Guess what product AMD is revealing!

[strata8]

It will be very interesting to see how the AMD A10 Micro-6700T performs in reality.

Possibly the dumbest name for a chip ever, though. I've seen retailers list the A4-5000 as the A4M-5000B so god knows how this will turn out.

 

[Vesku]

Nowhere do they indicate 2x performance per watt. It is interesting how they skirt around performance per watt with stuff as ridiculous as "productivity per watt". LOL. Really. That's certainly a new one. But. You won't find performance per watt anywhere in those slides. Just skewed stuff like "graphics" performance per watt (really??) and productivity per watt. Since AMD is so heavily skirting around the industry standard metric for mobile, you gotta wonder how actual devices will pan out. I'd say that since they're skirting around performance per watt so intentionally in their slides, that will probably show up in the chip. And not in a good way.

Because they missed their performance targets set in their preview, most likely to at least partially preserve efficiency claims. 15W TDP Beema was supposed to be ~20-24% faster in PC Mark 8 than 25W TDP A6-5200. In the leaked slides it scores slightly less. They almost made their performance target in 3DMark11 so the GPU portion is a nice improvement over Kabini.

 

[Shivansps]

Actually GPU improvement is most likely because of DDR3-1866, unless its updated to GCN 2.0, and thats the other problem, its unlikely to see that kind of memory on mobile devices.

 

[OCGuy]

Looks to be a TDP shell game, but hey if it gets them some OEM wins, then they are doing their job over there in marketing.

I don't know if I would be too excited, since basically they are getting closer to a target originally promised.

 

[blackened23]

Actually GPU improvement is most likely because of DDR3-1866, unless its updated to GCN 2.0, and thats the other problem, its unlikely to see that kind of memory on mobile devices.

So does it not support LPDDR3? And yeah, DDR3-1866 on a tablet device is a pretty ridiculous thought. Actually it sounds like this product will only be considered for sub notebooks or something along those lines.

 

[blackened23]

Because they missed their performance targets set in their preview, most likely to at least partially preserve efficiency claims. 15W TDP Beema was supposed to be ~20-24% faster in PC Mark 8 than 25W TDP A6-5200. In the leaked slides it scores slightly less. They almost made their performance target in 3DMark11 so the GPU portion is a nice improvement over Kabini.

The graphics improvements are nice, but with mobile devices it really is about the entire package. The entire package is what consumers care about - display resolution (this is COMPLETELY reliant on performance per watt, as it is a battery drainer), battery life, stuff along those lines. So with that in mind, performance per watt and total TDP seem to be the most relevant metrics which everyone uses.

I guess whenever the desktop portion of this chip is released it could be a decent low cost SOC. Better than Kabini on AM1 certainly, with a decent graphics bump. But, on a mobile device, i'm just not seeing it. I could be wrong though, hopefully we'll get some real (mobile) devices soon utilizing the chip. Then we can truly judge the merits of it.

 

[rtsurfer]

So does it not support LPDDR3? And yeah, DDR3-1866 on a tablet device is a pretty ridiculous thought. Actually it sounds like this product will only be considered for sub notebooks or something along those lines.

It says 4.5W TDP here.

Edit: You edited your post to remove the 15W from there.

Didn't read the slides, did you. ?

 

[blackened23]

No, I confused beema and mullins. Or maybe you're confusing them. One is for up to 25W TDP (configurable), the other is for 5W TDP. One obviously can't be used for tablets (10-25W TDP) while 5W TDP is the tablet TDP window.

Similar to Kabini and Temash. Those products both missed the mark for mobile, but one was for subnotebook (Kabini, configurable TDP 10- 25W IIRC) and tablets (Temash, 5-10W TDP). Same thing here. Beema and Mullins. One is low TDP the other is high subnotebook TDP. I keep confusing which is which.

Either way, you're not going to see 5W TDP chips paired with DDR3-1866 in a mobile device. Hence the question about LPDDR3.

 

[rtsurfer]

No, I confused beema and mullins. Or maybe you're confusing them. One is for up to 25W TDP (configurable), the other is for 5W TDP. One obviously can't be used for tablets (10-25W TDP) while 5W TDP is the tablet TDP window.

Similar to Kabini and Temash. Those products both missed the mark for mobile, but one was for subnotebook (Kabini, configurable TDP 10- 25W IIRC) and tablets (Temash, 5-10W TDP). Same thing here. Beema and Mullins. One is low TDP the other is high subnotebook TDP. I keep confusing which is which.

Either way, you're not going to see 5W TDP chips paired with DDR3-1866 in a mobile device. Hence the question about LPDDR3.

Fair enough.
It says DDR3L here.

I don't think they can include that fast RAM in a tablet.

Edit: Actually looks like the low TDP part will have DDR3-1333 & higher one will run 1866.

 

[monstercameron]

Fair enough.
It says DDR3L here.

I don't think they can include that fast RAM in a tablet.

Edit: Actually looks like the low TDP part will have DDR3-1333 & higher one will run 1866.

I read somewhere that the nvidia shield uses 1866Mhz ram, or atleast the modules do support that operation.
http://www.ifixit.com/Teardown/Nvidia+Shield+Teardown/16212
module found here^
http://www.skhynix.com/products/computing/view.jsp?info.ramKind=19&info.serialNo=H5TC4G63AFR
links to here ^
[pdf]
page 16

Rev. 1.0 / Apr. 2013
16
Table 1 -Timings used for IDD and
IDDQ Measurement-Loop Patterns
Table 2 -Basic IDD and IDDQ Measurement Conditions
Symbol
DDR3L-1066
DDR3L-1333
DDR3L-1600
DDR3L-1866
Unit
7-7-7
9-9-9
11-11-11
13-13-13
t
CK
1.875 1.5 1.25 1.07 ns
CL 7 9 11 13 nCK
n
RCD
7 9 11 13 nCK
n
RC
27 33 39 45 nCK
n
RAS
20 24 28 32 nCK
n
RP
7 9 11 13 nCK
n
FAW
1KB page size 20 20 24 26 nCK
2KB page size 27 30 32 33 nCK
n
RRD
1KB page size 4 4 5 5 nCK
2KB page size 6 5 6 6 nCK
n
RFC
-512Mb 48 60 72 85 nCK
n
RFC
-1 Gb 59 74 88 103 nCK
n
RFC
- 2 Gb 86 107 128 150 nCK
n
RFC
- 4 Gb 139 174 208 243 nCK
n
RFC
- 8 Gb 187 234 280 328 nCK

 

[rtsurfer]

I read somewhere that the nvidia shield uses 1866Mhz ram, or atleast the modules do support that operation.
http://www.ifixit.com/Teardown/Nvidia+Shield+Teardown/16212
module found here^
http://www.skhynix.com/products/computing/view.jsp?info.ramKind=19&info.serialNo=H5TC4G63AFR
links to here ^
[pdf]
page 16

Yes.
Also I know that my current phone, Samsung Galaxy S3 has its RAM clocked at 400Mhz LPDDR2. The RAM speed was raised to 533Mhz in the revised SOC included in the Galaxy Note 2.
These are products that were released 2 years ago.
Moreover both of them are phones.

I believe by the time AMD's SOC hits the market we would have enough technology to run the low powered memory in a tablet at 1333Mhz.

 

[Enigmoid]

...

Why I hate PC mark 8.

Non-linear performance increases. Not only does the A6-5200 have twice the number of cores but it runs at twice the frequency as the E1-2100. It has four times the theoretical performance but its score is not even twice as high. The a6-5200 is 33% faster than the a4-5000 but its pcmark 8 score is only 13% higher.

Its also too much of a system benchmark which while as a good indicator of system performance (debatable if openCL is even relevant for these systems) is too prone to flukes (look at IVB and its ridiculous scores because of the change in GPU). For instance

I'm not sure what this means for beema and mullins.

The systems are using different drivers (You previously posted this and the beema platform is using an 'unreleased' driver). This is definitely going to mess up pcmark 8 depending on OpenCL acceleration.

BTW can anyone find the E1-2100 with 8230?

http://products.amd.com/pages/NotebookAPUDetail.aspx?id=70

AMD lists it as having the 8210.

Only AMD knowns that.

Those look like SOC power measurements. If they are using their 'larne' reference platform the SSD + RAM is going to use more than those e-reader numbers. Though for instance those 1080p playback numbers look really high.

Anyway this looks good. There is a lot of room for improvement in kabini. I highly doubt though that these chips will be able to touch baytrail in terms of efficiency simply because of intels process node advantage. Kabini barely competes with IVB/HW in terms of perf/W (comparing to higher binned chips not bottom of the barrel celerons/pentiums).

My laptop with a 3630qm uses ~60-64 watts running cinebench R15 (540-560 score). The 5350 is around 160-170 points (30% the score) while using more than 30% more power. (Idle is about 23 W for my laptop-This includes the screen on min brightness for both measurments).

Even if you look at the high clocked 4770k you will notice it performs ~4x better in CB 11.5 while using around 4x the power (no DGPU). Its the much higher HW/IVB platform power that kills efficiency, not CPU efficiency.

There is a lot of low hanging fruit with kabini. I expect a significant drop in power consumption.

 

[cbn]

Any particular reason you're talking about desktop parts? If these SKUs hit the desktop they will be nothing more than low performance low cost SOCs just as Kabini or desktop BT are. Kabini was a mobile part that somehow ended up as a super cheap and low performance desktop chip a long time later. Beema and Mullins are the same.....intended to be mobile products. They are not designed with desktop as a prime consideration. For desktop as a prime consideration you should look at Kaveri. Beema + Mullins - These products are designed for mobile first and desktop as a last consideration.

Actually for desktop (more accurately server) I think I would rather use Kabini than Kaveri.

While Kaveri supports AES-NI like Kabini, Kabini (AM1 and I believe BGA) goes one step further with ECC support (though we have not seen any AM1 motherboards declaring ECC support yet)

With that mentioned, the major problem I see with using Kabini on the desktop/server is lack of SATA ports. Sure a person could add in a SATA controller card, but native support is always better.

If only AMD could add a couple more SATA to their cat core SOC (bring total to four). I guess it depends on their volume of mobile vs. projected server, but if the die size hit is not so bad for two more ports then I would gladly welcome the change.

 

[AtenRa]

Why I hate PC mark 8.

Non-linear performance increases. Not only does the A6-5200 have twice the number of cores but it runs at twice the frequency as the E1-2100. It has four times the theoretical performance but its score is not even twice as high. The a6-5200 is 33% faster than the a4-5000 but its pcmark 8 score is only 13% higher.

That is because it is not a CPU ONLY benchmark, it uses the iGPU as well especially if you run with OpenCL.

I have found it to be a very nice all around performance benchmark especially to monitor real world system usage performance.

 

[SilentRunning]

I thought from the picture it's clear that AMD is releasing a magnifying glasses

Look again there is clearly no magnifying going on. Or are you implying that AMD will produce defective magnifying glasses.