[AnandTech] Atom Z3770's Cinebench performance

[mikk]

I'm wondering is it valid to compare a Cinebench 32-bit result with a 64-bit result?

It is a bit flawed because 64bit version of Cinebench scores 10% higher.

 

[Abwx]

It is a bit flawed because 64bit version of Cinebench scores 10% higher.

With Piednoel , who was claiming that chinese early leaks
of HW were irrelevant , you can be assured that if it s flawed
it is forcibly , well , flawed in the "good" direction...

 

[Erenhardt]

Does the testing methodology differ from what we have seen in the past?
Testing mobile intel chip (CPU+iris pro 5200):

That doesn't look like laptop...

And I found this photo. It may, or may not have something to do with the atom testing...

That is heck of a tablet!

 

[MisterMac]

Well damn.

Never believe a Francois - but still damn.



Why is the mastermind of the p4 heating ovens working on lowpower?
he should be working on E or the highend mainstream teams for for us instead.

300w -E crazy SKUs please.



Anyhow - this looks like my tablet processor of choice.
I'd love to see some android OS based scores tho.

 

[inf64]

Results from more workloads(with measured power draw and clock speeds) is more than welcome.

 

[rootheday3]

re 'testing methodology' - Intel uses desktop chassis's like this for a lot of their internal mobile development boards rather than form factor devices (e.g. a notebook) for both cost and convenience (setting jumpers, swapping hard drives, attaching debugging tools, reading BIOS post codes, etc). Will the the thermals will be different in such a chassis versus a real laptop? sure.

But Intel also makes their own reference form factor devices (see Anandtech articles re the Medfield phones - industrial design and thermals were good enough that it was basically used verbatim by several carriers and slightly tweaked by several others).

And Intel is very careful when it makes public performance claims because of the legal implications. I would bet they do testing with thermal heads, etc to ensure that the benchmarks they publish will be representative.

If you are curious about Iris Pro performance in a real laptop, there are plenty of independent tests available now. See pclab.pl (1 & 2), notebookcheck.net, hardwareluxx, computerbase.de

regarding the image at the bottom - I have no idea what that is. Where did you get it? what does it have to do with Intel benchmarking practices and/or tablets? or are you just trolling?

 

[blackened23]

Iris Pro beats GT730M soundly while using less power and costing less (versus a mobile quad + mobile dGPU). Not bad. Very impressive indeed.

Looks like the pressure is on for nvidia and AMD, no wonder Apple decided to ditch nvidia for the macbook pros (in favor of Iris Pro). The main drawback is price - Iris Pro SKUs tend to be north of 500$ - but the 4950HQ quad mobile with Iris Pro is *still* cheaper than a quad mobile with a mobile dGPU while dramatically improving upon overall battery life as compared to quad mobile + discrete.

Personally I think this is awesome, even though it unfortunately comes at the expense of dGPU vendors - Intel is really pushing the boundaries here, I expect even better things with Broadwell quad mobiles. This has very good implications for small, lightweight devices with great graphics performance. Should I say it? I see a future involving ultrabooks with 10+ hours of battery life with outstanding iGPU performance. Not quite there (in terms of battery life) with Haswell, although Broadwell will come close. Can't wait.

 

[Nothingness]

The original Atom achieved 7+ hours fairly easily and that was on a dated lithography process.

The performance increase will take its toll, so BT might not last longer than CT. We'll have to wait for independent reviews to know.

 

[Broheim]

Even more impressive when you take into account that there's a noticeable difference in performance between running the 32 and 64-bit versions, in my case 9%:

well damn, if all of this turns out to be right on target then ARM has a problem, especially considering intel rolling out tic toc for atom.

 

[sushiwarrior]

Iris Pro beats GT730M soundly while using less power and costing less (versus a mobile quad + mobile dGPU). Not bad. Very impressive indeed.

Looks like the pressure is on for nvidia and AMD, no wonder Apple decided to ditch nvidia for the macbook pros (in favor of Iris Pro). The main drawback is price - Iris Pro SKUs tend to be north of 500$ - but the 4950HQ quad mobile with Iris Pro is *still* cheaper than a quad mobile with a mobile dGPU while dramatically improving upon overall battery life as compared to quad mobile + discrete.

Personally I think this is awesome, even though it unfortunately comes at the expense of dGPU vendors - Intel is really pushing the boundaries here, I expect even better things with Broadwell quad mobiles. This has very good implications for small, lightweight devices with great graphics performance. Should I say it? I see a future involving ultrabooks with 10+ hours of battery life with outstanding iGPU performance. Not quite there (in terms of battery life) with Haswell, although Broadwell will come close. Can't wait.

Okay two things, one, this is about Atom, not Iris Pro, two, you're comparing an old i5 to a brand new i7 at CPU limited speeds D: please, consider what exactly you are posting :|

 

[Aikouka]

Who the heck games on a portable ultrabook or tablet on a battery charge?

I don't think I've ever plugged my iPad in strictly because I'm going to play a game on it.

WHO THE HECK games on ultra portables?

I've played games on my Zenbook. It plays Torchlight II fine at 720p.

 

[blackened23]

Okay two things, one, this is about Atom, not Iris Pro, two, you're comparing an old i5 to a brand new i7 at CPU limited speeds D: please, consider what exactly you are posting :|

I've considered it. AMD has to worry on the low end and high end since Bay trail will perform great on the low end, and Iris Pro is apparently beating every AMD APU in terms of performance. It is also beating some nvidia discrete GPUs - that is an amazing feat, so this affects nvidia as well. Come next year, mobile dGPUs may be a dying breed for all except the beefiest full size gaming laptops. I don't think ultrabooks with dGPU will be a thing come this time next year.

That is aside from the fact, again, that Bay Trail should give all low priced x86 devices reasons to worry. If the performance difference is minimal (as compared to say, an AMD mobile APU), that just means that manufactures will use Bay Trail for low end Windows devices.

 

[Enigmoid]

Iris Pro beats GT730M soundly while using less power and costing less (versus a mobile quad + mobile dGPU). Not bad. Very impressive indeed.

Looks like the pressure is on for nvidia and AMD, no wonder Apple decided to ditch nvidia for the macbook pros (in favor of Iris Pro). The main drawback is price - Iris Pro SKUs tend to be north of 500$ - but the 4950HQ quad mobile with Iris Pro is *still* cheaper than a quad mobile with a mobile dGPU while dramatically improving upon overall battery life as compared to quad mobile + discrete.

Personally I think this is awesome, even though it unfortunately comes at the expense of dGPU vendors - Intel is really pushing the boundaries here, I expect even better things with Broadwell quad mobiles. This has very good implications for small, lightweight devices with great graphics performance. Should I say it? I see a future involving ultrabooks with 10+ hours of battery life with outstanding iGPU performance. Not quite there (in terms of battery life) with Haswell, although Broadwell will come close. Can't wait.

Gotta love the cherry picking.

Clearly the 730m result is borked somehow. Clearly the 650m is faster than iris in the 4 games tested (not really enough to make a conclusion). It looks like the 730m is 64/128 bit DDR3 which is why it loses so badly in Grid 2 (or its CPU limited as the low and high settings return the same results). 730m and 650m run at similar clock speeds so the main difference there is the VRAM type.

Iris is NOT cheaper than quad + dgpu. A GK107 chip (gtx 650) costs around $100 (I can get it for $80 on sale). That includes reseller markup (probably 20-30%) and the huge PCB + fans on the chip. The actual costs of the GK 107/208 chip + 1 GB GDDR5 + manufacturing to the OEM is probably something like $50. Don't forget according to the AT article intel charges ~$50 for the edram ALONE.

Also no dgpu really affects battery life under light/ nongaming loads. Optimus works.

 

[blackened23]

Amazing! "Cherrypicking. Biased results. Borked." I'm going to use these phrases when the results show something I don't want to see. Yeah.:thumbsup: Of course, more data is always desirable. I'm sure the usual roundup of websites will do their own benchmarks once Iris Pro is officially released (no devices on the market yet). So we will see in time.

As far as I can tell, the Iris Pro beats the 730M soundly with all metrics considered, because the 730M is roughly 5-6% faster than the 630M, and that's a generous estimate. Pretty much understandable why Apple ditched nvidia, and yes, dGPU does cause more power consumption than Iris Pro - the 750M alone is ~40W TDP. Meanwhile you get Iris Pro + quad mobile for a cheaper cost, total TDP = 47W. Yeah, the 750M is faster than the Iris Pro, but anything below that is undesirable. I'd guesstimate that a 740M is roughly equal to Iris Pro while having a higher TDP than a 47W Quad mobile/Iris Pro.

Also, the quad mobile 4900MQ is roughly 60$ less, overall, than the Iris Pro 4950QM. So yes, Iris Pro is cheaper than adding a dGPU. This situation will only get worse (for both AMD and nvidia, unfortunately) with Broadwell, intel is bringing the pain in mobile. Intel does have work to do on their iGPU drivers, although strides have been made in the past 2 months.

 

[Abwx]

TDP than a 47W Quad mobile/Iris Pro.

83W comsumption in benches , far from 47 , even
when accounting for the MB + RAM + SSD.

Besides , it s rather the i7 that beat the i5 in
theses benches than iris being better than a dGFX.

 

[blackened23]

TDP doesn't equal maximum power consumption. I don't know where you get that idea, that is not the case.

 

[Abwx]

TDP should be the effective comsumption
once you have integrated the inevitable
power peaks that are in principle of very
short times.

TDPs that can be exceeded for the needs
of benchmarketing are flawed numbers.

 

[Enigmoid]

Amazing! "Cherrypicking. Biased results. Borked." I'm going to use these phrases when the results show something I don't want to see. Yeah.:thumbsup: Of course, more data is always desirable. I'm sure the usual roundup of websites will do their own benchmarks once Iris Pro is officially released (no devices on the market yet). So we will see in time.

As far as I can tell, the Iris Pro beats the 730M soundly with all metrics considered, because the 730M is roughly 5-6% faster than the 630M, and that's a generous estimate. Pretty much understandable why Apple ditched nvidia, and yes, dGPU does cause more power consumption than Iris Pro - the 750M alone is ~40W TDP. Meanwhile you get Iris Pro + quad mobile for a cheaper cost, total TDP = 47W. Yeah, the 750M is faster than the Iris Pro, but anything below that is undesirable. I'd guesstimate that a 740M is roughly equal to Iris Pro while having a higher TDP than a 47W Quad mobile/Iris Pro.

Lol What?!?! The 630m on average (25 notebooks on slightly different configurations) gets an average of 1042 in 3dmark 11 GPU, highest 1218 @ 937/900, lowest 851 (clocks unknown but throttling suspected). Nominal Clocks for the 630m are 672 core, 900 mhz DDR3 Vram. The 730m gets an average of 1858 points from three samples, min 1739, max 2039. Using 3dmark we can estimate an increase of 78% at low resolutions because the vram is going to hold the 730m up when resolution is increased, especially when using the GK 107 chip with a 64 bit bus. In fact here we can see that in skyrim, the difference between the HIGHEST clocked 630m and a 730m is on average 43% (24% at low, probably CPU limited on ULV climbing to 63% on high with AA)

http://www.notebookcheck.net/Review-HP-Pavilion-Sleekbook-15-b004sg-Ultrabook.93576.0.html

In fact here (http://www.anandtech.com/show/5672/acer-aspire-timelineu-m3-life-on-the-kepler-verge/4) you can see the absolutely massive difference between the 640m and the 6650m. The 730m is slightly better than the 640m (+10%) with 128 bit memory and slightly worse with 64 bit memory (-20%). The 6650m is about 10-20% worse than the 630m.

Even if you want to compare raw specs, 630m is 96 cores @ 672/900 (hotclocked to 1344/900). 730m is 384 @ 725/900 (assuming 128 bit).

What is the TDP of the igp in Iris? For power consumption while gaming take that 40 watt tdp of the 750m and subtract about 80% of iris's tdp. Total platform power under stress will be down with Iris though. You also realize that a 740m with a 128 bit bus and GDDR5 is faster than a 650m, even faster than a 660m? Cutting the bus down to 64 bits and using DDR3 we can see than a handicapped 750m is very comparable to the highest iris SKU models.

EX) using http://www.anandtech.com/show/7238/acer-v7482pg9884-review-everything-you-need/4 and http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6 and selecting for highest playable settings we see that

(iris pro 4950 47w / neutered 750m)

metro ll 2033 768p med - (36.5/37.5).
bioshock infinite 900p very high - (24.2/30.1)
Sleeping dogs 900p high - (19.3/42.1)
Tomb Raider 900p high - (31.4/31.6)
Grid 2 900p high - (38.3/53.2).
In two games iris is competitive with a gimped 750m (1058 core but only ~14 GB/sec bandwidth). In three Iris gets murdered. Even a gimped 740m will be competitive with iris let alone one that has 128 bit DDR3 or 128 bit GDDR5.

Also, the quad mobile 4900MQ is roughly 60$ less, overall, than the Iris Pro 4950QM. So yes, Iris Pro is cheaper than adding a dGPU. This situation will only get worse (for both AMD and nvidia, unfortunately) with Broadwell, intel is bringing the pain in mobile. Intel does have work to do on their iGPU drivers, although strides have been made in the past 2 months.

Ark intel:
i7-4950HQ- Tray price $657
i7-4900MQ- Tray price $568
Iris costs $89 more but the comparison is flawed. Why? Because iris has lower clocks.
i7-4950HQ- 2.4 ghz base, 3.6 ghz turbo, 6 MB cache
i7-4900MQ- 2.8 ghz base, 3.8 ghz turbo, 8 MB cache
Drop down to the i7-4800MQ
i7-4800MQ- 2.7 ghz base, 3.7 ghz turbo, 6 mb cache $378.
Now there is a massive price difference. Does anyone know if that price includes the edram?

And yes a $60 dgpu easily outperforms iris. That cost to the OEM can buy a 750m with 2 GB GDDR5.

 

[aigomorla]

And I found this photo. It may, or may not have something to do with the atom testing...

That is heck of a tablet!

Da hell... thats one heavy tablet...

dual heat sinks... and another blower...

you sure thats not a All in One display unit where the PC is built onto the monitor directly?

 

[Enigmoid]

To get back on topic I expect that the atom 3770 to probably use around 5-10 watts of power. Don't forget that intel's tdp is slighty less than max CPU + max GPU (I have a 3630qm and furmark + cinebench uses around 45 watts without boost). Even intel's desktop CPUs are very close to or over tdp when the cores + IGP is stressed.

 

[Erenhardt]

Da hell... thats one heavy tablet...

dual heat sinks... and another blower...

you sure thats not a All in One display unit where the PC is built onto the monitor directly?

It was just a random SLI laptop photo to better express what I meant, some call it trolling, some think it is good joke.

 

[Sweepr]

To get back on topic I expect that the atom 3770 to probably use around 5-10 watts of power. Don't forget that intel's tdp is slighty less than max CPU + max GPU (I have a 3630qm and furmark + cinebench uses around 45 watts without boost). Even intel's desktop CPUs are very close to or over tdp when the cores + IGP is stressed.

There is this early 2013 slide:

Comparing to the recently leaked stuff: Valleyview D was up to 12W, now its up to 10W TDP; Valleyview-M was 4-6.5W now its 4.5-7.5W, Valleyview-T (Bay Trail-T) was up to 3W and there are no new leaks about its TDP. Anything from 3-5W TDP is my guess.

 

[monstercameron]

a bit off topic but doesnt grid 2 have aggressive optimization for intel hf graphics?
it goes to show optimizations can make or break anything...

http://software.intel.com/en-us/blo...-intel-iris-graphics-brought-me-into-the-fold

 

[blackened23]

Must be those dirty compiler tricks.

 

[Enigmoid]

a bit off topic but doesnt grid 2 have aggressive optimization for intel hf graphics?
it goes to show optimizations can make or break anything...

http://software.intel.com/en-us/blo...-intel-iris-graphics-brought-me-into-the-fold

Grid 2 tends to run better on AMD.