Trinity vs Ivy Bridge

That's a guru3D piece acknowledging your linked TPU news as source which is again referencing this HotHardware article, which does not mention DX11.1 at all. That's something TPU "invented" or just assumed from the HD7000 name and added to the story. The guy in the video they have there is actually saying the Trinity graphics has "second generation DX11 graphics", which basically means the NI generation which also fits with the VLIW4 shader architecture. It does not say DX11.1 or third generation or something like that.
And Anandtech just speculates about DX11.1 while flatly saying the slides only mention DX11 without the ".1".
 
Actually, the AMD material on the matter has been worryingly vague. They promise that 10-15 % of yearly performance increase for each new core. But nowhere is it clearly stated that those improvements refer to IPC alone or they do factor in the clock increase.
I vaguely remember some slide where it mentioned that bigger part of the increase comes from manufacturing improvements.
 
I'm not sure but I think this was not discussed here yet:

The score for the 2012 AMD A10-4600M on the “Pumori” reference design for PC Mark Vantage Productivity benchmark shows an increase of up to 29% over the 2011 AMD A8-3500M on the “Torpedo” reference design. The AMD A10-4600M APU has a score of 6125 and the 2011 AMD A8-3500M APU scored 4764. Scores rounded to the nearest whole number

http://blogs.amd.com/fusion/2012/03...ia-experience-for-our-“connected”-generation/

compare with this:

http://pc.watch.impress.co.jp/img/pcw/docs/486/640/html/graph09.gif.html

..especially the FX4100-score. Therefore the A10-4600M with 35W TDP (including GPU) is as fast as a FX4100, at least in the productivity benchmark.
 
Regarding the IPC improvements..I would expect Trinity to have a significantly higher single threaded performance than Llano. Given that BD and Llano were actually fairly close in terms of IPC per clock, Trinity wont gain much by its architecture changes alone (even if we see gains near the upper end of the 10-15% band given by AMD). Where it should gain the most is clocks. The architecture is designed for higher clocks. Its got a better implementation of Turbo than Llano and should be able to sustain higher turbo clocks for single threaded mode. Secondly Trinity is launching about a year and a half after Llano so the process should be a lot more mature. Look at the case of Phenom X6. By the time it launched, they could fit 6 cores at the same TDP of the 4 core launch variants (at the same clocks)

Edit: Also given the fact that Trinity has only 2 modules, there may very well be some corner cases where it actually loses to a 4 core Llano in highly threaded workloads
 
It looks like Trinity will be able to offer decent CPU performance especially for laptop gaming and light productivity.
Judging by FX-4100 scores compared to i3-2100 it mostly should be close contest. Still i5 looks to be out of reach for AMD core, but again, for laptop usage you usually don't need that much CPU speed.
I'm very curious about AMD's VCE engine, was hoping we can get some clues as to how it works and how flexible it is from GCN discreet card tests. Unfortunately this still isn't the case. Any fresh info regarding software support for VCE?
 
Trinity looks to shaping up quite nice!, I think if they can get 25% better 'serial' performance out of those Piledriver cores, and a decent 35% on the gpu for a 35w tdp..then that looks to be a winner.
Any idea what the Bandwidth situation is gonna be like on these puppies??
 
Trinity looks to shaping up quite nice!, I think if they can get 25% better 'serial' performance out of those Piledriver cores, and a decent 35% on the gpu for a 35w tdp..then that looks to be a winner.
Any idea what the Bandwidth situation is gonna be like on these puppies??

Supports DDR3-2133 compared to DDR3-1866 for Llano, so a very modest increase. But i doubt we're gonna see anything more than DDR3-1600 in the mobile space at least.
 
Ivy Bridge HD 4000 @ 3DMark Vantage Feature Tests:
attachment.php


sources:
HD 4000
HD 3000

In HD4000 link is also a new 3DM11 score included: P764.
 

Attachments

  • Auswertung HD 4000 vs HD 3000.png
    Auswertung HD 4000 vs HD 3000.png
    7.3 KB · Views: 672
Nice boost in POM and Particles tests -- those are texture fetch intensive, if I remember right. Colour fill in Vantage is FP16 RT blending, so it's primary limited by the narrow frame-buffer bandwidth. The new caching architecture is doing very well, it seems.
 
Intel is cooking something big up with Haswell...the only weakness in AMD's APU graphics that i can see is bandwidth....everything else they are well out in front of Intel..(although the gap is starting to close.)...if Haswell ups the EU's to the rumoured 40 and gets some special bandwidth sauce then who knows...
 
Intel is cooking something big up with Haswell...the only weakness in AMD's APU graphics that i can see is bandwidth....everything else they are well out in front of Intel..(although the gap is starting to close.)...if Haswell ups the EU's to the rumoured 40 and gets some special bandwidth sauce then who knows...

It's far too early to say "is starting to close" when we have only leaked material and some PR-slides of the other, 3DMark is irrelevant, we need to see IB vs Trinity in games to see if it's closing or not.
 
superPi and 3DMark06

Top Trinity desktop model superPi and 3DMark06 scores were leaked.

Its GPU has really made a noticeable advancement. However, the CPU part looks slightly clock-to-clock slower than the BD in superPi and on par in 3DMark (3.8/4.2GHz frequencies for this Trinity model).

http://bbs.pceva.com.cn/thread-39867-1-1.html
 
Top Trinity desktop model superPi and 3DMark06 scores were leaked.

Its GPU has really made a noticeable advancement. However, the CPU part looks slightly clock-to-clock slower than the BD in superPi and on par in 3DMark (3.8/4.2GHz frequencies for this Trinity model).

http://bbs.pceva.com.cn/thread-39867-1-1.html

http://wccftech.com/amd-trinity/

Super Pi 1M Calculations:
A10-5800K: 23.775 Sec
A8-3850: 26.039 Sec

3DMark 06 CPU Score:
A10-5800K: 4304 Marks
A8-3850: 3814 Marks

3DMark 06 GPU Score:
A10-5800K: 3285 SM 2.0/ 4067 SM 3.0
A8-3850: 2139 SM 2.0/ 2552 SM 3.0



However, we don't know what memory was used in the benchmarks.
If we assume DDR3-1600 for both, with Trinity having some serious 63% performance advantage over Llano, I think these are quite positive results on the GPU side. If it's bottlenecked by memory bandwidth, using DDR-1866 or DDR3-2133 will probably widen this advantage.

Then again, CPU performance remains uninmpressive. AMD really needs to start showing results (or ask reviewers to start benchmarking) in GPGPU tests.
 
http://wccftech.com/amd-trinity/

Super Pi 1M Calculations:
A10-5800K: 23.775 Sec
A8-3850: 26.039 Sec

3DMark 06 CPU Score:
A10-5800K: 4304 Marks
A8-3850: 3814 Marks

3DMark 06 GPU Score:
A10-5800K: 3285 SM 2.0/ 4067 SM 3.0
A8-3850: 2139 SM 2.0/ 2552 SM 3.0



However, we don't know what memory was used in the benchmarks.
If we assume DDR3-1600 for both, with Trinity having some serious 63% performance advantage over Llano, I think these are quite positive results on the GPU side. If it's bottlenecked by memory bandwidth, using DDR-1866 or DDR3-2133 will probably widen this advantage.
Mem clock is far less important for AMD than their mem controller. It was positively horrible in Llano.
 
Back
Top