AMD Ryzen 7 Parts Available for Pre-Order Now!

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

eva2000

Active Member
Apr 15, 2013
244
49
28
Brisbane, Australia
centminmod.com
I did get hands-on time with Ryzen 7 and have benchmarks (will release on March 2.) This is something I am personally pre-ordering.
ooh nice.. about time AMD gave Intel some competition again !

hmm for centos i usually compile nginx and php-fpm with intel cpu optimised GCC and clang compiler flags. So guess i need to look into AMD cpu flags. Wonder which version of GCC and clang I'd need for AMD Ryzen 7 support ?

maybe znver1 march ? x86 Options - Using the GNU Compiler Collection (GCC)

‘znver1’
AMD Family 17h core based CPUs with x86-64 instruction set support. (This supersets BMI, BMI2, F16C, FMA, FSGSBASE, AVX, AVX2, ADCX, RDSEED, MWAITX, SHA, CLZERO, AES, PCL_MUL, CX16, MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM, XSAVEC, XSAVES, CLFLUSHOPT, POPCNT, and 64-bit instruction set extensions.
@Patrick when you get access to Ryzen 7 and Naples on linux, can you also share output for this command
Code:
gcc -c -Q -march=native --help=target
cheers :)
 
Last edited:

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
@eva2000 AMD will run x86 out of the box. I think the compiler optimizations may not fully be in GCC yet.

I am going to see if I can get one of these up in the DemoEval lab, perhaps using a lantronix KVM, before my India trip.
 
  • Like
Reactions: eva2000

eva2000

Active Member
Apr 15, 2013
244
49
28
Brisbane, Australia
centminmod.com
@eva2000 AMD will run x86 out of the box. I think the compiler optimizations may not fully be in GCC yet.

I am going to see if I can get one of these up in the DemoEval lab, perhaps using a lantronix KVM, before my India trip.
@Patrick cheers. Just updated Centmin Mod LEMP stack to support both GCC 5.3.1 and GCC 6.2.1 along with native CentOS GCC versions, so hopefully AMD Ryzen 7 only needs GCC 6.2 and not GCC 7 to take full advantage of the cpu :)
 

mstone

Active Member
Mar 11, 2015
505
118
43
46
AMD didn't pretend that Barcelona wasn't a fiasco. They acknowledged there was a TLB issue and that's the main issue (which had a performance hit as a result). It just took longer before they rolled out fixed silicon for it and in the meantime had to use cpu-microcode patches to mitigate it.
AMD's story at the time was that it wasn't affecting deliveries, but the vendors couldn't get product. They tried to downplay the significance of the issue, but they didn't ramp up volume deliveries until after they had a handle on the bugs--and that most certainly affected future designs from a number of vendors, and their market share.

I highly disagree with this. They are particularly compelling parts
Well, everyone is certainly entitled to their opinion, but AMD's been making the stuff for a while and relatively few people seem interested in buying it. The performance of the CPUs has been kinda meh in comparison to intel at the same approximate price point, and if the GPU performance is better, who cares--nobody buys integrated GPU looking for performance. And it's not just the CPU, intel has a better story these days on the complete chipset (wireless, network, etc). You can find AMDs in the wild, but basically on budget systems where price matters more than anything else and for a small number of people who just really don't want to buy anything from intel.

What I was discussing was the successor to HT (hyper-transport) which is the "Infinity fabric" which Zen is utilizing and so will Vega as well. It's their current interconnect implementation
Yes, it's their HT replacement, and competitor to QPI (and future UPI). And it would be a lot more of a differentiator if they'd shipped it on time; even now they don't really need it because they haven't shipped anything that actually takes advantage of the theoretical improvements (the importance of the big xen + integrated high end GPU is to utilize that on-die link, there have also been discussions of things like on-die FPGA but that's also not here yet). At this point it's not clear they'll ship something before intel delivers UPI along with parts to utilize it (they're more focused on server environments, and integrating high speed networking).
 
Last edited:

zir_blazer

Active Member
Dec 5, 2016
355
128
43
Some questions related to Ryzen, which maybe Patrick can answer if they aren't NDA blocked:


1 - Does consumer Ryzen supports AVIC (Advanced Virtual Interrupt Controller)?

AVIC is AMD counterpart to Intel APICv for APIC Virtualization (Less overhead that emulating it or using paravirtualized Hyper-V Enlightnement hv-apic). So far, AVIC is supposed to be integrated in Carrizo, but there is only one Desktop part that uses it (The Athlon X4 845) and obviously no one bothered to mention if Hardware supports it and if it works or not (Based on Intel VT-d and AMD-Vi experience, even if the Processor supports it, no one would bother to add AVIC support in the Firmware).
Intel implements APICv on the HEDT platform since Ivy Bridge-E, but the consumer platforms Processors are missing it (Including Xeons E3). AVIC on consumer Ryzen would be a slight performance advantage for virtualization.


2 - Does the integrated PCIe Controller of Ryzen supports PCIe ACS (Access Control Services)?

PCIe ACS is useful if used in conjuntion with the IOMMU to provide PCIe Device isolation. Basically, PCIe by default allows for Peer-to-Peer data transfers, but doing these effectively bypasses the IOMMU and its purpose. ACS allows to disable PCIe P2P and force data to go through the IOMMU. This is important if using Passthrough/Device Assignment, because QEMU with VFIO works at an IOMMU Group level, which are a group of PCI Devices that aren't isolated from each other, thus you are expected to assign all the PCI Devices in a IOMMU Group to the same VM or at least leave them in a dummy Driver for isolation to work correctly. You can bypass isolation with a patch, but is considered a hack and will never be upstreamed because it is not the correct way of doing Passthrough, and incurs an unknow risk since the Devices aren't fully isolated.
So far, Intel HEDT platform support PCIe ACS, but the consumer Processors integrated PCIe Controller does not (Chipsets do). Basically, anything connected to the Processor PCIe Controller will always be in the same IOMMU Group, thus if you have two or three Video Cards, unless you use the ACS patch, you will not be able to pass them to different VMs.
Since Ryzen supposedly has 20 PCIe Lanes, and supports bifurcating them as 8x/8x/4x, it would be absolutely great if PCIe ACS works and you don't have to hack your way through it.


3 - Is possible to provide an example of a lspci, lspci -t, and IOMMU Grouping (May require enabling the IOMMU for the Linux Kernel with a parameter) of Ryzen in a production Motherboard?

No worries if we have to wait until launch, but it will be extremely helpful for early adopters of Ryzen that want to do Passthrough, as they will know exactly what they are getting. If you can get a 8C/16T Ryzen 7 1700 for 320 U$D and it has all those features, it would be simply amazing, and far better for virtualization power users that standard Windows users.
 
Last edited:
  • Like
Reactions: gigatexal

mstone

Active Member
Mar 11, 2015
505
118
43
46
Intel has had various hardware errata over the years so it's not just AMD who has TLB. Infact even Nehalem had TLB (contrary to the claims which say the opposite) and Intel acknowledged it too (which required bios workarounds and patches). Sure AMD's reputation was damaged with the TLB issue but it didn't mean they exited or have their market share drastically affected. That point only really came about with Bulldozer's launch and this verifies that (only a 5% drop in market share attributed to TLB):

http://3.bp.blogspot.com/_YzBo7Kz5y...hwkABvdOo7A/s1600/AMD_server_share_Sterne.GIF
I think you're misreading the graph. They were on a pretty steady upward trend until 2006 when the market was waiting for a quad core part. They flubbed it, their share dropped between 5 and 10 points (definitely more than 5) share--which was a decline of almost 50%--by the end of 07. They went up a little bit once they started shipping again but after that it was just a long slow downhill trend. They never recovered the positive momentum they had from 2003-2006.

[snipping a lot of fanboyism]
There's no point in trying to convince me that the parts aren't that bad, they haven't been selling and no amount of arguing on the internet is going to change that. If you take away the fact that there are a lot of cores, the single thread performance is still meh. So this comes down to "does the market really need a 8 core desktop processor"? The thing is, it's really hard to utilize a lot of cores. There are certain workloads which are embarrassingly parallel, and people doing that sort of thing are obviously happy. And certain overclocking performance fetishists will just love having a bunch of cores whether they are practical or not. But I'm unconvinced that many people will actually be able to maintain good utilization of an 8 core desktop--which is why intel hasn't been selling them (it's not a technical challenge). All of which raises the question: so why is AMD trying to sell high core count desktops instead of pushing these things right at the server space (where it's a lot easier to utilize cores, and where a cheap 8 core is a really compelling part next to a cheap quad core)? I'd guess at least part of the answer (coming full circle) is that AMD is having trouble convincing integrators that they can actually deliver on time and in quantity. I suspect they're also consciously focusing on the high end of each segment (they're debuting naples at 32 cores) and not releasing more practical parts in order to shake the bargain basement reputation they've acquired over the past few years. Eventually I assume they'll come out with 4 core desktop CPUs and 8 core server CPUs, and depending on pricing (and availability!) that might be where they really shake up the market.
 

cactus

Moderator
Jan 25, 2011
830
75
28
CA
So when are we going to see the UP workstation (E3/E5-16xx equivalent) version of Ryzen?
 

zir_blazer

Active Member
Dec 5, 2016
355
128
43
Beginning with the famous Socket 939 Opterons 1xx, AMD always had Opterons on the consumer platform but intended to be used with UDIMM ECC, directly comparable to the Xeons E3 (Which also uses the consumer platform but adds ECC support). Even AM3/AM3+ had such Opterons, but they are rather unknow. And I'm not 100% sure if you NEED an Opteron or/and a Server Motherboard to use ECC or the standard parts also supports it but is an unintended feature.
I don't know if the PRO parts have ECC, I think the only difference with those is that AMD commited to have these parts with a 5 years availability cycle or something else not very interesing.
 

PigLover

Moderator
Jan 26, 2011
3,184
1,545
113
... Even AM3/AM3+ had such Opterons...
Years ago I used to LOVE that about AMD chips - even at the lowest end of the scale with AM3/AM3+ if you got a MB with ECC support enabled the chip was happy to oblige.

My very first "low power" server build was on an AM3+ chip using ECC memory. I was ecstatic to get a system idle below 30 watts with the drives spun down (16GB & 4x 500GB IIRC - big time :)). I really believe that competing with this is why the Pentium & I3 chips always supported ECC - until Kaby Lake, of course, when Intel figured out that the lack of competition could allow them to reserve that to E3 and price accordingly.

Regardless of how each of us happens to analyze what happened in the past, nobody can argue that the lack of anyone pushing the boundaries - the lack of a credible disruptor - has allowed Intel to slow down and milk each product in ways you didn't see when there were two horses on the track. Whether Ryzen 7 proves its value or not I'm glad to see AMD in the race again rather then being the "also ran" they've been lately.
 

cactus

Moderator
Jan 25, 2011
830
75
28
CA
Well my 1680v3 gets stomped on by all three models... at least in R15.
Yeah, I am thinking feature equivalent vs performance. 10G onboard with ECC, maybe some more PCIE. Video, SAS Raid for local spinners, 10G card maxes out a X370 board's PCIE in full ATX.