Ryzen 5 coming April 11th, 6c12t and 4c8t options available

[juri-han]

39 minutes ago, lots of unexplainable lag said:

From the article:

 

"We have confirmation from AMD that there are no silly games going to be played with Ryzen 5. The six-core parts will be a strict 3+3 combination, while the four-core parts will use 2+2. This will be true across all CPUs, ensuring a consistent performance throughout."

 

Unless AMD can fix the super slow communication speeds between CCXs I fear that the 4c8t R5 parts won't really hold up to the i5 until games/Windows realize how Ryzen is set up. 

cant you already get 7700k performance by turning on performance mode? 1600x/1500x/1400 should be the same and once MS rolls out of their bed made out of $100 bills and optimise windows then add the extra threads you would have to be crazy to buy a i5 till they add hyper threading 

[Droidbot]

1 minute ago, juri-han said:

cant you already get it 7700k by turning on performance mode? 1600x/1500x/1400 should be the same and once MS rolls out of their bed made out of $100 bills and optimise windows then add the extra threads you would have to be crazy to buy a i5 till they add hyper threading 

The CCX issue makes Ryzen have very low performance with apps that take advantge of core to core communication like Watch Dogs 2, which takes advantage of every thread you chuck at it happily. 

 

 

 

[MyName13]

One would think that these CPUs are just good enough for gaming and nothing else when it's the opposite, why does AMD talk so much about gaming anyway?

[MageTank]

7 hours ago, stconquest said:

Haven't watched it yet, but: 

 

 

"Consumers wanted faster memory, so here we are working on that. This is 2933 vs 2133". Unless they made 2933 work with 2DPC, this removes the last ounce of hope I had for this IMC getting fixed. We already have 2933 on Ryzen 7 in 1DPC. We even have people running 3200-3400 with single rank, 1DPC setups. I feel like AMD is mocking us at this point x.x

[stconquest]

20 minutes ago, MageTank said:

"Consumers wanted faster memory, so here we are working on that. This is 2933 vs 2133". Unless they made 2933 work with 2DPC, this removes the last ounce of hope I had for this IMC getting fixed. We already have 2933 on Ryzen 7 in 1DPC. We even have people running 3200-3400 with single rank, 1DPC setups. I feel like AMD is mocking us at this point x.x

They might not get "fixed" it for a while... they might not get done it until the next CPU launch (Zen2/whatever).  It has to be a concern for them.  The vid was showcasing less than what we already know on these forums.

 

You know IMCs better than I.  I still have hope that the IMC can be fixed, if not... 2DPC is the only way to get the most out of the platform. 

[MageTank]

Just now, stconquest said:

They might not get "fixed" it for a while... they might not get done it until the next CPU launch (Zen2/whatever) launch.  It has to be a concern for them.  The vid was showcasing less than what we already know on these forums.

 

You know IMCs better than I.  I still have hope that the IMC can be fixed, if not... 2DPC is the only way to get the most out of the platform. 

I could almost stomach 2DPC not working, but the fact that multi-rank kits have to be crippled in order to even post, bothers me greatly. Rank Interleaving is a boon I benefit greatly from. If I have to cripple the clocks and timings to get rank interleaving, then it completely defeats the purpose of having it. If they'd give us access to those timings, I know for a fact that we could fix it, or work around it. It's the only thing I ask from AMD at this point. 

[Dabombinable]

1 minute ago, stconquest said:

They might not get "fixed" it for a while... they might not get done it until the next CPU launch (Zen2/whatever) launch.  It has to be a concern for them.  The vid was showcasing less than what we already know on these forums.

 

You know IMCs better than I.  I still have hope that the IMC can be fixed, if not... 2DPC is the only way to get the most out of the platform. 

I know this-AMD's had shit IMC for a while (dual channel DDR3 1333 with an A8 4555M as slow as dual channel DDR3 800 with an i5 470UM)

[stconquest]

2 minutes ago, MageTank said:

I could almost stomach 2DPC not working, but the fact that multi-rank kits have to be crippled in order to even post, bothers me greatly. Rank Interleaving is a boon I benefit greatly from. If I have to cripple the clocks and timings to get rank interleaving, then it completely defeats the purpose of having it. If they'd give us access to those timings, I know for a fact that we could fix it, or work around it. It's the only thing I ask from AMD at this point. 

Rank-Interleaving is used in 8GB capacity sticks and up, right?  I did not even know that Ryzen was having trouble with that. 

[stconquest]

2 minutes ago, Dabombinable said:

I know this-AMD's had shit IMC for a while (dual channel DDR3 1333 with an A8 4555M as slow as dual channel DDR3 800 with an i5 470UM)

It is really hard to tell what a new gen will bring though, even if past gens were lacking.  That is the entire point of making changes, to improve.

[MageTank]

Just now, stconquest said:

Rank-Interleaving is for 8GB capacity sticks and up, right?  I did not even know that Ryzen was having trouble with that. 

For DDR3, yes. On DDR4, 8gb sticks can still be single rank. 16GB is the only guaranteed dual-rank sticks for DDR4. As for the multi-rank info...

 

[Dabombinable]

1 minute ago, stconquest said:

It is really hard to tell what a new gen will bring though, even if past gens were lacking.  That is the entire point of making changes, to improve.

The trend has been clear for a while though. For example, even with an IMC K10 had worse memory performance than the NBMC found in Conroe era motherboards. AMD will always be a bit behind when it comes to the memory controller (unless they've managed to get their shit together with Naples).

[Dabombinable]

I'll put up a link to the results I found with Aida 64 memory bench with my various CPU as an example (though it won't have the results from my Phenom II P920 and N970-the laptop doesn't have a HDD at the moment). I've only gotten the cache speed table written up so far (averaged across 5 runs, put into a spreadsheet using a top-of-the-line 2003/2004 Compaq Presario R3000 with Office 2007)

[stconquest]

1 minute ago, MageTank said:

For DDR3, yes. On DDR4, 8gb sticks can still be single rank. 16GB is the only guaranteed dual-rank sticks for DDR4. As for the multi-rank info...

 

Hmmm, should be interesting to see if 2x16GB has frequency limitations due to rank interleaving... or how much 8GB sticks would suffer in dual rank mode (when possible)

 

1 minute ago, Dabombinable said:

The trend has been clear for a while though. For example, even with an IMC K10 had worse memory performance than the NBMC found in Conroe era motherboards. AMD will always be a bit behind when it comes to the memory controller (unless they've managed to get their shit together with Naples).

Damn you go way back.    I did not know the memory performance was comparatively bad back then.

[Dabombinable]

6 minutes ago, stconquest said:

Hmmm, should be interesting to see if 2x16GB has frequency limitations due to rank interleaving... or how much 8GB sticks would suffer in dual rank mode (when possible)

 

Damn you go way back.    I did not know the memory performance was comparatively bad back then.

Oh it was. Wait until you see how abysmal Trinity (September 2012) was compared to Arrandale (November 2010)

Here's the release date+info (shame I no longer have access to an i5 540M):
http://www.cpu-world.com/Compare_CPUs/AMD_AM4555SHE44HJ,Intel_CN80617005190AI/

[leadeater]

5 minutes ago, Dabombinable said:

The trend has been clear for a while though. For example, even with an IMC K10 had worse memory performance than the NBMC found in Conroe era motherboards. AMD will always be a bit behind when it comes to the memory controller (unless they've managed to get their shit together with Naples).

Problem is I highly doubt anything relating to the memory controller on Naples will make it down to desktop platform, maybe never. The Naples platform is based around Gen-Z which is likely to be far too costly to implement in desktop chips.

 

My understanding is you also have to program specifically in mind for Gen-Z and that just won't be done by game developers and all hardware in the chain must have Gen-Z support inbuilt.

 

http://genzconsortium.org/about/

[Dabombinable]

3 minutes ago, leadeater said:

Problem is I highly doubt anything relating to the memory controller on Naples will make it down to desktop platform, maybe never. The Naples platform is based around Gen-Z which is likely to be far too costly to implement in desktop chips.

 

My understanding is you also have to program specifically in mind for Gen-Z and that just won't be done by game developers and all hardware in the chain must have Gen-Z support inbuilt.

 

http://genzconsortium.org/about/

If Ryzen's Naples implementation is anything like Haswell/Boradwell E, it would probably happen but 2-3 years latter.

[leadeater]

4 minutes ago, Dabombinable said:

If Ryzen's Naples implementation is anything like Haswell/Boradwell E, it would probably happen but 2-3 years latter.

Only problem is the memory controller on those is rather standard, be it quad channel though.

 

Gen-Z on the other hand has no support from Intel or Nvidia so that makes it a non starter for the desktop world on that alone, increased effort in coding for different platforms and the hardware requirement. You'll get no benefit from Naples/Gen-Z with an Nvidia GPU.

 

Edit:

Treat Gen-Z like NVlink, both have similar constraints and full implementations will only be found in servers.

[The Benjamins]

On 3/15/2017 at 6:05 AM, huilun02 said:

Don't know why they got to give the 4c such miserable 200Mhz turbo. Also this means R3 lineup will be 2c4t and 4c4t. TDP not as low as I'd like because the 1700 is already 65w. Maybe just conservative numbers.

at least the 1500x with XFR can in theory do 3.7/3.9 without any manual OC, and with hopefully easier to manage temps we can get it slightly past 4.1GHz

[Valentyn]

2 hours ago, lots of unexplainable lag said:

From the article:

 

"We have confirmation from AMD that there are no silly games going to be played with Ryzen 5. The six-core parts will be a strict 3+3 combination, while the four-core parts will use 2+2. This will be true across all CPUs, ensuring a consistent performance throughout."

 

Unless AMD can fix the super slow communication speeds between CCXs I fear that the 4c8t R5 parts won't really hold up to the i5 until games/Windows realize how Ryzen is set up. 

 

It's a real concern, and bar upping memory Frequencies it'll be very hard for a normal consumer to try and address the issue.

The Interconnect is apparently only 22GBs according to Josh Walrath, where as Infinity Fabric would apparently scale as high as 512GBs.

AMD mentioned several times they know exactly where they can get better performance for Zen2, and I believe sorting out that interconnect and that 50% Memory Speed cap would really help.

Ryzen has the IPC, we see it in production work, but the latency is killing performance when it really should not. 

I guess the saving grace is that Zen2 is socket compatible with AM4.

[The Benjamins]

5 minutes ago, Valentyn said:

 

It's a real concern, and bar upping memory Frequencies it'll be very hard for a normal consumer to try and address the issue.

The Interconnect is apparently only 22GBs according to Josh Walrath, where as Infinity Fabric would apparently scale as high as 512GBs.

AMD mentioned several times they know exactly where they can get better performance for Zen2, and I believe sorting out that interconnect and that 50% Memory Speed cap would really help.

Ryzen has the IPC, we see it in production work, but the latency is killing performance when it really should not. 

I guess the saving grace is that Zen2 is socket compatible with AM4.

Lets not forget that R5 has A LOT more cache then intels parts, which will help.

[Valentyn]

Just now, The Benjamins said:

Lets not forget that R5 has A LOT more cache then intels parts, which will help.

 

It certainly well, it's just a damn shame they went with two CCXs really. 
It does make me wonder how well the up coming APUs will be.

Single CCX + graphics. Will they be a little bit faster in gaming then normal R5 dual CCX counter parts? Who knows.

 

In the meantime AMD and motherboard manufacturers really need to sort out the RAM issues. Higher frequencies are shown time and time again to really help Ryzen's gaming performance; and it really needs it.

Atleast G.Skill have started putting out AMD specific ram kits that will run properly, as with Intel XMP kits it's rather random.

[grimreeper132]

On ‎15‎/‎03‎/‎2017 at 8:40 AM, dkspooder said:

So with the whole kerfuffle about the scheduler, and the CCX rumors, I'd like to say that I want to think positively about the 4c8t models. If AMD manages to significantly improve clock speeds with less cores on the package, we may see their equivalent to the 6700k, or even the 7700k. Ryzen 5 is not meant to compete with the Extreme edition processors, so hopefully it's at least price competitive to the consumer i7. 

well the 6/12 sorta does compete with some of the extreme, due to the same number of cores, but the fact it's significantly cheaper will mean that it's a better choice for workstation PCs, as performance looks like it'll be good, maybe not as good, but still good, especially if you take the pricing into consideration, and the fact that if you were willing to pay the intel tax for their 6 core you would then realise that there is an 8 core which is a better workstation CPU at that price and go for it. it looks like intel is going to get beaten in one area of the market of the workstation PCs, maybe not for gaming, but certainly there, and that will mean AMD might break even, maybe even make a large profit and then down the line start expanding into gaming CPUs and other areas

[SpaceGhostC2C]

3 hours ago, MageTank said:

For DDR3, yes. On DDR4, 8gb sticks can still be single rank. 16GB is the only guaranteed dual-rank sticks for DDR4. As for the multi-rank info...

 

I wouldn't treat those as hard limits, though. Piledriver DDR3 official support is 1866 in 2 dimms, 1600 in 4 dimms, and even 1600 in 2 dimms if the board has 4 slots (¿?). Yet you can run any DDR3 speed, even in 4-dimm configurations. Granted, motherboard manufacturers were advertising higher speeds than the official ones, but so they do now as well. I'm under the impression 2667 is the 1866 of their previous platforms, i.e., the highest number they allow manufacturers to put without adding (OC) next to it  

For comparison, Intel is even more conservative, officially stating 1600 in their DDR3 platforms, and 2133 (Skylake) and 2400 (Kaby Lake) DDR4. Yet we know that's not the limit...

[Spork829]

HYYYPPPPPEEEEE. Glad I've sat on my Z97 motherboard and locked i5 for a year patiently waiting for the 4790K price to drop, which it'll pretty much have to do unless the 1500X/1400 have serious issues. And if Intel doesn't lower it then screw 'em, I'll just get a new mobo and Ryzen chip.

[weez]

Intel will still have top performance for gaming but with the pricing of the R5 AMD has the potential to have some of the best performance per dollar we've seen in a long time. Obviously we will need to wait for benchmarks but this is exciting.