Samsung Exynos 5250 - production starting in Q2 2012

[Laurent06]

Meaning it still exists but will go for tablets only or what? <- confused as hell....

Meaning Samsung made a dev/demo tablet with it

 

[Ailuros]

Meaning Samsung made a dev/demo tablet with it

That doesn't help one bit; I was hoping for an A15 based smartphone platform soon, in order to actually have at least some comparison material to custom core CPUs based on ARM ISAs

 

[Ailuros]

By the way does anyone have a clue why NVIDIA is listing only 32 GFLOPs for the Mali T604 (under 5250) in its recent marketing slides? They're acknowledging 4 TMUs@533MHz, so in order to reach those 32 GFLOPs it should have less than 32 ALU lanes. What am I missing?

 

[Nebuchadnezzar]

I caught that too, it should be 72 Gflops right?

 

[Ailuros]

I caught that too, it should be 72 Gflops right?

Theoretical peak yes; but since IHVs lately also seem to count for instance SFU single FLOPs, it's probably where that 72 peak probably comes from. Assuming 64 ALU lanes at 533MHz 68 GFLOPs.

 

[Rys]

I think it's fair to say that it's 72 GFLOPs for T604 at 533 MHz, peak. Not that there are FLOPs from SFU units in their pipe, but any SFU FLOPs should be 'valid' if you can trivially retire instructions from the units in general shading.

 

[Ailuros]

I think it's fair to say that it's 72 GFLOPs for T604 at 533 MHz, peak. Not that there are FLOPs from SFU units in their pipe, but any SFU FLOPs should be 'valid' if you can trivially retire instructions from the units in general shading.

The point is where NVIDIA got the 32 GFLOPs only from for the T604 (5250) in the infamous T4 slide. Under that reasoning the T604 should have 30 ALU lanes only and a crapload of SFU units (79?) to reach those 72 GFLOPs all together at 533MHz

 

[Rys]

I think it's fair to say that NV technical marketing is struggling to get the finer details of the competition right at this point

 

[Ailuros]

I think it's fair to say that NV technical marketing is struggling to get the finer details of the competition right at this point

Are you sure you want to call it "technical" marketing? A mistake like that doesn't suggest even any basic technical understanding.

 

[Alexko]

I think it's fair to say that NV technical marketing is struggling to get the finer details of the competition right at this point

I think it's fair to say that NV technical marketing isn't struggling to get the finer details of the competition right, because a struggle implies an attempt.

 

[Jubei]

Meaning it still exists but will go for tablets only or what? <- confused as hell....

Someone must have forgotten to give this man the memo that Octa is not coming to smartphones

http://youtube.com/watch?v=-rz8aPVxv24

There is some hope left

 

[Turbotab]

Someone must have forgotten to give this man the memo that Octa is not coming to smartphones

http://youtube.com/watch?v=-rz8aPVxv24

There is some hope left

The Galaxy Note 3 will still technically be a smartphone, despite the phablet tag and 3000 mAh ++ battery

 

[Nebuchadnezzar]

Someone must have forgotten to give this man the memo that Octa is not coming to smartphones

http://youtube.com/watch?v=-rz8aPVxv24

There is some hope left

He's also stating the 70% power advantage over the 4412, while at CES the slides were clearly stating that that 70% decrease was relative to the 4210. He's exactly what his title says he is, director of marketing and nothing else.

 

[juicytuna]

There's still hope for Exynos Octa in SIV:


http://www.sammobile.com/2013/03/03...s-leak-confirm-exynos-octa-powervr-sgx-544mp/

 

[Laurent06]

These AnTuTu results seem too low, or the chip is not running at the claimed 1.8GHz.

 

[tangey]

As per the comments on that page, the 5th screenshot says "my device nexus 4" ??

 

[french toast]

These AnTuTu results seem too low, or the chip is not running at the claimed 1.8GHz.

Agreed..wayy to low.

 

[Turbotab]

These AnTuTu results seem too low, or the chip is not running at the claimed 1.8GHz.

Are you basing that off the very high scores of the Tegra 4 reference tablet? If we compare the CPU scores of the purported GS4 Exynos to the Nexus 10, they look very reasonable, considering that Antutu is highly threaded.

However the Tegra 4 CPU scores are very suspicious. N10 Integer 3,127 vs T4 Integer 8002 & N10 Floating Point 2060 vs T4 12,275 that is far beyond the result you would expect for doubling the core count & a 12% frequency increase, on the same uArch. I wonder if Nvidia has been fully honest in that benchmark, or are they using a new revision of the A15 core??

There is a history of cheating on the Antutu benchmark, but surely Nvidia would not pull such a dirty trick.
http://www.antutu.com/view.shtml?id=3909

Nexus 10 - Officially tested by the Antutu team (AnTuTu Benchmark 3.0.3)
http://www.antutu.com/i/view?id=4255
Dual Core A15 @ 1.7 GHz

Total: 13,750
RAM: 2,340
CPU integer: 3,127
CPU float-point: 2,060
2D graphics: 1,470
3D graphics: 3,842
Database IO: 565
SD card write: 150
SD card read: 196

Purported GS4 - Exynos Octa - (AnTuTu Benchmark 3.0.3)
Quad Core A15 @1.8 GHz (yes I know + 4 A7 cores)
http://www.sammobile.com/2013/03/03...rvr-sgx-544mp/screenshot_2013-03-03-13-09-15/

Total: 24,894
RAM: 5,144
CPU integer: 5,228
CPU float-point: 5,536
2D graphics: 1,621
3D graphics: 6,449
Database IO: 565
SD card write: 150
SD card read: 201

Tegra 4 scores from MWC 2013
http://www.miaplicacionandroid.com/wp-content/uploads/2013/02/0aeftegra4-antutu-680x372.jpg

Quad A15 @ 1.9 GHz

Total: 36,375
RAM: 7,666
CPU integer: 8,002
CPU float-point: 12,275
2D graphics: 1,636
3D graphics: 5,925
Database IO: 565
SD card write: 150
SD card read: 156

 

[juicytuna]

The standard Android kernel lacks a lot of the latest Linaro CPU optimizations. I wouldn't really doubt those T4 CPU scores.

 

[Nebuchadnezzar]

The standard Android kernel lacks a lot of the latest Linaro CPU optimizations. I wouldn't really doubt those T4 CPU scores.

They barely make for such a difference, I tested this myself. At least on the kernel side they are insignificant. The Android libraries might be something else.