The Achilles' heels of Ryzen no one is talking about

I'm hoping AMD will release a better motherboard chipset soon to support Ryzen, but again I doubt they will unless they can adapt something from the server market to fulfill this role. The fact is, AMD targeted Intel's mainstream segment, not the HEDT segment.
 
I'm hoping AMD will release a better motherboard chipset soon to support Ryzen, but again I doubt they will unless they can adapt something from the server market to fulfill this role. The fact is, AMD targeted Intel's mainstream segment, not the HEDT segment.

I can almost guarantee that AMD will target the high end workstation market with a server socket compatible ATX motherboard.
 
I wouldn't doubt it but if they do it will be almost certainly be variation of something specifically designed for the server market.
 
I'm hoping AMD will release a better motherboard chipset soon to support Ryzen, but again I doubt they will unless they can adapt something from the server market to fulfill this role. The fact is, AMD targeted Intel's mainstream segment, not the HEDT segment.
What they've released so far doesn't seem bad considering the intended market. They haven't shown the APUs yet and conceivably could bring HBM along with the APU acting as system memory or even an enormous cache for the CPU. That would have a rather significant impact on the platform that may have been overlooked. Quad channel DDR4 wouldn't approach that capability. Not to mention the possibility of a GPU that isn't eating into that PCIE3 limit.

If you recall from the Kanter interview, Zen was a bit lacking in instruction decoders, AVX, and MADD capabilities from a typical design. It wouldn't be surprising if the true high performance chips were designed as APUs. All the tasks Zen appears weak in would be inherently strong for a GPU. I'm sure they could pull more robust parts from the server market, but what they've shown so far looks capable considering costs. If someone really wanted one of those over the top systems I'm sure they could use Naples with it's 48(?) PCIE3 lanes and crapload of cores. Keep in mind "7" series parts from AMD are typically the upper mid range and not high end.
 
Comparing apples to apples, I think the Z270 and X370 are far closer than X99 to X370 is. Even then, I think Z270 is better on paper at the very least. I think the issue is that AMD targeted Haswell and it's platforms which is why the X370 motherboard platform is so underwhelming.
I'm very sad to see the x370 already behind prior to release. I see some solid attempts at high end boards but those fancy multiple m.2 connectors are being held back.
 
I'm very sad to see the x370 already behind prior to release. I see some solid attempts at high end boards but those fancy multiple m.2 connectors are being held back.

multiple m2 slots are not for everyone. Heck I am yet to jump on to one. I can bet around 90+% of the consumer base wouldn't be hurt by this. Remember AMD needed something solid on CPU side and something decent to support it. More features and other things will come in future when Zen sells and generates more revenue that goes back in R&D. At this point to have a decent high end board for AMD is enough, I am surprised they even have one.
 
The MSI Titanium has two M.2 slots. A comparable z270 Titanium has 3. Big loss there.:rolleyes: *eye roll*
 
The MSI Titanium has two M.2 slots. A comparable z270 Titanium has 3. Big loss there.:rolleyes: *eye roll*
Nothing is stopping them adding a M2 (or two) off the south bridge, it's just it has to share those four lines with many other devices.
 
If you need more PCI-E lanes than the "mainstream" X370 or Z270 chipsets provide, then pay 3x more for X99 or X299 when that comes out. Hell, wait for Naples since you're running some sort of server or rendering/number crunching machine.
 
From the 2 M2s on my Z170 board, 1 of them disables 3 SATA ports, the other disables a PCI-E port, so I'm not sure how that is any different :p
 
Comparing apples to apples, I think the Z270 and X370 are far closer than X99 to X370 is. Even then, I think Z270 is better on paper at the very least. I think the issue is that AMD targeted Haswell and it's platforms which is why the X370 motherboard platform is so underwhelming.

I mean AMD had to target skylake/haswell-e because of the core count. I think if they would of targeted the 7700k it would of been odd. They are selling top tier processors for mainstream prices. Just my 0.02c
 
From the 2 M2s on my Z170 board, 1 of them disables 3 SATA ports, the other disables a PCI-E port, so I'm not sure how that is any different :p

It isn't. This is why I've been saying we need more PCIe lanes. Not less, even in the consumer or mainstream segments.
 
AMD X370 has 24 PCIe lanes which is the same as Intel Z270


http://www.legitreviews.com/amd-ryzen-learned-ces-2017_190305

No it doesn't - that is describing how the CPU lanes are being routed.

Ryzen CPU = 24x PCI-E 3.0 lanes
X370 Chipset = 8x PCI-E 2.0 lanes

Kaby Lake/Skylake/Haswell/Ivy Bridge CPU = 16x PCI-E 3.0 lanes
Z270 chipset = 24x PCI-E 3.0 lanes
Z170 chipset = 20x PCI-E 3.0 lanes

The Intel chipsets communicate from chipset to CPU via the DMI 3.0, which is PCI-E 3.0 x4.
 
From the 2 M2s on my Z170 board, 1 of them disables 3 SATA ports, the other disables a PCI-E port, so I'm not sure how that is any different :p
I don't think my board does any of that (I have read the manual,) but it's Z270 and matx. It will do nvme raid too. But I only have a single m2 card for now and use only 3 of the sata ports. The BPX benchmarks very well. I only want to raid it for more space but added speed is a bonus.
 
Last edited:
Lets keep in mind that this is the first release of Ryzen. They will enhance both the process and architecture. I expect to see expanded I/O next year.
 
AMD is clearly competing with Intel in a space clean between their mainstream and HEDT parts, and it is paying off in spades. Very wise marketing choice. That being said, I would love to see a Ryzen with a slightly lower clock and more lanes on it in future. For now these things are basically glorified console processors, which is a great for gamers.

Also, I didn't know any Intel processors officially supported over 2666, which I see Zen does. Boards are saying 3200, so on that note I always buy 1 level up from what the board suggest, and got me 16GB of 3600 :)
 
It really scares me that so many people buy hardware without reading or understanding the underlying specs.... Yes, that board has two m.2 slots. If you bothered to actually read the spec sheet, you would see that one of them is completely gimped in that it is running PCIe 2.0. So, can you connect the current (or upcoming) fastest m.2 drives to it? Absolutely. Will you be getting anywhere near their top speeds? Absolutely NOT. Especially with the projected speeds that should be featured in upcoming upcoming factored in.
Ya man. I need all that speed to load up my porn sites faster. It doesn't bother me that it's gimped. I won't be using 2 m.2 drives anytime soon.
 
It really scares me that so many people buy hardware without reading or understanding the underlying specs.... Yes, that board has two m.2 slots. If you bothered to actually read the spec sheet, you would see that one of them is completely gimped in that it is running PCIe 2.0. So, can you connect the current (or upcoming) fastest m.2 drives to it? Absolutely. Will you be getting anywhere near their top speeds? Absolutely NOT. Especially with the projected speeds that should be featured in upcoming upcoming factored in.

The thing that bother me is that people complain and they are using generally no M2 drive, one graphics card and no I/O slots... or the new common one: cpu, one nvm, 1 sata, 1 hdd and a Titan on a X99 board...
I have no issues with people saying they will hold off until a better chipset arrives or go intel due to Pci-E lanes for their 3x M2 drives, pci-e nic, audio, 2 titan x+++

If one likes cost to everyone, sure give em billions of pci-e lanes, but as the market is 95% will not utilize the pci-e lanes they've offered and it's not a drawback for them, it's a cost benefit.

Also worth noting is that AMD effectively have 4-8 PCI-E lanes covered with integrated usb controller, run a external PCI-E card through usb for a NVM drive is so be, for me I doubt I'll use that usb port and a 2nd nvm drive for years! like 95% of the world :)
 
I don't think my board does any of that (I have read the manual,) but it's Z270 and matx. It will do nvme raid too. But I only have a single m2 card for now and use only 3 of the sata ports. The BPX benchmarks very well.

dmark_1.jpg


I only want to raid it for more space but added speed is a bonus.

Which board are you using :)? Mine also does NVME Raid so I don't think that is a guarantee that you will not lose sata / pcie slots.
 
I'm not using m2 sata but pcie. The only pcie 4x slots on it are the 2 m2 slots. There are also two 1x slots (and the 2 gpu slots that are direct to the cpu.) I don't think it even uses all of the pcie lines fron the Z270, but there is also the onboard wifi/bluetooth and usb 3.1.
 
Last edited:
No it doesn't - that is describing how the CPU lanes are being routed.

Ryzen CPU = 24x PCI-E 3.0 lanes
X370 Chipset = 8x PCI-E 2.0 lanes

Kaby Lake/Skylake/Haswell/Ivy Bridge CPU = 16x PCI-E 3.0 lanes
Z270 chipset = 24x PCI-E 3.0 lanes
Z170 chipset = 20x PCI-E 3.0 lanes

The Intel chipsets communicate from chipset to CPU via the DMI 3.0, which is PCI-E 3.0 x4.

This. Ryzen is interesting because it allocates more PCIe lanes to the CPU which may or may not prove beneficial. Ryzen has the potential to be faster in multi-RAID NVMe configurations as it shouldn't have the same bottleneck DMI 3.0 creates. Time will tell what if any difference this theoretical difference makes.
 
This. Ryzen is interesting because it allocates more PCIe lanes to the CPU which may or may not prove beneficial. Ryzen has the potential to be faster in multi-RAID NVMe configurations as it shouldn't have the same bottleneck DMI 3.0 creates. Time will tell what if any difference this theoretical difference makes.

So then, for Intel, does the 4x PCI-E come from the 16/24 or it is not counted there(same for Ryzen)
 
Have no-one ever tested pci-e lane scaling, 8\8x works fine for titan X with no penalty, why should you care about "ohh noes, no 16\16X 16\16X SLI?"
https://www.pugetsystems.com/labs/articles/Titan-X-Performance-PCI-E-3-0-x8-vs-x16-851/

You can start to see differences at 8x/8x and 16x/16x with two Titan X's. Even single card 8x compared to 16x but still small. Meaning for single card PCIe 3 will probably not be limiting for at least 2 generations or more - basically you will need a single card with performance of two Titan X performance to start seeing it. For dual cards I think over time 8x/8x of course will be more restrictive but even for two TitanXs I would not worry about it.
 
https://www.pugetsystems.com/labs/articles/Titan-X-Performance-PCI-E-3-0-x8-vs-x16-851/

You can start to see differences at 8x/8x and 16x/16x with two Titan X's. Even single card 8x compared to 16x but still small. Meaning for single card PCIe 3 will probably not be limiting for at least 2 generations or more - basically you will need a single card with performance of two Titan X performance to start seeing it. For dual cards I think over time 8x/8x of course will be more restrictive but even for two TitanXs I would not worry about it.

I still play Fallout 4 and actively mod it with others. I need every ounce of performance I can get, but I won't go SLi for certain reasons. I would have liked to have seen like 32 lanes on the CPU itself, or more on the chipset. If I find I need them, Skylake X will become an option.
 
The thing that bother me is that people complain and they are using generally no M2 drive, one graphics card and no I/O slots... or the new common one: cpu, one nvm, 1 sata, 1 hdd and a Titan on a X99 board...
I have no issues with people saying they will hold off until a better chipset arrives or go intel due to Pci-E lanes for their 3x M2 drives, pci-e nic, audio, 2 titan x+++

If one likes cost to everyone, sure give em billions of pci-e lanes, but as the market is 95% will not utilize the pci-e lanes they've offered and it's not a drawback for them, it's a cost benefit.

Also worth noting is that AMD effectively have 4-8 PCI-E lanes covered with integrated usb controller, run a external PCI-E card through usb for a NVM drive is so be, for me I doubt I'll use that usb port and a 2nd nvm drive for years! like 95% of the world :)

Yeah, I'd be using 1 m.2 slot... eventually. And SATA for bulk storage on either a SSD or a HDD (as I am now).

1 GPU is ideal due to so many random performance/stability/support issues with SLI/Crossfire in general.

Other than that, I have... keyboard/mouse/microphone via USB............ and nothing else.
 
https://www.pugetsystems.com/labs/articles/Titan-X-Performance-PCI-E-3-0-x8-vs-x16-851/

You can start to see differences at 8x/8x and 16x/16x with two Titan X's. Even single card 8x compared to 16x but still small. Meaning for single card PCIe 3 will probably not be limiting for at least 2 generations or more - basically you will need a single card with performance of two Titan X performance to start seeing it. For dual cards I think over time 8x/8x of course will be more restrictive but even for two TitanXs I would not worry about it.

I see several graphs that don't jive with your conclusion in this. Most of the ones that skew towards your conclusion are like 4k x triple wide screen insanity....


x8/x8 beats x16/x16 in 1080p? Does worse FPS in 1080p than 4k with x16/x16? What?
pic_disp.php


And again with losing at 1080p? Loses by ~2.5% at 4k? But does better with x8/x8 than x16/x8?
pic_disp.php
 
looks like margin of error to me.

there are some anomalies but to draw a conclusion out of it is silly.
 
This. Ryzen is interesting because it allocates more PCIe lanes to the CPU which may or may not prove beneficial. Ryzen has the potential to be faster in multi-RAID NVMe configurations as it shouldn't have the same bottleneck DMI 3.0 creates. Time will tell what if any difference this theoretical difference makes.
This is what I'm most interested in, that and if any of the motherboards out right now will state if their u.2 connector is using the chipset or the direct CPU path.

I'm hoping during the release this is more fleshed out.
 
looks like margin of error to me.

there are some anomalies but to draw a conclusion out of it is silly.

That's definitely within a margin of error. Even if it is consistently reproducible, that hardly qualifies as a tangible difference. I'm a performance whore and I wouldn't sweat that.
 
It could be game dependent too, some games show this off better than others. [H] did a test on this a while back with different gens of PCIE and it showed to be negligible with certain games having better performance.
 
I see several graphs that don't jive with your conclusion in this. Most of the ones that skew towards your conclusion are like 4k x triple wide screen insanity....


x8/x8 beats x16/x16 in 1080p? Does worse FPS in 1080p than 4k with x16/x16? What?
pic_disp.php


And again with losing at 1080p? Loses by ~2.5% at 4k? But does better with x8/x8 than x16/x8?
pic_disp.php
Also helps if you use all the data and not single out a particular one. Of course make your own evaluation and conclusions. I think there is a difference starting to show but not very significant in the end. Also who would buy two Titans and game at 1080p :LOL:? That would be more appropriate for 4K and higher. For a game the highest difference was 6% with Grid Autosports single Titan - Now some don't mind coming in 2nd place, I rather be first ;)

For Divinci Resolve 3 of the 4 favored 16x/16x, one test was 14% faster with 16x/16x over 8x/8x -> Sorry but it is showing clearly in certain cases that 8x/8x or just using 8x is starting (STARTING NOW) to bottleneck with a TitanX. Next generation or Volta I would expect this to be more clear.
 
Also helps if you use all the data and not single out a particular one. Of course make your own evaluation and conclusions. I think there is a difference starting to show but not very significant in the end. Also who would buy two Titans and game at 1080p :LOL:? That would be more appropriate for 4K and higher. For a game the highest difference was 6% with Grid Autosports single Titan - Now some don't mind coming in 2nd place, I rather be first ;)

For Divinci Resolve 3 of the 4 favored 16x/16x, one test was 14% faster with 16x/16x over 8x/8x -> Sorry but it is showing clearly in certain cases that 8x/8x or just using 8x is starting (STARTING NOW) to bottleneck with a TitanX. Next generation or Volta I would expect this to be more clear.

I'm glad I can play DaVinci Resolve.... is that multiplayer?

It's a small margin, for an even smaller margin of the market, some "Achilles Heel". Bring it back when AMD makes a fuck up like pinning their CPU's to RD-RAM that costs more than their competitor's whole CPU+MOBO+RAM combo.
 
Last edited:
  • Like
Reactions: noko
like this
It could be game dependent too, some games show this off better than others. [H] did a test on this a while back with different gens of PCIE and it showed to be negligible with certain games having better performance.
Since those tests there are more powerful GPU's that require more bandwidth - results previously done don't necessarily reflect what will happen now. Otherwise we would never have to go beyond PCIe 1.0 or 2.0, if a more powerful GPU's will always used the same bandwidth. Now what AMD is doing, if they are successful is to use the bandwidth more efficiently - as in loading only what is needed and not a whole bunch of other stuff constantly; for example just one small mipmap texture vice all the mipmap 4K textures and smaller for a far away distant object.

As for PCIe 16x and a single GPU - I don't see it as being limiting in the next 3-5 years. 8x/8x could be more limiting next gpu generation for multiple top end GPU's.
 
Back
Top