Trinity vs Ivy Bridge

On a 4GHz P4 under Linux I ran suprpi in under 15 seconds. Same CPU on windows took nearly 3x longer. They are compiled with different compilers and the difference is HUGE.

 

When using some software, it tells nothing about memory bandwith to gaphics because:

1) The poor result from those benchmarks are likely becuase the specific software fails to understand the interleaved memory rows only reading from one channel each time, a more random pattern is more likely to use full bandwidth.

2) There is a special link between graphics and the memory controller, it does not have the same limits as the CPU.


Personally I doubt the performance level of Llano would be possible if it was limited to the bandwidth measured by Sandra...

 

AMD announced that Trinity and Brazos 2.0 systems will soon be coming our way.

When should we expect some reviews on desktop and notebook systems? Is anyone aware of a NDA date?

 

superpi has also wild differences between CPU implementations. it's best used to look at overclocking or memory timing results, and to know if your PC is stable enough for it to not diverge right away.

 

SiSoft Sandra measures 250GFLOPs on HD 4000@1,15GHz: www.computerbase.de/artikel/grafikk...s-hd-4000-und-2500/10/#abschnitt_gpucomputing (last test)

 

Wow, in that computerbase review the IvyBridge lags quite a bit behind Llano in gaming benchmarks.
In many cases, it seems to perform near a discrete HD6450 (160 VLIW5 sp, 8 TMUs, 4 ROPs @ 750MHz).

 

Is this benchmark even reliable? The GT430 score is some 15% above its theoretical ALU throughput.

 

I'm a bit surprised that in some games performance is barely faster than HD3000 but in some it's much faster. I certainly didn't expect fully uniform scaling but not that wild numbers neither.
In any case, it seems to be just about fast enough to catch the fastest Llano on the mobile front (reason of course being gpu clocks are essentially the same for intel but much lower for amd compared to desktop). It should also catch some of the slower Llano desktop parts though the good news for amd is that of course these are in a different price class, even the chips with HD2500 (which certainly aren't close to the faster desktop Llano parts) are more expensive. And of course Trinity should make sure there will be a lead again in gpu performance for the mobile parts (well at least I'd hope so...).

 

Now, that's curious:

http://www.chip-architect.com/news/2012_04_19_Ivy_Bridges_GPU_2-25_times_Sandys.html

Two IVB revisions with different GPUs? And why Intel is providing only a die-shot of the "older" revision?

 

I expected a 10-20% better performance based on the earlier leaked Vantage results. It looks like IVB is 35-40% faster than 2600k/2700k HD3000 (differs from review to review) and about 50% in the mobile segment. That would mean that the GPU gap between Intel and AMD doesn't change much with Trinity. It could be another story for the 17W versions though. Apart that the DX11 implementation looks pretty good especially Tessellation scaling. Also the anisotropic filtering improved a lot finally.

 

A link, please?
Thank you.

 

http://www.computerbase.de/artikel/...0-und-2500/11/#abschnitt_tessellationleistung


Llanos performance hit with Tessellation is bigger.

 

Thanks.
Sorry for the laziness.
I expected images and image quality comparisons, anisotropic filtering, because you said "improved a lot" and I wondered what that was.

http://www.computerbase.de/artikel/grafikkarten/2012/test-intel-graphics-hd-4000-und-2500/4/

Although the HQV Benchmark 2.0 result is the same: 158 for both Ivy and Sandy.

http://www.computerbase.de/artikel/grafikkarten/2012/test-intel-graphics-hd-4000-und-2500/6/

 

Regarding the AF:

http://www.computerbase.de/artikel/...00-und-2500/4/#abschnitt_anisotrope_filterung
http://ht4u.net/reviews/2012/intel_ivy_bridge_core_i7_3770k/index15.php
http://www.anandtech.com/show/5771/the-intel-ivy-bridge-core-i7-3770k-review/8

They improved the angle dependency significant.

 

The retail 3770K die shown on this website shows 160mm2 though.

http://www.overclock.net/t/1249419/pcevaluation-intel-i7-3770k-temperature-measured-without-ihs

I used the 6th picture there

CPU Package: 618x618 pixels = 37.5mm x 37.5mm
CPU die: 324x134 = 159.9mm2

That's pretty much what's expected.

 

That's correct except that the package has "INTEL CONFIDENTIAL" on it so
it's not a retail version.

There's another 160 mm2 die which is actually a retail version but it's from a
Xeon which doesn't use the iGPU (which is different on the two dies)

So for the moment it's still a bit of a mystery.

Hans

 

Given the amount of die space Intel devoted to the HD4000 on its 22nm process, AMD seems to have quite a large architectural advantage. I guess it's expected, but good to see as AMD definitely needs it.

Trinity is probably going to show us why AMD bothered with VLIW4 for only one part on the desktop side, as they seem to have made quite a leap without a new process. I hope they can deliver on the 17W part.

 

Llanos iGPU allocation is much bigger (28-30% vs 40% of the overall chip). Even without the shrink the iGPU part would be far smaller than Llanos.

 

was it mentioned in any review if Ivy has the same limitation of Quicksync only being available if the IGP is used?
More importantly, does GPCompute (OpenCL/DirectCompute) work with a discrete GPU as primary card - thinking of games using the IGP as GPCompute card and leaving the discrete one for graphics.

 

I thought that on Sandy Bridge you could use quicksync if you had a Z68 chipset and no graphics card, whereas you couldn't run it at all with a P67 chipset.
Well, I know remember it, it's available if you run the Lucid Virtu software. or even if you use a dummy VGA dongle to make your computer think a display is attached. it's funny and ridiculous but has been done I believe.