EarlyMon

OMAP 3630 destroys Snapdragon 8650?

I think it's only fair to clarify that the OMAP isn't just a cpu+gpu.

OMAP 3640=ARM Cortex A8 plus (PowerVR SGX530 GPU plus C64x+ DSP core (likely at 430 MHz, not floating point, requires software for that)

The way that Android Central portrayed things included saying that the Droid X is faster because it has a separate GPU, implying the Snapdragon seems to lack a GPU - not so. And a lot of forum participants across the web are getting plus marks on their posts explaining that the Snapdragon also lacks a DSP module - also not so.

Snapdragon 8650=Scorpion plus (AMD Z430 GPU plus 600 MHz DSP core (floating point, if I'm not mistaken[*]))

The following goes slightly into detail:

Inside DSP Articles - Snapdragon Scorpion

Inside DSP Articles - OMAP

Various forums claim that the Snapdragon is an ARM Cortex, A8 or maybe even A9, but it's only safe to say that it's Qualcomm proprietary with many similarities to A8 or (maybe even) A9 - EVO, many others besides Snapdragon - they're all ARMv7 (read: Cortex) variants. OMAP is TI's answer to ARMv7. Snapdragon is Qualcomm's answer to ARMv7.

Both of these processors - OMAP 3630 and Snapdragon 8650 - are remarkably advanced. They feature integration details that both TI and Qualcomm can be mighty proud of. Both offer state-of-the-art performance in mobile devices. The Nexus One's Snapdragon 8250 is HSPA only, whereas the 8650 is dual mode HSPA/EV-DO, so, a carrier difference only.

Benchmarking is a nearly lost art for the public - it must always be a suite of tests and while the temptation is there to aggregate the final result to a weighted number, doing so means nothing. On any number of the individual measurements for these integrated processors, I've no doubt that each will outneck the other. Neither one seems in an way whatsoever to be a slouch.

But - all we care about (righteously) are the end results in products.

As bedwetting over champing the competition for mindshare reaches tidal proportions, reviewers race to show how the Droid X outperforms sliced bread, including every ancient device, like the EVO.

In graphics. Because HTC put a governor on them. For whatever reason.

But not necessarily because of a raw processor deficiency - for us outsiders, it was done for reasons unknown - the very reason for this thread.

As it is, the Linpack tests showed the Droid X significantly faster than an EVO in MFLOPS. However, processor benchmarks seem to put the Snapdragon over the OMAP in Dhrystone MIPS.

While the video was fun for me too, I pretty much consider it a parlor game - just something we do while we all wait for Froyo and whatever updates we'll all incorporate in six months.

ONLY as fast at that Nexus One?

I can't argue with your logic in any way whatsoever. Your conclusion that the Droid X is going to smoke Snapdragons with 2.2, projecting forward, makes sense.

But - I thought if we're going to make projections, maybe it's wise to compare Nexus One from 2.1 to 2.2 - the results are intesting:

The Droid X at 2.1 is showing 8 MFLOPS. At 2.1, the Nexus One started at around 7 MFLOPS in the LinPack test. Why then is the EVO only coming in at 5? Sense? Other overhead? [**]

If you look only at the video (timecode 0:35), you'd think that the OMAP is 37.5% faster than a Snapdragon in MFLOPS. It's not. It's 12.5% faster when comparing to non-impeded device.

By the way, I'm glad the OMAP has a higher MFLOPS (million floating point operations per second) score. As its DSP relies on software to get to floating point, it's gonna need it. And it's got it. So it's OK.

Now let's examine those Quadrant results. If we believe Android Central, the Droid X is running 2.1. And it's getting 2.2-like results already.

Out of the box.

A PowerVR SGX530 GPU plus a C64x+ 430 MHz DSP core is outrunning an AMD Z430 GPU plus a 600 MHz DSP core? By a factor of about 2.5 to 1?

And the Droid X was originally announced to be released with 2.2?

And Motorola did nothing under the hood 2.2-related for this 2.1 release?

And we know that for a fact before the thing is even released?

Either we're very smart, and the OMAP is such a Snapdragon killer that Qualcomm better start apologizing for their existence - or there's some shenanigans in there somewhere.

I'm gonna claim shenanigans. I could be wrong, however. I often am.

BTW, here are supporting links, I think it's only right to always include references:

Snapdragon (processor) - Wikipedia, the free encyclopedia

Qualcomm unveils 1 GHz mobile phone processors - News - Linux for Devices

C64x+ - Texas Instruments Embedded Processors Wiki

Verizon spoils Apple's party, unveils the Droid X with Android 2.2 and Flash Player 10.1 ? Cell Phones & Mobile Device Technology News & Updates | Geek.com

http://gadgetsdna.com/wp-content/uploads/2010/06/iphone4_droidx_incredible_evo4g_nexusone.jpg

Android 2.2 Froyo Benchmarking - iandouglas.com

Quadrant Benchmark Screenshots - Page 2 - Nexus One Forum - Google Phone Forum

Dhrystone - Wikipedia, the free encyclopedia

Qualcomm Products and Services - The Snapdragon Platform

[*]I don't have a link supporting that the Snapdragon DSP is floating-point capable, but I have seen that suggested - so take that with a grain of salt. Either way, it's referred to as a 6th generation DSP, so it's not like they just threw some junk in there.

Qualcomm News and Events - Press Releases - Qualcomm Acquires Handheld Graphics and Multimedia Assets from AMD

By the way, I'm not claiming to be an expert or anything, but my DSP code was used by Cray to debug their compilers prior to running their benchmarks, and I was responsible for the benchmarking validation at the world's largest supercomputing site back in the day, so I've done just a little bit with benchmarking and thought my 2 cents might be a little bit helpful. Those were only like in the 130 to 800 MFLOP range, and not any graphics beyond the stuff we wrote for Tektronix outputs. These new processors are kinda way out there for me compared to that and so while I don't have any background in benchmarking this new stuff, maybe I have a little background in calling shenanigans on speed claims.

PS - A little note on triangles/second, a popular raw GPU benchmark.

(let's call a million triangles per second, Mtps)

The Snapdragon 8650 is rated at 22 Mtps.

The Palm Pre with a PowerVR SGX 530 is rated at 14 Mtps.

The iPhone 3GS with a PowerVR SGX 535 is rated at 28 Mtps.

The Samsung Galaxy S processor is said to do a whopping 80 to 90 Mtps (like a Snapdragon 8672 in that regard).

An SGX 530 at any speed doing so much better in graphics with that sort of benchmark?

Again, references:

Tegra or Snapdragon - xda-developers

Snapdragon

[**] Huh. Whatdyaknow? On a lark, I ran Linpack on my stock/rooted rom that's not running too much bloatware - still, running Sense. I get 6.9 MFLOPS. Added in some bloatware, re-ran: I got just over 5. This just in: benchmarks should only reported with specified or controlled configurations. Alert the media. We could make a ground-breaking video and plaster it all over YouTube right there: Dramatic 25% speed increase on EVO PROVEN by benchmark! Watch the video!

Post your linpack score.

Are we done yet?

Last edited by EarlyMon; June 28th, 2010 at 11:51 PM. Reason: changed 3630 to 3640 - it's still 45nm OMAP3