AMD Bristol/Stoney Ridge Thread

[Shehriazad]

IIRC, there will be a 2400 hard limit with these chips?

Well yea...was more speaking of the future. Also, I could understand if "mobile" BR won't really be able to benefit much past 2400...but Desktop?

I mean Desktop Kaveri could benefit all the way up to DDR3 3000...why exactly would BR be different? Or am I missing some architectural change?

 

[majord]

The only thing you're missing is the lossless delta colour compression in BR, which by AMDs own estimates gives about 5 to 6% relief, or equivalent to extra bandwidth, depending how you look at it.

So yes it would still benefit from faster RAM without doubt.

The top RAM clock is a disappointment, plain and simple, especially when the likes of apple are pushing 50gps with lpddr4. (though not directly comparable due to the nature of supporting standard dimms vs a custom on board design)

Given AMD's position of compromise though, it's understandable, and hardly a disaster given it will still keep them on top of the mainstream igp game.

In other words, its more important they get out a stable product, on time without steering recourses away from Zen based development. Rather than push memory standards to the limit .

If they're not pushing it with raven ridge though, well that would be a real problem IMO

 

[deasd]

Well yea...was more speaking of the future. Also, I could understand if "mobile" BR won't really be able to benefit much past 2400...but Desktop?

I mean Desktop Kaveri could benefit all the way up to DDR3 3000...why exactly would BR be different? Or am I missing some architectural change?

I couldn't find bench about APUs pair with lot more than DDR3 2400. Could you provide links/source about this?
I assume if there's no or little hard limit inside APUs, the IGP should operate at higher bandwidth than uncore, I guess IGP use dedicated channel through memory which we don't know, and very likely it could't be adjusted through any BIOS/software.

 

[cbn]

Given AMD's position of compromise though, it's understandable, and hardly a disaster given it will still keep them on top of the mainstream igp game.

Remember AMD are also releasing Polaris 11 which will be low power, better performance, and have a better feature set compared to Bristol Ridge's iGPU. This, not to mention, whatever Nvidia plans on launching for lower end desktop (eg, another cut down GM206 variant and/or GP108). This, in addition, to the projected glut of used 28nm cards that will hit the second hand market after various people upgrade to FinFET dGPU.

Furthermore, Remember Intel has GT4e coming to desktop which will be faster than Bristol Ridge so that very tiny niche is gone also.

Then after factoring all of the above, consider the BGA version of Bristol Ridge at 35W or 45W would be just as fast as the 65W AM4 Bristol Ridge in most scenarios.

All in All, it seems to me it is almost impossible to justify Bristol Ridge for AM4. (Athlon x4...yes, dual core with 384+sp....yes. Quad core with 512sp....No)

P.S. Regarding the noise level of a 35W Bristol Ridge mobile chip in a laptop. Yes, it will be louder than the 15W Bristol Ridge (all things being equal).....but I think many people would judge the performance boost to be quite impressive (much greater than say Stoney Ridge at 35W vs. 15W). With this noted, I'm sure the laptop makers could make a 35W notebook quieter (at the compromise of either size or battery life, etc)

 

[majord]

Remember AMD are also releasing Polaris 11 which will be low power, better performance, and have a better feature set compared to Bristol Ridge's iGPU. This, not to mention, whatever Nvidia plans on launching for lower end desktop (eg, another cut down GM206 variant and/or GP108). This, in addition, to the projected glut of used 28nm cards that will hit the second hand market after various people upgrade to FinFET dGPU.

Furthermore, Remember Intel has GT4e coming to desktop which will be faster than Bristol Ridge so that very tiny niche is gone also.

Then after factoring all of the above, consider the BGA version of Bristol Ridge at 35W or 45W would be just as fast as the 65W AM4 Bristol Ridge in most scenarios.

All in All, it seems to me it is almost impossible to justify Bristol Ridge for AM4. (Athlon x4...yes, dual core with 384+sp....yes. Quad core with 512sp....No)

P.S. Regarding the noise level of a 35W Bristol Ridge mobile chip in a laptop. Yes, it will be louder than the 15W Bristol Ridge (all things being equal).....but I think many people would judge the performance boost to be quite impressive (much greater than say Stoney Ridge at 35W vs. 15W). With this noted, I'm sure the laptop makers could make a 35W notebook quieter (at the compromise of either size or battery life, etc)

I agree Polaris 11 and nv equivalent will put a dent in BR's appeal, but remember , you can't even buy a mainstream CPU without an IGP, (with the exception of athlons which you can't even buy in the channel here) so the competition isn't really dGPU's anyway.

As for 35w being as fast , I don't think so. 65w will sustain higher CPU and gpu frequencies when both are loaded to the brink by a modern game

 

[cbn]

I agree Polaris 11 and nv equivalent will put a dent in BR's appeal, but remember , you can't even buy a mainstream CPU without an IGP, (with the exception of athlons which you can't even buy in the channel here) so the competition isn't really dGPU's anyway.

For AMD it is a bit of a moot point as their pricing in OEM desktops is not competitive:

http://forums.anandtech.com/showpost.php?p=37937094&postcount=343

 

[cbn]

As for 35w being as fast , I don't think so. 65w will sustain higher CPU and gpu frequencies when both are loaded to the brink by a modern game

The 35W can be set at 45W via cTDP.

And at 45W or 65W, memory bandwidth will be the limiting factor in gaming situations. (The probable exception being CPU intensive games like GTA V)

 

[Shehriazad]

I couldn't find bench about APUs pair with lot more than DDR3 2400. Could you provide links/source about this?
I assume if there's no or little hard limit inside APUs, the IGP should operate at higher bandwidth than uncore, I guess IGP use dedicated channel through memory which we don't know, and very likely it could't be adjusted through any BIOS/software.

Well, the highest I ever saw "confirmed" to run was "DDR3-2711". BUT I have to admit that a lot of boards would not properly let the system boot if you went past 2400 or at least had to use BCLK.

I do remember some site that managed to test up to 2800 but I absolutely fail to find it again...so I guess I'm not very reliable. On my old 7850K I actually ran 2590ish speeds and games actually still showed small performance differences.

The performance scaling from 2400 up to 2600 actually seemed pretty linear, if I recall the performance scaling drops down past that....but CONTINUES to affect it.


So it's not about it not having a benefit...it's about it no longer running stable/booting at all. If AMD worked out the stability issues even 3000mhz would show visible perf gains.

 

[cbn]

If AMD worked out the stability issues even 3000mhz would show visible perf gains.

Yes, Remember HD7750 (512sp @ 800 MHz) with either DDR3 (1600) or GDDR5 (4500):

http://www.hardware.fr/focus/76/amd-radeon-hd-7750-ddr3-test-cape-verde-etouffe.html

(Up to 50% lower frame rates with DDR3! Or another way of phrasing that would be that HD7750 GDDR5, in some cases, had almost double the FPS compared to the DDR3 version!)

 

[The Stilt]

On A10-7870K or A10-7890K (866MHz iGPU) > 25% of the graphics performance is lost even at the maximum possible memory bandwidth (2400MHz, dual rank). The performance difference between single and dual rank modules is pretty much worth a single speed class on APUs; DR modules at 2133MHz match SR modules at 2400MHz.

The only way to increase the memory frequency from 2400MHz on 15h designs is to increase the BCLK frequency. On APUs everything run from the BCKL, so it is certainly not a good idea. The AHCI in FCH is lost at 104 - 106MHz BCLK, analog signals are lost at > 110MHz and certain PCI-E devices can get confused too. AMD dGPUs usually handle ~120MHz in PCI-E 3.0 mode and higher in older modes.

 

[zentan]

Just as though, no unlocked SKUs it seems

Yea that seems the way with excavator on desktop at least for now.
You have put out lot of info about AMD APUs/CPUs in last 1 or so year in the forums and most of what we are seeing is going the direction that some of your results showed and some other good info shared by you.
For all these insights I think a lot of people are thankful to you.

 

[cbn]

One thing I am concerned about (and I mentioned it in the other thread) is how a person might try to predict or extrapolate gaming performance of the 35W BR vs. 15W BR based on the closeness of the advertised CPU speeds:

But what happens during actual gameplay? What kind of CPU (and GPU) clocks will the 15W have under load? (I am guessing much lower than the 35W)

Worst case scenario: The gaming oriented person is not satisfied with the 15W BR and (based on predicting/extrapolating from marketing specs) also writes off the idea of using the 35W model as well (thinking there is very little chance it could be significantly better than the 15W).

EDIT: Perhaps for advertised CPU frequency there could be three numbers used instead of the two used now. (Example: "Max/Base" CPU frequency changed to "Max/Base/Gaming load" CPU frequency)

 

[cbn]

In other words, its more important they get out a stable product, on time without steering recourses away from Zen based development. Rather than push memory standards to the limit .

If they're not pushing it with raven ridge though, well that would be a real problem IMO

The memory controller (which they get from Synopys) should be faster on Raven Ridge.

According to the Synopys website (link here) the latest IP is up to DDR4 3200 ....but I don't know if this will be the one used or not? It depends on when it was released and how long it takes for AMD to integrate it. It could Raven Ridge uses a slower controller.

 

[DrMrLordX]

DDR4-3200 is already becoming old hat. Gotta do better than that!

 

[ET]

I decided to estimate how gaming capable Stoney Ridge is.

I have a Radeon 5550 with 800MHz DDR2 (according to HIS) in my HTPC. It's used for family gaming, and copes with LEGO games at 720p, and the likes of Never Alone, Ducktales, Brothers - A Tale of Two Sons, King's Quest, etc. In short, a pretty decent family gaming card, even though it's somewhat borderline and it would be nice to get some more performance (like LEGO at 1080p) and I plan to upgrade it to Bristol Ridge AM4.

So how would the A9-9410 compare to my Radeon 5550?

I took the Radeon 7750 vs. 5770 as an estimate of the architectural difference. They performed about the same when the 7750 was released, at close core and memory speeds. So I estimate from this that the 5770's 800 cores is equivalent to the 7750's 512 cores.

The A9-9410 has 192 cores, so by this ratio it's equivalent to 300 cores of the 5000 family. This is very close to the Radeon 5550's 320 cores. But the 5550 has a core clock of 550Mhz, while the A9-9410 has a 800Mhz clock, which in total gives it 36% better performance than the 5550. Even the A6-9210 has a 600MHz clock, which makes it about equivalent to the 5550.

When Stoney Bridge has a problem is memory bandwidth. The 5550 has DDR3-1600 with a 128 bit bus, compared to Stoney Bridge's DDR4-2133 with a 64 bit bus. End result is 67% of the 5550's bandwidth. Other the other hand, GCN 1.2 has lossless compression which should help alleviate the bandwidth problem. On the gripping hand, this bandwidth it shared with the CPU.

So bandwidth would be an issue, and it's hard to estimate its impact. However, I'd still estimate Stoney Bridge can be about as fast as my 5550, which isn't that bad. The 5550 is certainly not a gamer's card, but it functions well for family gaming.

So this puts these chips in some more perspective for me. All in all, nice to see how far along AMD's low end chips have come since the E-350 in my Thinkpad X120e. If Lenovo released a new low cost Stoney Bridge based Thinkpad X, I would have considered getting it as an upgrade. (Although I'd be surprised if that happens, and I already ordered a Dell XPS 13, so that's just in theory.)

Edit: Nothing new about Bristol Ridge AM4 and I'm a little bored.

Re-read this and noticed "800MHz DDR2" up top and "DDR3-1600" in the calculation. I think that HIS is wrong. AMD's site specifies RAM as DDR3 at 800MHz, which is DDR3-1600 (that's consistent with the bandwidth mentioned). I therefore assume that the card's memory runs at an effective 1600MHz.

Also, I calculated that the A9-9410's GPU is 6x as fast as the Radeon 6310 in the E-350 (based on the architecture core equivalence assumption above and clock speed) and it has twice the memory bandwidth, before taking compression into account.

 

[monstercameron]

Truly too bad netbooks died...

 

[cbn]

VP9 decode is OpenCL based on both Carrizo (UVD 6.0) and Stoney Ridge (UVD 6.1).
Even the initial GCN4 based designs won't support VP9 HW decoding.

Most likely a highly optimized 3rd party SW codec (such as LibAV) will provide significantly better experience than the OpenCL solution. That was the case with Strongene HEVC OpenCL decoder at least.

I asked AMD which softwares currently support the acceleration, let's see what they say. Hopefully it is a driver feature at least.

Have you got a response yet from AMD?

 

[dark zero]

Truly too bad netbooks died...

Intel Atom killed them with their mediocre products until today (Bay Trail being the sole exception)

 

[monstercameron]

Intel Atom killed them with their mediocre products until today (Bay Trail being the sole exception)

What?

 

[The Stilt]

Have you got a response yet from AMD?

Yes and no...
Let's just say we need to wait for the official announcement (no idea when will this happen).

But the UVD / VCE related information at https://www.x.org/wiki/RadeonFeature/ is absolutely correct

 

[Abwx]

Yes and no...


But the UVD / VCE related information at https://www.x.org/wiki/RadeonFeature/ is absolutely correct

No crossfire support since the R500 to Island chips according to this site but we should be confident that it s up to date for features that are way more recent...

Yeah, absolutely correct, no doubt...

This page is only for free Radeon drivers using KMS. radeon (xf86-video-ati) for 2D; radeon, r200 Mesa and r300, r600, radeonsi Gallium drivers only. THIS PAGE IS NOT FOR FGLRX/CATALYST DRIVERS PROVIDED BY AMD/ATI.

 

[monstercameron]

No crossfire support since the R500 to Island chips according to this site but we should be confident that it s up to date for features that are way more recent...

Yeah, absolutely correct, no doubt...

huh? the guys who maintain the kernel drivers also maintain these pages.

 

[cbn]

There is no mistake, 15W TDP SKUs will only use 1866MHz speed.

With the 15W Stoney Ridge rated at DDR4 2133, I do wonder if there are some cases where it would be faster than a 15W Bristol Ridge laptop (shipped with a single SO-DIMM running at DDR4 1866).

Maybe particularly with older games (or newer games like War Thunder) that only load one or two CPU cores? (But then BR would still have better floating point due to the ability to use one integer core for each module for games that primarily load two cores)

I guess it depends on how dependent the game @ X resolution/detail settings is on memory bandwith vs. CPU? (And then if CPU, how well it can spread the load out?)

 

[dark zero]

http://www.notebookcheck.net/Bristo...0-9600P-Against-the-Competition.168477.0.html

Ok, time to revive the thread.

I need to say something first.
The way AMD is trying to match their performance is like this:
Core i7 -- FX
Core i5 -- A12
Core i3 -- A10
Pentium U -- A9
Celeron U/Pentium N -- A6
Celeron N -- E2


The only real match from AMD is the E2. In my opinion, the way on how they are need to compare are like this:

Core i7 / Unmatched (need to say more?)
Core i5 / AMD FX
Core i3 / AMD A12
Pentium U / AMD A10 (due ST performance)
Celeron U / AMD A9 (due threads available)
Pentium N / AMD A6 (due cost)
Celeron N / AMD E2

Using the comparison from AMD, is not a good product, it falls behind even Broadwell. However using the realistic way (comparing to Pentium U), it has a good chance to be a decent competitor.

 

[majord]

well it is the bottom of the pack in the A10s

..but anyway, an interesting point of discussion in that test - the dGPU, which is an old 320SP Oland @ 844Mhz boost, fed with just 64bit DDR3 dedicated - thats a measly 16GB/s. And they just happend to test it on its own next to the iGPU's which is extremely convienent.

and it easily beats the shit out of the r5 iGPU. When you look at the sustained clockspeeds of the iGPU it's easy to see why. (400-500Mhz)

Would like to see clockspeeds and performance of the higher SKU's with the full 512SP, as i'm sure with better binning they will be somewhat higher but it seems to me the DDR4-1866 isn't the main hindrance to performance. Especially when when we're talking about a GCN1.3 architecture here - which is much less sensitive to bandwidth than GCN1.0 Oland DGPU

Seems to me AMD need 14nm Polaris based IGP's more than anything in the 15w envelope though - the frequency uplift will be massive based on what we've seen at the pointy (i.e unfavorable) end of the spectrum with Polaris 10