Titan X...So the number of GPUs with External connectors will be very limited. It will also make GPU and specially custom coolers much simpler to design and allow for better designs as well. Also will help users with low end and small cases.

This will make the building a PC better is every way.

It's not going to affect the consumer market for at least the next 2-5 years. AIBs will still have to make cards compatible with older systems that are limited to PCIE 3.0 during that timeframe. Basically that means that power cables will still be required for consumer graphics cards.

PCIE power changes will have the most immediate impact in the server market where a high ASP can make up for the limited market of PCIE 4.0 supporting mainboards.

For consumer application, the most immediate impact of PCIE 4.0 is likely going to be internal storage (M.2 4xPCIE 3.0 can already be saturated), external storage (you can already saturate eSATA much less USB 3.0) and 10 Gig Ethernet (including NAS) will benefit as it'll become more cost effective to bring out a consumer facing solution.

There will also be the potential for external PCIE 4.0 attached graphics solutions. But it remains to be seen how compelling that is to consumers.

Regards,
SB

 

[Grall]

For consumer application, the most immediate impact of PCIE 4.0 is likely going to be internal storage (M.2 4xPCIE 3.0 can already be saturated), external storage (you can already saturate eSATA much less USB 3.0)

You actually can't saturate* M.2, and you wouldn't be able to saturate SATA either* if you could run the ports in NVME mode instead of AHCI.

*Other than for straight large-size block transfers, which are very uncommon generally speaking. Most/vast majority file I/O is small block size, pseudo-random, and there today's I/O connections are far from being maxed out.

 

[Silent_Buddha]

You actually can't saturate* M.2, and you wouldn't be able to saturate SATA either* if you could run the ports in NVME mode instead of AHCI.

*Other than for straight large-size block transfers, which are very uncommon generally speaking. Most/vast majority file I/O is small block size, pseudo-random, and there today's I/O connections are far from being maxed out.

The latest Samsung NVME M.2 SSD has a higher transfer rate than X4 PCIE M.2 can support. So yes, it can be saturated.

Regards,
SB

 

[Grall]

You miss my point, which is that you only saturate SSDs by doing large block transfers. Which rarely happens in the real world. I hear digital movie editing can do it, but most other file accesses are random enough to never saturate the link to the storage device.

 

[Alessio1989]

Teardown Asus RX 470 Strix

Here is the review (Italian): http://www.bitsandchips.it/9-hardware/7335-asus-rx-470-strix-4gb


Meanwhile my friend Michele is working on this comparison:

 

[xEx]

Wow the Nitro 4GB that I have ordered (no stock) for 230 on amazon is now 270....Would prefer a 250 1060...wtf AMD...

 

[Deleted member 13524]

Wow the Nitro 4GB that I have ordered (no stock) for 230 on amazon is now 270....Would prefer a 250 1060...wtf AMD...

What makes you think AMD is the one responsible for the price hiking of these cards and not the OEMs and/or retailers in response to demand?

 

[trinibwoy]

https://app.scoutfin.com/document.h...b759fd048&_branch_match_id=297670879870932296

"AMD is only one of two companies who can offer high performance graphics"

Geez AMD can't help but downplay itself. That should be "one of only two companies"

 

[Grall]

Vega is now 1H '17. ...Which means, NOT first quarter.

Lucky AMD that Titan X is a cut-down GPU. I don't buy cut-down graphics cards, so there's still a chance they'll get to sell me something, even though it will be painful to wait potentially until june next year.

 

[entity279]

I'm also kinda forced to wait on Vega since I've got a Freesync monitor.

 

[Malo]

What makes you think AMD is the one responsible for the price hiking of these cards and not the OEMs and/or retailers in response to demand?

I think it's more that AMD can't actually provide the cards.

 

[bdmosky]

"NVidia is only company who can offer high performance graphics."

Fixed that for you.