Trinity vs Ivy Bridge

[Mintmaster]

IVB GPU size is 45 mm² (~28% of the overall die) according to official wafer and die pictures. If you add a 28% big GPU to the Sandy Bridge non iGPU space you end up way smaller. Llanos iGPU is much bigger, it reaches around 40% of the overall die of 228 mm².

Look at that link. Official IVB die pictures aren't the retail ones. Your method of is incorrect, as IVB has additions to it, so the non-iGPU space is not equivalent to that of SB. Your math is incorrect, as you wouldn't add X% to it, but rather add a fraction equal to X/(100-X).

One again, just look at the link. Even if Intel does transition back to the the smaller die, it's 71 vs. 91, but Llano's perf advantage is much bigger than that and it's going to get even wider with Trinity.

 

[Paran]

Look at that link. Official IVB die pictures aren't the retail ones.

Do you have an official confirmation from Intel? I made a new calculation based on this and this. I have to correct my previous number slightly, I got 28,8% and 29,2% iGPU space from the overall die. I think most people don't exclude the free space on the bottom, otherwise I would get something around 32%, but this would be an inaccurate calculation. With 29% of 160 mm² (which is the official IVB size) iGPU size would be 46,4 mm². Llano is in a different class, die allocation much more towards iGPU.

 

[Mintmaster]

Fine, use that second photo of yours (very nice, BTW), line the CPU cores with an SB die shot, and measure. You''ll find that IVB would be 270mm2 on 32nm, making the IGP 79mm2. That's smaller but not in a different class from Llano's 91mm2 IGP.

Not to state the obvious, but Intel's process advantage (temporarily) makes up for architectural deficiencies in the mobile sector, as the 45W IVB has the same clocks (actually, slightly higher IGP turbo) as the desktop version. That makes it roughly on par with the gimped A8-3500M. Let's see what Trinity is capable of in the various power brackets...

 

[nAo]

By excluding the free space and IO (?) stuff at the bottom and the piece of L3 that at the top I get 26.2%, not 28%.

 

[Mintmaster]

By excluding the free space and IO (?) stuff at the bottom and the piece of L3 that at the top I get 26.2%, not 28%.

What would be your equivalent analysis of Llano?

I just assumed this guy knows what he's talking about:
http://www.chip-architect.com/news/2012_04_19_Ivy_Bridges_GPU_2-25_times_Sandys.html
With this diagram, I get similar numbers to what Paran was saying.
Intel's diagram unfortunately involves some photoshop (possibly of SB).

 

[Paran]

By excluding the free space and IO (?) stuff at the bottom and the piece of L3 that at the top I get 26.2%, not 28%.

I only excluded the free space at the bottom, that's why my percentage is a bit higher. It would be 42 mm² with your 26.2%.

What would be your equivalent analysis of Llano?

I just assumed this guy knows what he's talking about:
http://www.chip-architect.com/news/2012_04_19_Ivy_Bridges_GPU_2-25_times_Sandys.html
With this diagram, I get similar numbers to what Paran was saying.
Intel's diagram unfortunately involves some photoshop (possibly of SB).

1) it's very unlikely that Intel made such a big change just before mass production from E0 to E1 because validation times for CPUs are very long. If Intel released Retail IVB in Fx stepping maybe it would be be a little bit more plausible, but this isn't the case
2) die measurements from E1 stepping confirmed 160 mm²
3) they measured a i5-3550. Problem: it has only GT1 graphics onboard which should be way smaller, very odd!
4) they compared an old IVB schematic and claimed old. They even don't know if this "old" schematic was ever accurate.

 

[fellix]

What would be your equivalent analysis of Llano?

I just assumed this guy knows what he's talking about:
http://www.chip-architect.com/news/2012_04_19_Ivy_Bridges_GPU_2-25_times_Sandys.html
With this diagram, I get similar numbers to what Paran was saying.

The middle die is actually from an early Haswell ES sample, not IVB.

 

[Paran]

The middle die is actually from an early Haswell ES sample, not IVB.

Omg you are right, I just checked the number at the bottom and it's identical to the Haswell sample showed at the IDF *lol*. This is some amateurish work from chip-architect. It's probably a GT2 version of Haswell, there is no surprise that Haswell will get a bigger iGPU. What makes it even more funny that they claimed it's a i5-3550 - a GT1 IVB chip.

 

[Hans de Vries]

The middle die is actually from an early Haswell ES sample, not IVB.

The middle die is from here: http://www.eetimes.com/electronics-news/4370843/Analysts-start-Intel-Ivy-Bridge-CPU-teardown

UBM TechInsights has an Ivy Bridge processor marked as a 3.3GHz Core i5-3550 chip packaged in Malaysia. It has a die size of 170 mm2,

Hans.

 

[Blazkowicz]

I would like an IOMMU function, AMD has it on llano and successors, and on sempron/athlon II X2 and higher if you get a 970 chipset or higher.

we could get it on socket 1150 unless market segmentation still doesn't allow it.
if it works you can assign the IGP to a linux or BSD virtual machine and the lowest end one is enough for basic display and unaccelerated video playback.

 

[Mintmaster]

Omg you are right, I just checked the number at the bottom and it's identical to the Haswell sample showed at the IDF *lol*.

That's mentioned on the chip-architect page, except their logic is opposite of yours - "fits perfectly with a mystery processor which was assumed to be a very early Haswell sample several months ago." It's not an IDF photo. It was supposedly taken by UBM TechInsights, not chip-architect.

1) it's very unlikely that Intel made such a big change just before mass production from E0 to E1 because validation times for CPUs are very long.

Fine, let's assume somebody messed up. Using your die photo, and scaling it up so the CPUs match those of a SB die shot (do you have a better one of these, too?), we find that IVB would be 25% bigger than SB if built on 32nm. 216*1.25*0.292 = 77mm2. 216*1.25*0.262 = 71mm2.

That's not "a different class" than Llano, assuming your 40% figure is accurate. I actually get ~35% from this die shot, not including display interfaces (which are no doubt somewhere on the IVB periphery as well), which puts it at 80mm2:
http://i.haymarket.net.au/News/20110615094344_llano-diagram.jpg

It may be a different class from Trinity's IGP, given that estimates put it at 240mm @ 32nm.

Still, IVB did really well in keeping the IGP clocks high for the mobile part. I think AMD may have a tough time there.

3) they measured a i5-3550. Problem: it has only GT1 graphics onboard which should be way smaller, very odd!

I think Intel uses the same die for i5 and i7 on SB and IVB, right?

 

[mczak]

I think Intel uses the same die for i5 and i7 on SB and IVB, right?

Hmm using just "i5" and "i7" here might not make sense as that doesn't really have much to do with what die will be used - there's even dual-core i5 after all and let's not talk about the mobile dual-core i7...
In any case, there are now 4 Ivy Bridge dies (for socket 1155 that is) - two dual core (GT1/3MB, GT2/4MB) and two quad core dies (GT1/6MB, GT2/8MB), whereas previously there were only 3 with SandyBridge - one quad and two dual core ones (again GT1/3MB, GT2/4MB).
Hence a i5-3550 could indeed use a smaller die as the top dog, but if it actually does might be a different question, it might be possible it also exists as a partially disabled part, no idea there...

 

[Mintmaster]

Ahh, okay. I knew Intel had dual and quad core dies, but thought they just disabled silicon for the HD2000/3000 variants. I guess when you do the kind of volume that they do then it's not worth wasting even a little silicon.

I guess the dualcore IVB chips will be really tiny. Intel really could put them in sub-$500 tablets as rumored and still make money.

 

[DavidC]

I guess what's possible is that they have been considering the version with more EUs but concluded that 16EUs were fine.

http://www.anandtech.com/show/4830/intels-ivy-bridge-architecture-exposed/5

I mentioned that Ivy Bridge was designed to scale up, unfortunately that upwards scaling won't be happening in IVB - GT2 will be the fastest configuration available. The implication is that Intel had plans for IVB with a beefier GPU but it didn't make the cut.

Hmm, that makes it even more likely...

 

[mczak]

I guess what's possible is that they have been considering the version with more EUs but concluded that 16EUs were fine.

I don't think intel could just scale EUs/samplers like with GT1/GT2. If they had a GT3 variant I think they'd need to at least scale ROPs too, looks very underpowered otherwise there to me. Don't know how easy it would have been to make that scalable but I guess we'll see how it's going to look like with Haswell...

 

[Deleted member 13524]

After seeing that 11.6" Clevo W110ER (4-core Ivy Bridge + GT650M), I think that AMD's only chance really is to largely under-price the Intel+nVidia solutions.

Though this seems to be happening, as they're saying this HP G6-2002AX is being sold for ~600€:

Crappy 44Wh battery in a 15" model is still crappy though..


There's also this guy who seems to have a HP-2020AU with an A6-4400M (single module 2.7->3.2GHz, don't know which GPU) and it's being sold for ~380€. Synthetic benchmark esults are scattered around that thread, starting in page 5.


May 15 is taking ages.. I want to know more about the Trinity 17W versions, and if we'll see 35W versions on slim 13" models.

 

[swaaye]

Stupid gloss finishes. I have an HP dv2 that's magnesium but has the piano black gloss finish and the awful Alps touchpad is literally a mirror. It's designed to be operated while wearing white gloves, I think...

 

[Blazkowicz]

I've found an ARM netbook which is likeable enough to buy, 189 euros shipping included, 1024x600 screen and 3G modem. only it does have this mirror finish.
I guess it's ok and even something of a nice attention from the vendor. it is a fragile device afterall, so it actually is designed to be operated with hand gloves.

if it gets too annoying we can paint it with a spray bomb..

 

[lanek]

Stupid gloss finishes. I have an HP dv2 that's magnesium but has the piano black gloss finish and the awful Alps touchpad is literally a mirror. It's designed to be operated while wearing white gloves, I think...

Oooh yes, i know what you mean... and if you are cigarette addict, forget thoses things, it will quickly get some strange white line on the surface... ( they disappear if you clean it all the time, but seriously, who clean a TV monitor 2x by weeks ) ..

I think this trend come from Samsung.. they use and have over use this piano black in all their products (Including Apple product they have design, but that is another story ) .

 

[RudeCurve]

I think this trend come from Samsung.. they use and have over use this piano black in all their products (Including Apple product they have design, but that is another story ) .