Speculation: Ryzen 3000 series

[Kedas]

I did put the specs of Ryzen 2000 and the leaked Ryzen 3000 specs (from AdoredTV) on a chart so you can see that it can be true, obviously we don't know how much the design change requires more power and TDP isn't really an exact value but enough to compare the quality range they are in. 3850X stands a bit out but would come much later so.....
The values for 3000 are divided by 2 (7nm 50% power)

 

[dlerious]

While I agree that UEFI support for ASRock's X370 boards has been pretty bad, you'll notice that they shifted their UEFI support to X470 and B450 boards instead. So when the 5-series boards launch, you can bet that ASRock (and others) will prioritize support for those new boards. You will want to buy a new board.

If you are expected good UEFI updates for X370 boards, eh, forget about it. Not really gonna happen.

I can kind of see lesser support for the budget line-up, but I expect a little more when it comes to the flagship boards. I don't like this mobile model where they stop once a new model is released.

My X370 Crosshair VI had a 3 year warranty (still does). I'd like to see them supporting it at least as long as the warranty. My Zenith Extreme was $550 when I bought it - still waiting for an update with the new AGESA. I'm at the point where I'm ready to avoid ASUS products all together. Workstation and server shouldn't be something you need to upgrade every year.

 

[Mopetar]

I'm not interested in a platform w/o PCIE 4.0 + 7nm going forward.
EPYC will ahve PCIE 4.0
Ryzen and threadripper better as well.. Otherwise, I have no interest.

I'd hope they shoot for much more PCIE slots beyond the current 32 slots offered on Ryzen and PCIE 4.0

Any reason you need more lanes? Don’t the current top of the line GPUs not even need 16x PCIE 3 to give full performance?

Given that multi-GPU use is lower than its ever been, I’m just left wondering what you’d want (or need) all of that for, especially in a consumer chip. Sure you’d want PCIE 4 if you’re trying to future proof a build, but I can’t see needing that many lanes.

 

[Insert_Nickname]

My X370 Crosshair VI had a 3 year warranty (still does). I'd like to see them supporting it at least as long as the warranty. My Zenith Extreme was $550 when I bought it - still waiting for an update with the new AGESA.

The CH6 got an update with AGESA PR1006 recently (version 6401). I haven't got around to installing it yet though. Dunno about the Zenith.

 

[ozzy702]

Any reason you need more lanes? Don’t the current top of the line GPUs not even need 16x PCIE 3 to give full performance?

Given that multi-GPU use is lower than its ever been, I’m just left wondering what you’d want (or need) all of that for, especially in a consumer chip. Sure you’d want PCIE 4 if you’re trying to future proof a build, but I can’t see needing that many lanes.

He doesn't play games and doesn't "need" PCIE 4.0, at this point nobody does. PCIE 4.0 will be short lived and quickly replaced with PCIE 5.0.

 

[thigobr]

More PCIE bandwidth is always good for storage even on mainstream. GPUs are fine at 3.0 8x but with PCIE M2 drives getting cheaper and cheaper you could have 2 drives in your system running in 4.0 2x instead of 1 drive 4x 3.0...

 

[Zucker2k]

Any reason you need more lanes? Don’t the current top of the line GPUs not even need 16x PCIE 3 to give full performance?

Given that multi-GPU use is lower than its ever been, I’m just left wondering what you’d want (or need) all of that for, especially in a consumer chip. Sure you’d want PCIE 4 if you’re trying to future proof a build, but I can’t see needing that many lanes.

Oh no, you didn't! Here comes the NVME onslaught!!

 

[scannall]

Oh no, you didn't! Here comes the NVME onslaught!!

NVME drives are the single best upgrade you can do. Particularly if you are coming from a hard drive. And they take lanes. Wanting at least faster lanes isn't a bad thing.

 

[moinmoin]

Is there a way to split one PCIe 4 lane into two PCIe 3 ones? If there is that would offer a lot more possibilities for different AM4 board configurations.

 

[Tuna-Fish]

Is there a way to split one PCIe 4 lane into two PCIe 3 ones? If there is that would offer a lot more possibilities for different AM4 board configurations.

Only by using a switch, like the one in all chipsets.

 

[Reinvented]

While I agree that UEFI support for ASRock's X370 boards has been pretty bad, you'll notice that they shifted their UEFI support to X470 and B450 boards instead. So when the 5-series boards launch, you can bet that ASRock (and others) will prioritize support for those new boards. You will want to buy a new board.

If you are expected good UEFI updates for X370 boards, eh, forget about it. Not really gonna happen.

I just noticed that they released an update for B350's and X370. Finally! I'm going to give them a try to see how it works with 2700X.

 

[DrMrLordX]

I just noticed that they released an update for B350's and X370. Finally! I'm going to give them a try to see how it works with 2700X.

Good luck. I get nothing but reduced memory speed on my 1800x every time they update my X370 Taichi.

I can kind of see lesser support for the budget line-up, but I expect a little more when it comes to the flagship boards. I don't like this mobile model where they stop once a new model is released.

My X370 Crosshair VI had a 3 year warranty (still does). I'd like to see them supporting it at least as long as the warranty. My Zenith Extreme was $550 when I bought it - still waiting for an update with the new AGESA. I'm at the point where I'm ready to avoid ASUS products all together. Workstation and server shouldn't be something you need to upgrade every year.

Beware updates. With AM4, adding compatibility with new chips sometimes breaks functionality with existing product. Yes, ASRock will (slowly) update their 3-series AM4 chipset UEFIs. But they can't/won't maintain the same max memclocks, and they introduce other funky bugs. It's been a hair-pulling experience.

 

[ub4ty]

Any reason you need more lanes? Don’t the current top of the line GPUs not even need 16x PCIE 3 to give full performance?

Given that multi-GPU use is lower than its ever been, I’m just left wondering what you’d want (or need) all of that for, especially in a consumer chip. Sure you’d want PCIE 4 if you’re trying to future proof a build, but I can’t see needing that many lanes.

It's always good to have more I/O. NvMe storage is one particular reason for more PCIE lanes.
Not sure if you guys heard the news or earlier discussions but AMD is likely exposed Infinity fabric over PCIE 4.0. So, there's that major reason and the fact that it's future proof'd buying a mobo w/ PCIE 4.0. I'd settle on Ryzen 3 7nm for the long haul (5 years+) if it came w/ PCIE 4.0.

PCIE 4.0 has lower latency and higher bandwidth.
AMD also supports bifurcation on threadripper allowing one to break out a single x16 into x4/x4/x4/x4.
My Ryzen 7 systems have I/O fully saturated. I have a x8 card and x16 card/nvme and sata. I'd like to have more options on it. I use GPUs for compute related flows not gaming. Lower latency and higher bandwidth (i expect navi to be pcie 4.0) are needed.

On my threadripper builds, I have all of the PCIE 3.0 slots x8/x16/x8/x16 filled with cards. I also have the PCIE 2.0 occupied. I also have 3 nvme drives and a slew of sata drives. So, there I could use more I/O as well. I am also highly interested in an I/O chiplet as I have several devices that can DMA data too and from each other w/o CPU intervention and I'd like to see what interesting things can be done on the I/O chiplet when AMD exposes the internals more.

Essentially I want the future Now and have real uses for it. AMD is headed towards HSA and (infinity fabric) over PCIE 4.0 is one big stepping stone. I already have some PCIE 4.0 devices being eyeballed and there is the obvious progression for several devices :

https://news.samsung.com/global/sam...-factor?utm_source=mainkv&utm_medium=internal
The NF1 SSD features a brand new, high-performance controller that supports the NVMe 1.3 protocol and PCIe 4.0 interface, delivering sequential read speeds of 3,100 megabytes per second (MB/s) and write speeds of 2,000MB/s.

PCIE 4.0 is also how AMD drop kicks Nvidia's proprietary Nvlink in the face and this goes all the way up to the Power9 architecture and their ridiculously priced DGX-2


Big things will fall apart if AMD opens up Infinity fabric over PCIE 4.0 :

 

[ub4ty]

He doesn't play games and doesn't "need" PCIE 4.0, at this point nobody does. PCIE 4.0 will be short lived and quickly replaced with PCIE 5.0.

I'd love to see some professional analysis regarding PCIE 4.0 being short lived and quickly replaced by PCIE 5.0.
As far as PCIE 4.0 and AMD :
https://www.servethehome.com/xilinx-alveo-u280-launched-possibly-with-amd-epyc-ccix-support/
The U280 acceleration card includes CCIX support to leverage existing server interconnect infrastructure for high bandwidth, low latency cache coherent shared memory access with CCIX enabled processors including Arm and AMD. (Source: Xilinx Alveo U280 whitepaper WP50 (v1.0) accessed 16 November 2018)

https://www.arm.com/solutions/storage

Yeah, it's a big deal. It's what AMD has been working towards regarding HSA. If launched on PCIE 4.0, it's here to stay and will change a great deal of things.

And true enough I don't play video games on my workstations. However, I do have a need for PCIE 4.0 and have already spec'd towards its coming and changes therein.

 

[ub4ty]

Only by using a switch, like the one in all chipsets.

https://www.anandtech.com/show/12961/gigabyte-aorus-pcie-x16-m2-for-four-nvme-drives-x399-compatible
https://www.techarp.com/computer/8tb-ssd-ddr4-speed-guide/

http://www.thessdreview.com/raid-en...iew-4-x-m-2-ssd-raid-at-10gb-s-and-932k-iops/

 

[ub4ty]

Good luck. I get nothing but reduced memory speed on my 1800x every time they update my X370 Taichi.



Beware updates. With AM4, adding compatibility with new chips sometimes breaks functionality with existing product. Yes, ASRock will (slowly) update their 3-series AM4 chipset UEFIs. But they can't/won't maintain the same max memclocks, and they introduce other funky bugs. It's been a hair-pulling experience.

I have dedicated a machine to evaluating various hardware updates. It indeed has been a bumpy ride. Seems a AGESA 1.0.0.6 update came out on 12/21 for Asrock. Going to wait until chatter begins on its stability. Running a suite of stress test/mem test and benchmarking apps is a Pain in the butt.

 

[jpiniero]

I'd love to see some professional analysis regarding PCIE 4.0 being short lived and quickly replaced by PCIE 5.0.

Intel's Tinsley platform (Saphire Rapids/Granite Rapids) is rumored to support PCIe 5 (and DDR5 for that matter). When you might see that, however...

 

[ub4ty]

Intel's platform ...

Thanks for the laugh

 

[Reinvented]

I have dedicated a machine to evaluating various hardware updates. It indeed has been a bumpy ride. Seems a AGESA 1.0.0.6 update came out on 12/21 for Asrock. Going to wait until chatter begins on its stability. Running a suite of stress test/mem test and benchmarking apps is a Pain in the butt.

So far so good. No problems on my end with a 1700 @ 3.8 1.23v, and no problems with a stock 2700X. Using GTZRX B-die Trident Z RGB.

 

[TempAcc99]

I'd love to see some professional analysis regarding PCIE 4.0 being short lived and quickly replaced by PCIE 5.0.

This is because pcie 4.0 took forvever to specify and due to that they changed how it was done to make it quicker. So pcie 5.0 final spec came out rather quickly after 4.0 AFAIK only 1 year later. hence it was assumed that 4.0 is very short lived.

 

[ub4ty]

This is because pcie 4.0 took forvever to specify and due to that they changed how it was done to make it quicker. So pcie 5.0 final spec came out rather quickly after 4.0 AFAIK only 1 year later. hence it was assumed that 4.0 is very short lived.

Every PCIE versioned spec has historically been released in this manner. PCIE 5.0 will be no different. The spec for 5.0 is set to come out in 2019 which means you probably wont see production until 2020 with availability being 2021/2022 in select enterprise markets. Which means you probably wont be seeing this in consumer markets until 2023/2024. PCIE 4.0 will be as long lived as the other PCIE versions. Furthermore, no mention of the increased costs and signal issues related to the increased speeds. This is not going to be a cheap feature. For this reason, PCIE 4.0 will likely stay around for as long as PCIE 3.0 in consumer markets and PCIE 5.0 will be much further out. I could easily foresee PCIE 5.0 being heavily restricted to specific enterprise markets.. Doubling the speed and signal issues with PCIE 5.0 makes it a non-event on consumer mobos.

Since PCIE 5.0/DDR5 system memory is being claimed by Intel, I think you can go ahead and ignore it.
I thought there was something more to this claim besides final spec release dates. This is historically have PCIE has been released. All of the releases are long lived and more costly. PCIE 5.0 might be used as a short run interconnect to the chipset (something Intel loves doing : parceling out shared bandwidth through the chipset). However, there is literally no consumer on the other side for home users that has that kind of bandwidth requirement. This is more likely for enterprise 400Gb nics that can actually drive that much traffic. The board traces and components are going to be very expensive.

Technical details :
But as the transfer speed increases, the signal decays and devices/components need to be crammed together to maintain higher signal availability. It is stated that with PCIe Gen 1.0, signal retained strength up to 20 inches in mainstream FR4 boards. With PCIe 4.0, the signal would decay only after 3-5 inches, hence requiring the need of extender cables and signal amplifying chips.

Retimer chips for a full 16-lane full PCIe 4.0 could cost $15 to $25 — if you can find them. Upgrading an adapter card from Megtron-2 to Megtron-4 materials might only add a dollar or so. However, the cost of a similar upgrade for a motherboard is about $100, and if the upgrade is to even higher quality Megtron-6 it would cost about $300.
“The data center will go to Megtron-4 for PCIe 4.0 and that will add maybe $10 cost — and you may still need retimers,” said Krause. “For version 5.0, people will weigh even higher-cost PCB materials and retimers or move to cables.”
“What we have been using for 4.0 and expect to use for 5.0 is twinax cables and firefly connectors,” he added. “The cost is very low compared to retimers, you can get whatever you want in distance, and the latency is really good.”
Indeed, Krause noted that “there’s been a lot of interest in using cables … for every inch on a board, you can go 10 inches on cables for the same power and loss budget, but cables have costs in being routed and connected.”

https://www.eetimes.com/document.asp?doc_id=1333388&page_number=1

So, you can pretty much forget about PCIE 5.0 coming to the desktop any time soon.
PCIE 4.0 and DDR4 is where the action will be. Intel is just blowing smoke up people's behinds.

 

[ub4ty]

So far so good. No problems on my end with a 1700 @ 3.8 1.23v, and no problems with a stock 2700X. Using GTZRX B-die Trident Z RGB.

Excellent. I'll set it up on one of my machines and do a stability test. Thanks for the info and nice 1700 SPECS

 

[NostaSeronx]

PCIE 5.0 will be ready sooner than later...

https://blogs.synopsys.com/expressyourself/2018/12/19/first-pcie-5-0-32gt-s-phy-ip/
https://www.synopsys.com/dw/ipdir.php?ds=dwc_multi_protocol_32g_phy
https://www.synopsys.com/dw/ipdir.php?ds=dwc_pcie5_phy => PCIe 5.0 PHY, TSMC16FFC x4, North/South (vertical) poly orientation

It is expected to be not much to fix between 0.9 now, and 1.0 in Q1 2019. So, PCIe 5.0 is effectively done.

DDR5 is a must to support in phy, but not necessarily in product.
https://news.synopsys.com/2018-10-2...-Power-Efficient-DDR5-and-LPDDR5-IP-Solutions
https://www.synopsys.com/dw/ipdir.php?ds=dwc_ddr54_phy
DDR5/4 PHY - TSMC 16FFC => 4.8 GHz
DDR5/4 PHY - TSMC N7 => 6.4 GHz (It states 4800 Mbps, but this will quickly get up to peak and beyond in custom)

 

[ub4ty]

PCIE 5.0 will be ready sooner than later...

https://blogs.synopsys.com/expressyourself/2018/12/19/first-pcie-5-0-32gt-s-phy-ip/
https://www.synopsys.com/dw/ipdir.php?ds=dwc_multi_protocol_32g_phy
https://www.synopsys.com/dw/ipdir.php?ds=dwc_pcie5_phy => PCIe 5.0 PHY, TSMC16FFC x4, North/South (vertical) poly orientation

It is expected to be not much to fix between 0.9 now, and 1.0 in Q1 2019. So, PCIe 5.0 is effectively done.

DDR5 is a must to support in phy, but not necessarily in product.
https://news.synopsys.com/2018-10-2...-Power-Efficient-DDR5-and-LPDDR5-IP-Solutions
https://www.synopsys.com/dw/ipdir.php?ds=dwc_ddr54_phy
DDR5/4 PHY - TSMC 16FFC => 4.8 GHz
DDR5/4 PHY - TSMC N7 => 6.4 GHz (It states 4800 Mbps, but this will quickly get up to peak and beyond in custom)

Read my comments about the component and motherboard costs and the physical restrictions.
That alone is a non-starter for the desktop environment.
Next up is the complete lack of use cases as there is absolutely nothing that has such bandwidth requirements on the desktop as far as PCIE cards. Then there's the history of PCIE releases.. Essentially, you're not going to see the tech until they have allowed a good enough period for the previous standard. Enterprise will see it first for some time. So, until this is a thing on CPUs and boards in enterprise, this is a non-story.

I'm not sure what bringing this up achieves as PCIE 4.0 and DDR4 is most definitely being delivered and supported first and it will be a standard for some time. A couple of more weeks and that will be confirmed via Rome. If the suggestion is to wait on some bogus pronouncement from intel's imaginative platform, that's what makes this all the more laughable.


https://www.servethehome.com/qct-daytona-server-ready-for-amd-epyc-rome-2/
An example of this is PCIe Gen4 which is coming to x86 in 2019 first with AMD EPYC 2 codename Rome. At the recent AMD Next Horizon event, we saw a QCT platform designed for Rome.
This looks like an AMD EPYC server that has been updated for PCIe Gen4...

PCIE 5.0 and DDR5 system memory is a non-story until it becomes a story... and what I mean by that is it existing in a shippable CPU and on enterprise boards and there's a discussion about the costs and limitations. As far as spec drafts, etc.. Yes, many here are fully aware of that and the history in which that correlates in noway with actual availability. DDR5 memory modules have been demo'd btw, it currently sits at CL42.

AMD's next gen is shipping with PCIE 4.0 and DDR4. That's all that matters. 7nm and AMD's new platform is the big upgrade. PCIE 4.0 is a nice have. DDDR4 is more than sufficient. I could easily ride such a platform for 5 years+ until PCIE 5.0/DDDR5 do or don't become a thing. Enterprise will be equipping heavily on PCIE 4.0/DDR4. The orders are already coming in for Rome. This is what keeps new standards from proliferating as it always has in computing.

 

[dlerious]

Read my comments about the component and motherboard costs and the physical restrictions.
That alone is a non-starter for the desktop environment.
Next up is the complete lack of use cases as there is absolutely nothing that has such bandwidth requirements on the desktop as far as PCIE cards. Then there's the history of PCIE releases.. Essentially, you're not going to see the tech until they have allowed a good enough period for the previous standard. Enterprise will see it first for some time. So, until this is a thing on CPUs and boards in enterprise, this is a non-story.

What's this history you keep bringing up? PCIe 1.0 launched in 2003, 2.0 launched in 2007, 3.0 launched in 2010, and 4.0 was 2017. With 5.0 slated for next year, that would be the shortest period between releases.