Who is looking to upgrade to Zen2

Issue is that, even though they promise backward compatibility, existing TR mobos don't support PCIe 4, whereas TR3 should support PCIe 4 with the new boards that come out. So you're potentially looking at basically a full platform update.

Also, as for TR1 deals... https://hardforum.com/threads/amd-tr4-1920x-249-and-1950x.1982139/

Except pcie 4 is a big fat meh. Maybe in a few years it'll matter at which point mobos will be way cheaper, but right now nothing can use it except nvme drives doing large file transfers. If I cared much about that I could nvme raid on existing x399 for way cheaper than a new pcie4 mobo anyway.
 
Except pcie 4 is a big fat meh. Maybe in a few years it'll matter at which point mobos will be way cheaper, but right now nothing can use it except nvme drives doing large file transfers. If I cared much about that I could nvme raid on existing x399 for way cheaper than a new pcie4 mobo anyway.

Nah there is way more to it than gpu and nvme...

Al is/was the president pf PCISIG who developed the spec.

upload_2019-6-17_4-39-53.png
 
That rumored x590 chipset (take with a grain of salt) is throwing a curveball into the mix if it really has more PCIe lanes. The decision between TR and Ryzen platforms for those looking to upgrade to a 16 core just got more complicated.
 
That rumored x590 chipset (take with a grain of salt) is throwing a curveball into the mix if it really has more PCIe lanes.

It can have a larger PCIe switch that can break out the available lanes from the CPU into more / different stuff for peripherals, but bandwidth won't be seeing an increase. That's still not a bad thing, but users will have to be mindful about what connected devices they're trying to load simultaneously.

If you're actually looking to load the whole system down, you're going to want HEDT with more direct lanes to the CPU.
 
  • Like
Reactions: mikeo
like this
I got my eyes on the 3900x and the 3950x but I'm also wondering if Intel gets that new 10core cpu out this year as well. Not planning on building new again til November anyway so I should have some nice choices either way.
 
I want a 7nm TR4 likely not the 64C version even though I could easily afford it (well as long as I remain employed at the current job - which is 22 years now as primarily a c++ programmer doing medical imaging research).

Even that is beyond what I can imagine needing, but it could be the last workstation like this I would ever be in position to purchase
 
I was going for the 16c but after seeing the price leaks of the motherboard it just makes more since to wait for a 7nm thread ripper. 1800x lives on sadly
 
  • Like
Reactions: mikeo
like this
I was going for the 16c but after seeing the price leaks of the motherboard it just makes more since to wait for a 7nm thread ripper. 1800x lives on sadly

I'm considering getting a x470 board to match with Ryzen 3900x or 3950x. When PCIe 4.0 devices are abundant and more compelling I'll probably be on a new CPU and board by then.
 
I'm considering getting a x470 board to match with Ryzen 3900x or 3950x. When PCIe 4.0 devices are abundant and more compelling I'll probably be on a new CPU and board by then.

So long as the configuration has been tested to work well, this is what I expect to be recommending.
 
PCIe 4.0 is a nice-to-have, but probably not a zOMG game changer. Gains will be mostly academic, though if someone has a use case that will visibly benefit from it, good for him.

Alone, it is not a reason to prefer AMD over Intel. As a freebie, hey, sure, I'll take it.
 
I think you'll want to wait for reviews. I certainly do, and this is one of my personal concerns, and one I've brought up before.

Where we'll see this be an issue is in frametime consistency in games that use more cores, well, more cores than are present on one die. Crossing CCX boundaries on one die has been mostly addressed in Zen+ and looks to be addressed in Zen 2, so generally speaking the twelve core part might be fine relative to the eight core part, as having six cores enabled on one die is overkill for most games now (and more than adequate for the rest).

Always wait for the reviews. But that being said, things are looking very favorable for Zen 2 in general.
 
I know most technology is obsolete soon after buying it, but this is getting ridiculous..

PCIe 2.0 isn't even obsolete yet.

These 'standards' are pretty easy to write down and perhaps demonstrate in a lab, but mass producing products with them will be another challenge entirely.
 
Well, PCIe 4.0 will be obsolete very soon.
Problem is simple. The GPU had increased compute power by some factors in the last years and nothing moved since PCIe 3.0, so Nvidia and AMD have figured out new kind of non standard links. Also those GPU need to talk to each other when you put several of them, and they need to talk fast. The CPU on the other side has seen its compute power double and double again in a couple of years, not because of Intel. And this is going to double again and with packed chiplets it's going to get huge. So PCIe 3.0 is already obsolete but 4.0 will be very soon and so on til 2021 PCIe 6.0... and then maybe it will come to optical connection to go even faster.
 
Well, PCIe 4.0 will be obsolete very soon.
Problem is simple. The GPU had increased compute power by some factors in the last years and nothing moved since PCIe 3.0, so Nvidia and AMD have figured out new kind of non standard links. Also those GPU need to talk to each other when you put several of them, and they need to talk fast. The CPU on the other side has seen its compute power double and double again in a couple of years, not because of Intel. And this is going to double again and with packed chiplets it's going to get huge. So PCIe 3.0 is already obsolete but 4.0 will be very soon and so on til 2021 PCIe 6.0... and then maybe it will come to optical connection to go even faster.

In a word: no.

You might want to reference the definition of the word 'obsolete' and its connotations with respect to computing.
 
  • Like
Reactions: N4CR
like this
In a word: no.

You might want to reference the definition of the word 'obsolete' and its connotations with respect to computing.
What he said. You will not see PCI-E 6.0 on the consumer side before DDR5 IS available and in Mass production so Intel/AmD can launch new chipsets with them.


If you guys think PCI-E 4.0 enabled boards are pricey, what til you see what a 6.0 board will cost if they find they cannot use the same trace lengths as 5.0. it is assumed they will be able to but we won't know until it is available to buy. I'm guessing 2021 for 5.0 and 2023 sliding into '24 for 6.0.

The Sig governing body is just now getting ready to open up mass certification for 4.0 devices in late July or early August.
 
What he said. You will not see PCI-E 6.0 on the consumer side before DDR5 IS available and in Mass production so Intel/AmD can launch new chipsets with them.


If you guys think PCI-E 4.0 enabled boards are pricey, what til you see what a 6.0 board will cost if they find they cannot use the same trace lengths as 5.0. it is assumed they will be able to but we won't know until it is available to buy. I'm guessing 2021 for 5.0 and 2023 sliding into '24 for 6.0.

The Sig governing body is just now getting ready to open up mass certification for 4.0 devices in late July or early August.

While trace lengths will be more difficult with PCIE 5.0, they should not get any more difficult with PCIE 6.0.

With PCIE 6 they are moving away from only two states. (0 or 1) to four states like some high high end switches do.
01
10
00
11

This will allow them to double the bandwidth while halving the frequency.

I doubt we will see this outside of data centers and supercomputer before 2024.
 
While trace lengths will be more difficult with PCIE 5.0, they should not get any more difficult with PCIE 6.0.

With PCIE 6 they are moving away from only two states. (0 or 1) to four states like some high high end switches do.
01
10
00
11

This will allow them to double the bandwidth while halving the frequency.

I doubt we will see this outside of data centers and supercomputer before 2024.

Yeah I agree that the move to 4 signal states from 2 will hopefully go as they expect, I am still curious if it will require thicker PCBs and such. I like you do not expect it in th consumer space for quite a while.

Hopefully it proceeds the way they are planning. It's crazy how fast they are moving all of a sudden.
 
Back
Top