Desktop Trinity benchmarks are up.

[grimpr]

Better IPC than BD, real turbo, nice overclocks, strong chip.

http://www.tomshardware.com/reviews/a10-5800k-a8-5600k-a6-5400k,3224.html

 

[Soulkeeper]

I just read the article and I'm only mildly impressed with the FM2 platform as a whole.

I'm still bitter about the lack of an upgrade path with FM1, and I get the feeling FM2 will be a repeat.

On the cpu side, it takes a full 1GHz higher clock to barely beat llano in most things.
Generally the performance increases don't seem noticeable to people that have purchased an AMD computer in the last 1-2yrs.

The 4 module variants with L3 will be much more interesting to me.
I really wish the APUs and FX chips shared a socket, then maybe people would have an upgrade path.

On the bright side, it looks like they've managed to push the clocks a little higher with lower voltage.

 

[ShadowVVL]

It looks good but the graph for power is a bit confusing so idk if the power consumption is good or bad.

But since amd is making us wait I'm just going to head over to intel for my htpc build.

 

[PhoenixEnigma]

Just started reading. Fairly solid looking 15% increase over BD clock-for-clock, and that's without the benefit of L3 cache. Between that and what looks like a somewhat decently working turbo mode, +20% doesn't sound out of the question. On the other hand, multithreaded floating point performance still blows compared to Stars.

It's a good step in the right direction, though, I will give them that.

 

[Shaydza]

Pretty much every game is playable there so when the hit the channel I am gonna build myself an htpc with one of these suckers.

If Vishera shows the same percentage increases that trinity does it is going to mean something very bad for the console market.

Remember guys we are looking at a very early release with an immature motherboard and drivers.

Can't wait!

 

[Borealis7]

when is Vishera expected?

 

[AtenRa]

when is Vishera expected?

Im not sure but i believe Q3-4 2012

 

[frozentundra123456]

Not bad, but still kind of meh. Not really terrible at anything, but not really outstanding in anything either. Not really good enough for gaming at moderate to high resolutions and decent settings. If one is interested in gaming, I would still go with a discrete card. Trinity could be nice in the mobile sector, but at their current level of performance, I dont see much place for them in the desktop if you are interested in gaming.

 

[bill3]

If you figure they can get desktop Piledriver top model up to 4.2 ghz (which I am assuming, didn't they acquire some resonant clock mesh technology or something for clocking higher?), that's ~10% more clock, plus well I picked one particular bench, compared it to 8150 and found an 11% speedup over Bulldozer in IPC (even though the article beginning shows a couple benches at +15% I went with my figure),

Well it's a long way of saying I estimated 22% more performance from a 4.2 ghz clocked desktop piledriver lol.

So if you go and simply multiply a lot of the 8150 bench scores in their 8150 review by 1.22, to replicate a supposed 4.2 ghz Piledriver, basically you get a chip that competes well or beats a lot of the top intel chips in Tom's benches.

For example the PCMark 7 benches on this page http://www.tomshardware.com/reviews/ivy-bridge-benchmark-core-i7-3770k,3181-12.html


PCmark 7 overall suite score

3770k=5487
2550k=5079
8150=4189
8150x1.22=5111

PCmark 7 productivity suite score

3770k=5436
2550k=5103
8150=4105
8150x1.22=5008

Creativity suite score

3770k=5655
2550k=5138
8150=4792
8150x1.22=5846

And so on, I just glanced over those, but you see there 8150 goes from kind of a joke to suddenly highly competitive when sped up by 22% (even beating the $359 3770K on the third bench).

Having painted that rosy picture, I'd still worry about floating point workloads, and gaming. Still if my presupposition AMD can get 10% more clock out of Piledriver is true, they'll at least get that speedup even in those loads, possibly more I'm not sure how much the Piledriver IPC enhancements will help there.

 

[Olikan]

15% better IPC than bulldozer D:
Same IPC than Llano in many benchmarks...


humm.... the 35W laptop revew was indeed bottlenecked by the 10W less than Llano...

 

[ShintaiDK]

Look at this:

And then this:

I would be very careful about any IPC claims yet. Note the 1 module, 2 core 5400K is the fastest. A whole 7 seconds faster than the same clocked 2 modules 4 cores 5600K. Not to mention the 5600K got a +100Mhz turbo more. Or roghly something that equals that the 5400K is around 10% faster in IPC than the 5600K in this benchmark.

 

[bill3]

Look at this:


I would be very careful about any IPC claims yet. Note the 1 module, 2 core 5400K is the fastest. A whole 7 seconds faster than the same clocked 2 modules 4 cores 5600K. Not to mention the 5600K got a +100Mhz turbo more.

Whats your point? Just some bug or anamoly I guess but I dont see how it in any way opposes higher IPC claims, since 5400k and 5800k are the same chip. It's not as if it's a Bulldozer based core beating it.

If anything your evidence maybe the IPC increase is greater than Tom's figures, maybe the 5800k had a bugged benchmark there and when that's sorted the IPC increase would be even greater based on the 5400k score.

Plus, they had a second benchmark also backing up the 15% increase. Sure, I too would love to see a more detailed look into the IPC stuff though, I hope Tom's wasnt the only site that got their hands on chips to review. I also am extremely interested in how trinity's piledriver cores perform in gaming with a discrete GPU only and hope a site looks at that.

 

[T_Yamamoto]

This looks more promising than bd

 

[Olikan]

This looks more promising than bd

well, at least vishera won't be worser than Phenom 2

 

[IntelUser2000]

Looks like to me the IPC is now on par with Thuban with the module penalty kicking in with multiple threads. That's pretty good of an advancement.

 

[inf64]

The test against FX8150 in iTunes is more relevant since in this one they disabled all power/Turbo options and run both at the same clock (3.8GHz).
The results that show "stock" numbers use the power/Turbo options and that's why you can see some variance there since some chips run cooler (like 1M version).
edit: 3dsmax show the similar gain of 15% in favor of PD at same 3.8Ghz clock.So it's pretty much consistent.

It's rather safe to say that ~10% with no L3 is the uplift we will see with PD core. Add on L3 cache which generally accounts for smallish 2-3% (in some corner cases it can make a big difference though) and Vishera may end up being ~15% faster than BDver1 . They need to hit ~4Ghz clock with this thing which coupled with IPC uplift (which is real and not imaginary) can make it a very good FX successor,being from 15 to 25% faster. Coincidentally this is just what they need to match or beat 2700K/IB in plenty of MT workloads. ST will still lag behind but less than it is now,which is a step in right direction.

They need to have similar increases with SR core next year. 15% IPC uplift with 10-15% clock uplift would be perfect for them.Just about enough to combat Haswell.

edit2:
To be precise,if you take the results (@ 3.8Ghz) and turn them into seconds,PD core has exactly 108/91=1.18 or 18% shorter runtime in 1st test and 336/287=1.17 or 17% shorter runtime in second test. This is more than being 15% faster as THG stated.

 

[pelov]

15% better IPC than bulldozer D:
Same IPC than Llano in many benchmarks...


humm.... the 35W laptop revew was indeed bottlenecked by the 10W less than Llano...

IPC will vary on the workload and I wouldn't expect ~10% average but in the single digits. Trinity will still be behind Husky/Stars in IPC but only by a slight margin but the clocks will more than make up for the deficit.

I'm wondering what the power consumption was when OC'd. If the chips hit 4.4/4.5ghz with an early BIOS and poor board then maybe they do get to 4.6>4.7ghz. Though that's not that far from where it sits in Turbo anyway. Still, I wouldn't expect power consumption to be as bad as BD is considering Trinity mobile at equal TDP footing beats even Intel's 22nm parts.

The other question is how high will Vishera clock at stock? Llano, with K10.5 Husky cores clocked at ~3ghz due to APU TDP limitations. On 45nm the same architecture went as high as 3.7ghz on Denebs. If Vishera starts in the low 4 range and hits up to mid/high 4ghz turbo then it could very well be a decent chip.

 

[sm625]

Note the 1 module, 2 core 5400K is the fastest. A whole 7 seconds faster than the same clocked 2 modules 4 cores 5600K.

It is an interesting observation. Can that much performance degredation be caused by scheduling and context switching and all of its associated overhead? If so then I'd say there is something seriously wrong with this architecture. It is already hugely disappointing that it just barely beats llano even though it is running 30% faster.

I dont see how this competes with a llano 3870k when both are overclocked. The trinity is very likely to top out at around 4.4GHz, which wont beat a 3.3GHz llano much less a 3.6GHz llano.... (gaming aside)

 

[Kalessian]

It is already hugely disappointing that it just barely beats llano even though it is running 30% faster.

Might be worth it to note that (from the article):

As far as its role in Trinity, the benchmarks will show that the Piledriver architecture generally outperforms Llano’s Stars design, particularly in applications that emphasize integer math. When you start taxing Piledriver’s shared floating-point resources, older Llano-based APUs still wind up delivering better performance, though generally by slim margins.

Are any of the llano v trinity tests single-threaded and run at the same clock rate w/o turbo? I would like to see that.

 

[IntelUser2000]

Might be worth it to note that (from the article):

Are any of the llano v trinity tests single-threaded and run at the same clock rate w/o turbo? I would like to see that.

The iTunes test and Lame test I think only takes advantage of 2 threads. But other than that it runs at default, meaning Trinity runs at higher clock speeds. The only clock for clock comparison is against Bulldozer.

 

[veri745]

edit2:
To be precise,if you take the results (@ 3.8Ghz) and turn them into seconds,PD core has exactly 108/91=1.18 or 18% shorter runtime in 1st test and 336/287=1.17 or 17% shorter runtime in second test. This is more than being 15% faster as THG stated.

You're using the new CPU as the baseline instead of the old CPU.

287/336 = 0.854, i.e. 15% faster.

 

[Nemesis 1]

Where are you guys getting IPC increase of 15%. Certainly not based on the benchmarks in this preview. OR do we do % differant for AMD than intel . If trinity is a 15% increase in IPC intel IVB is 25% faster IPC than SB. Funny you AMD guys are . Alot like the Dems.

 

[Kalessian]

Well looking at this graph we can get a little more insight into ph2 vs piledriver (seems the wav file is a little shorter in this ph2 test, also the ph2 runs at 3.7ghz, not 3.8ghz and itunes version isn't the same so it's far from perfect):

The 8150 is probaly turboing to 3.9ghz so call it 3.8ghz average? Wav seems to be ~5% longer, but ph2 980 is ~2% slower [than 3.8ghz]... I scaled the ph2 at 3.8ghz to be 1:25 compared to 1:31 for trinity Really dirty comparison but trinity looks to have just about caught up with ph2. Is llano better than ph2 clock for clock?

 

[Don Karnage]

Toms really fail's at benchmark reviews.

 

[beginner99]

Looks like to me the IPC is now on par with Thuban with the module penalty kicking in with multiple threads. That's pretty good of an advancement.

Yeah, now has similar IPC as a 2007 intel CPU...great achievement.