• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Trinity Internal Benchmarks Surface

btarunr

Editor & Senior Moderator
Staff member
Joined
Oct 9, 2007
Messages
47,677 (7.43/day)
Location
Dublin, Ireland
System Name RBMK-1000
Processor AMD Ryzen 7 5700G
Motherboard Gigabyte B550 AORUS Elite V2
Cooling DeepCool Gammax L240 V2
Memory 2x 16GB DDR4-3200
Video Card(s) Galax RTX 4070 Ti EX
Storage Samsung 990 1TB
Display(s) BenQ 1440p 60 Hz 27-inch
Case Corsair Carbide 100R
Audio Device(s) ASUS SupremeFX S1220A
Power Supply Cooler Master MWE Gold 650W
Mouse ASUS ROG Strix Impact
Keyboard Gamdias Hermes E2
Software Windows 11 Pro
"Trinity" is the codename of AMD's next-generation performance accelerated processing unit (APU) family. Based on the new socket FM2 package, these chips will take advantage of AMD's next-generation Piledriver processor core architecture and VLIW4 GPU stream processor architecture. Together, Trinity promises increased general, visual, and parallel compute performance. Some of the slides detailing AMD's own performance estimates were put up by DonanimHaber in their recent video bulletin. We screen-grabbed the performance graphs from the low-resolution video, hence the grainy images.

To begin with, AMD is promising noticeable performance improvements over the current "Llano" APU. It spread its benchmarks across three categories: visual performance (using 3DMark Vantage), general performance (using PCMark Vantage), and parallel compute (GPGPU) performance (calculated CTP SP GFLOPs). With 3DMark Vantage, Trinity A8 (quad-core), A6 (triple-core), and A4 (dual-core) APUs are seeing a roughly 32% improvement over their respective Llano-based counterparts; with general performance, the improvement is a candid 13.8% on average; but with GPGPU performance, the improvement is a massive 56.3% on average. This could be attributed to the VLIW4 architecture. Lastly, there are notable CrossFire dual-graphics performance improvements.



View at TechPowerUp Main Site
 
Looks promising if it's not bollocks.

I'm trying to build as cheap yet effective a budget PC for my mate that will play Civ V and Football Manager (the only games he plays). I'm aware Civ V is atrociously optimised. I had been looking at the Llano A8. This looks like a markedly better bet.

He will be using Photoshop CS 5 though which i think is CUDA optimised?
 
Blurry AMD internal benches mean pigeon poo...but, this does show kinda what I was expecting Trinity to be. Gonna be a bit more impressive than Llano, but not totally mature.

Is nice to see they got the dual GPU working better. But I am curious to see the real world benches. So many of those had dual GPU slower than single.
 
member.php
i dont always trust rumors, but when i do they are from donanimhaber

LOL.
i hope these arent like the bulldozer ones.
 
actually these ones seem quite realistic and believable. compared to bulldozer anyway.
 
So the Llano A8 is around 6670 performance GPU-wise, so the Trinity one is now 6790 I guess?
 
Fool me once, shame on you, fool me twice...
 
It will be close to a 5570. When it hits 5670 I'll be impressed.
 
So the Llano A8 is around 6670 performance GPU-wise, so the Trinity one is now 6790 I guess?

You (and I) wish... Llano A8 is more like a HD5570. Next gen will probably be the equivalent of HD6650. HD6670 has GDR5 instead of DDR3 than apu will work with, so it will be better. Hope they will do some tweaks and pilledrive apu will be faster than HD6550 but i don't think it will better than HD6670.
 
This where amd shines.
 
Fake.

I'm thrilled I bought my A8-3850 instead of waiting like a sucker for a bulldozer-based FM2 part. I would be willing to bet that it will be a turd, apart from the GPU improvements.

And guys, its marketed at a 6550 discrete part.
 
I guess 2013 will be the first proper BALANCED Fusion, with what? SteamRoller by then? CPU + probably HD79xx series's CU architecture in the GPU combined with 22nm process(HOPEFULLY)
 
So the A6 is exclusively triple core now? If that's the case, then triple core Trinity is beating the pants off quad core Llano (unless they're comparing against the one triple core Llano). Very impressive.
 
Looks promising, but when these APU's have at least 800 stream processing units and Phenom II CPU performance, I will be definitely interested ;), but that might be awhile :(
 
Looks promising, but when these APU's have at least 800 stream processing units and Phenom II CPU performance, I will be definitely interested ;), but that might be awhile :(

Llano already has comparable CPU performance to Phenom II. You only need to wait for stream processors.
 
I looked at the pics, saw the word "ESTIMATED", and decided that this is nothing more than hot air.


These are DESIGN TARGETS, not actual perforamnce.

Thanks.

:D
 
any news about price ?
which would be a good choice
amd trinity + 6750
or
i3 2100+6750 ?
 
Last edited:
now with 60% more blurriness!!!
 
So the A6 is exclusively triple core now? If that's the case, then triple core Trinity is beating the pants off quad core Llano (unless they're comparing against the one triple core Llano). Very impressive.

No triple cores after Llano; unless they disable ONE integer core in the second cpu module.
That doesn't sound right to me... 1 module, 2 integer cores. 2 modules, 4 integer cores. 3 integer cores? weird.
This sounds fake to me.
 
Looks promising, but when these APU's have at least 800 stream processing units and Phenom II CPU performance, I will be definitely interested ;), but that might be awhile :(

High-end Llano doesn't make sense. You won't see it anytime soon.

Problems:
-Huge price
-Too much heat to dissipate
-Huge die
-Memory bandwidth couldn't feed 800sp

mo problems
 
No triple cores after Llano; unless they disable ONE integer core in the second cpu module.
That doesn't sound right to me... 1 module, 2 integer cores. 2 modules, 4 integer cores. 3 integer cores? weird.
This sounds fake to me.

Yeah, it's kinda weird, but you can disable one of the cores in a module on an FX processor. An 8 core FX processor with only four enabled cores (exactly one per module), not only works, but usually performs better in single threaded tasks than if all cores were enabled. This is because a non bulldozer-privy scheduler might put another thread (maybe some background thread, or what not) on the other core of the same module that has a core running the single-threaded task. Then, resources become shared, and slow things down. Such worse case scheduling doesn't always happen, but when it does, it's not all bad, as power consumption will be theoretically lower in such case. Just a little extra info for those who think a magical Windows 8 scheduler will make FX chips much faster - it'll help minimize the occurrence of the worse case listed above, and for the most part, that's it.
 
Yeah, it's kinda weird, but you can disable one of the cores in a module on an FX processor. An 8 core FX processor with only four enabled cores (exactly one per module), not only works, but usually performs better in single threaded tasks than if all cores were enabled. This is because a non bulldozer-privy scheduler might put another thread (maybe some background thread, or what not) on the other core of the same module that has a core running the single-threaded task. Then, resources become shared, and slow things down. Such worse case scheduling doesn't always happen, but when it does, it's not all bad, as power consumption will be theoretically lower in such case. Just a little extra info for those who think a magical Windows 8 scheduler will make FX chips much faster - it'll help minimize the occurrence of the worse case listed above, and for the most part, that's it.

Hyper-Threading Plus. 8 "Core" FX CPU's are just Quad-Cores with 50% more resources and bottlenecks. You can claim it's a full core because it has enough individual resources, but the bottom line is 2 threads move through 1 core\module\unit\banana-phone at a time. I find it humerous that a 8-"Core" FX CPU with half it's "Core"s disabled performs on par or better in tasks that only utilize up to 4 threads.
 
Back
Top