Intel Z3770 vs Snapdragon 800 in Kraken 1.1

[dragantoe]

thought I would throw in my tegra 4 which just got 4568 on kraken 1.1

 

[VengenceIsMine]

thought I would throw in my tegra 4 which just got 4568 on kraken 1.1

# would mean a lot more if it had some additional data like platform, clock speed browser etc. Kraken is HIGHLY dependent upon which browser you run it in.

My Core I7 desktop Kraken 1.1
IE 11: 2372
Firefox 23: 1577
Chrome 29: 1463

Same machine. See how much the browser alone matters? That was why given the circumstances I was in, running Kraken on two similar code bases had more relevance than a random score like what you just threw out.

 

[BallaTheFeared]

If these were done with the same OS using the same browser than it looks like the elephant we've heard coming has arrived.


Edit: NM looks like RT vs x86, not sure if it has any merit.

 

[VengenceIsMine]

If these were done with the same OS using the same browser than it looks like the elephant we've heard coming has arrived.

It was as close as a comparison as I could make it under the circumstances.

Windows 8.1 Preview Z3770 1.46 base clock, turbo to 2.3 Ghz

vs

Windows 8.1 Preview Snapdragon 800 that maxed at 2.2 Ghz, I didn't write down base clock but I think the Qualcomm was definitely more likely to be final silicon than the Atom.


build #'s were close, wasn't public preview.

Obviously Kraken favors single threaded perf some which we expect to favor Intel based on Anands earlier articles. Also we can expect Windows 8.1 to be somewhat more optimized for x86 than RT for Arm v7 but Microsoft has been playing the multiple processor support game for a long time since NT 3.5 days if I remember correctly so I suspect they are pretty good at it.

So yes, it looks like the elephant is imminent.

 

[beginner99]

We're dealing with one thread, period. So straight away we have best case for the z3770 (who would have believed that?) and its burst turbo.

And how many threads do browser use in real world scenarios to render web pages? exactly, one. That's why single-threaded performance matters a lot and quad core phones are pretty much non-nonsensical (except for marketing reasons).

That'll be why the S800 has all the design wins and Intel has...?

Design wins are mostly determined before launch. Qualcomm already mocked Intel over their lack of wins.

Just use your brains, if it was THIS good it would be the chip in all the cool new tablets launching next month. Guess what? Yep it's in NONE OF THEM.

We are talking about performance and performance/w not about sensible pricing. I've always believed that intel can only fail with too high pricing.

 

[dragantoe]

# would mean a lot more if it had some additional data like platform, clock speed browser etc. Kraken is HIGHLY dependent upon which browser you run it in.

My Core I7 desktop Kraken 1.1
IE 11: 2372
Firefox 23: 1577
Chrome 29: 1463

Same machine. See how much the browser alone matters? That was why given the circumstances I was in, running Kraken on two similar code bases had more relevance than a random score like what you just threw out.

it was on chrome, and it's a stock tegra 4 you could have just asked "what browser were you using?"

 

[mavere]

Well, ignoring IE11 on ARM for a moment, I got some numbers from an old laptop on dualcore Penryn at 2GHz.

IE Preview 11.0.9431.0
5600.0ms +/- 0.9%

Chrome Canary 31.0.1612.1
2930.5ms +/- 0.4%

Assuming scaling holds up on that Win8.1 tablet, we'd could be looking at ~4000 ms in Chrome (which would be pretty decent for a low wattage device).

The performance gap here is actually a lot smaller than what I'd expect given Anand's comments in the past:

On single threaded performance, you should expect a 2.4GHz Silvermont to perform like a 1.2GHz Penryn. [...] Keep in mind, I’m talking about single threaded performance here. In heavily threaded applications, a quad-core Silvermont should be able to bat even further up the Penryn line

but there's a lot of variables here anyway.

 

[erunion]

it was on chrome, and it's a stock tegra 4 you could have just asked "what browser were you using?"

I see you didn't answer the question of what device it was in. Im assuming its in your Shield, which is an apple to orange comparison.

Your shield, unlike a tablet, has a heatsink and fans! So obviously its Tegra 4 is clocked higher than any tegra 4 in a tablet would be.

 

[AtenRa]

delete

 

[monstercameron]

are there any current products equivalent with the atom in terms of the gpu in intels lineup[or will there be any differences in the gpus?]?

 

[AtenRa]

are there any current products equivalent with the atom in terms of the gpu in intels lineup[or will there be any differences in the gpus?]?

No, but im expecting lower performance than HD2000

 

[SiliconWars]

You repeatedly tried to assert or imply that I was either fibbing or biased without the slightest evidence other than you didn't like the data I was presenting. My approach was a balanced as circumstances allowed. I simply presented a data point and while you strenuously tried to denigrate it, it is a better & more valid data point than any other I am currently aware of comparing these 2 SOCs.

You think so do you? First you claim

I know Kraken's limitations quite well but this result was somewhat consistent with some other leaked results so I'm not completely surprised.

then you say

it is a better & more valid data point than any other I am currently aware of comparing these 2 SOCs.

What leaked results show Atom some 64% faster than S800? I must have missed those.

Most that have been presented to date in terms of comparing these 2 SOC's are on differing platforms, browsers or are using what is obviously early engineering silicon. It is more valid because the code & platform are nearly identical (barring slight differences in optimization) and the silicon involved is either final or closer to it than has been previously reported.

You know nothing about the silicon in the other reports.

Obviously Kraken does favor single threaded performance but guess what, single threaded performance & single threaded javascript performance is still critical to user experience, ESPECIALLY in tablet usage. So in short, you thrashed and squealed and contributed very little other than mentioning that Kraken favors single threaded performance.

Yes and given time to take benchmarks, this is what you chose. You profess to know Kraken like a champion yet I was the one who had to prove it favoured Intel and is about as single-threaded as a benchmark can be while you were telling everybody how much it blows the socks off S800. Don't you think that was relevant information for your first post?

Anyone who thinks it is 64% faster than the S800 needs a head examination, just like they did after the AnTuTu results.

SiliconWars, get off it. At this point you're being a jackass
-ViRGE

 

[Arachnotronic]

Anyone who thinks it is 64% faster than the S800 needs a head examination, just like they did after the AnTuTu results.

Hmm...we'll really have to see about that.

 

[Nothingness]

Hmm...we'll really have to see about that.

According to leaked presentation Z3770 is 53% base/60% rate faster than S800 on SPEC CINT2000. Compared to A15 it's 28% base/33% rate faster.

Of course this is probably a good benchmark for Intel (icc is very good at compiling SPEC). But that is another interesting data point.

 

[Enigmoid]

Based on Windows single threaded performance in a benchmark that already favoured Intel, it's pretty clear I won't be surprised by performance in what is actually relevant.

Singlethread performance is the single most important metric of performance for a Soc like this.

 

[SiliconWars]

Singlethread performance is the single most important metric of performance for a Soc like this.

If that were true then surely Clover Trail would be in everything? Certainly according to Anand's benchmarks which has the Z2760 blowing away all the competition.

(Javascript - All of these were run in Modern IE10 on the Windows devices.)

Does that look anything like reality, or do we constantly read about how Clover Trail is dog slow? It sure isn't 50% faster than Tegra 3 unlike what is shown here.

Come on, we're looking at best case for Intel again and you're all falling for it again. If this was some first time poster telling us that Steamroller was 50% faster than Haswell you'd all be asking serious questions, and you should be with this as well.

 

[JoeRambo]

The problem with this comparison is that in fact a very critical component of browser - Javascript engine is being compared between ARM and x86 instead. I think it is safe to say that browser makers have improved JS performance by order of magnitude on same hardware in last decade, and there still exists a sizable gap between major browsers - for example atm IE (10) is like 3-4 times slower than chrome on same machine in Kraken. Also - in recent times IE (~ 9 i think) 64bit used to have abysmall JS performance compared to 32bit due to not having same JS engine.

So it is complete waste of time to compare arm vs x86 scores even if they are on "same" browser - IE.

Still, it should not stop us from seeing the obvious:
1) Scary amount of performance is available in tablet at that thermal envelope.
2) There is an elephant in this room - I highly doubt that IE11 was built with ICC or by compiler that is even aware about new Atom architecture, yet it is turning out impressive numbers. A quick IE11 preview run of same benchmark on 4.5 Haswell gives ~1500, so that tablet is just 5 times slower than what x86 ST performance champion achieves.

I am looking forward to this CPU, an Atom that is finally good at chewing generic x86 code without recompilation and at that performance is great news.

 

[sxu]

A website is reporting that Intel by mistake released some informations about Bay trail (http://www.tomshardware.fr/articles/intel-atom-z3700-baytrail,1-45599.html).

 

[SiliconWars]

Graphics performance seems very close to the S600 based on that so the S800's graphics should be around 70% faster.

 

[beginner99]

If that were true then surely Clover Trail would be in everything? Certainly according to Anand's benchmarks which has the Z2760 blowing away all the competition.

Does that look anything like reality, or do we constantly read about how Clover Trail is dog slow? It sure isn't 50% faster than Tegra 3 unlike what is shown here.

Well you can always go on youtube and look at real-world usage comparisons.

http://www.youtube.com/watch?v=SILUVIwiNF0

Here you see the RAZR I (=Medfield, Single-core) against the Galaxy S4 international (Exynos Quad-core). You can see that the 4 cores don't lead to any advantage at all for the GS3.

At 11:30 you can see what single core performance can make a difference.

In general the RAZR I appears just better overall.

 

[liahos1]

Graphics performance seems very close to the S600 based on that so the S800's graphics should be around 70% faster.

Using ice storm to determine GPU strength apples to apples

S800 w/ Adreno 330 about 1.71 faster than HTC One with Adreno 320

Adreno 320 about 1.76x faster than Adreno 225 in HTC One X

Intel saying they are 2.4x faster than Adreno 225 sooo

Intel GPU appears to be 1.36x faster than Adreno 320 and about 20% slower than Adreno 330 in this benchmark.

CPU performance appears to be 53%-60% faster than the s800 in CPU performance. All of this is going to be I’d bet you at a much better power envelope than the s800.

 

[Sweepr]

Very impressive CPU performance for something that should fit iPad-like tablets (50-60% faster than a S800 @ SCPEint!). Better than S600 GPU performance would be a huge jump from Clovertrail days and suitable for a modern tablet, even if its not an Adreno 330 killer.
This is just a 7-gen GPU with 4 EUs, if rumours are true they will have two new Atoms next year, each based on a new graphics architecture (Gen 8 and Gen 9) + 14nm technology to fit a lot more EUs.

 

[SiliconWars]

Using ice storm to determine GPU strength apples to apples

This is a better way of doing it -

Take the Z2760 score of 2439 and multiply it by 4.82, giving 11756, which is basically the same as the S600.

CPU performance appears to be 53%-60% faster than the s800 in CPU performance. All of this is going to be I’d bet you at a much better power envelope than the s800.

No specint performance is up to 60% faster, that's a long way away from "cpu performance".

 

[blackened23]

So the take home message here is that kraken is worthless as a benchmark. Sorry, but the pre-requisite for any benchmark being taken is consistency - kraken gives dramatically different results just from using a different browser. This opens up an entire can of worms when comparing results of different platforms with different operating systems. There is absolutely no consistency here, even when testing on the same system.

I don't think there are any good benchmarks for the ARM SOC platform, and krakn seems to be no exception.

 

[liahos1]

fair enough!

do you think we'll see s800's in sub 200 dollar tablets? Not a leading question. It seems to me that the s800 is purely a premium end device and the s600 and s400 should be the low end to mid end comp? So you have bay trail which should be according to intel playing in price points from sub 200 to premium end. I just wonder how much gpu performance matters here vs cpu performance and performance per watt?