Desktop Trinity benchmarks are up.

[Olikan]

4.5GHz at 1.152V

 

[Homeles]

Well The fact that you said I said there was no IPC increase is amusing . No where did I make such a claim . . But if your going to Figure IPC . Its has to be the same method that is used to determine Intel IPC increases. Clocks normalized and run the benchmarks ./ Its that simple . IVB was run against the same clocked SB . normalized. A 15%IPC increase for trinity is = to about a 10% increase in IVB.

I didn't say that... obvious straw man is obvious.

 

[Riek]

My first impression from this preview is a bitter disappointment. Still relies on high clock to be relevant. Mini-Netburst, all over again. But let's wait for the retail samples first

The frequency is a design choice and won't change.

And netburst was brilliant... Northwood at least . If AMD doesn't do something like prescott, Bulldozer -> trinity -> steamroller should be great transisions for them.
Just don't expect a frequency oriented design( compared to others) turn around in an ipc oriented design.

When looking at the cpu picture alone, the progress isn't really there. But when taking into account the extra gpu power, the same TDP budget,t similar die size and the same flunky process... the result is impressive imo. (GLOFO 32nm isn't as lucky as intels 32nm or their own 45nm for that matter... its closer to AMD 65nm process )

4.5GHz at 1.152V

That is assuming the screenshot isn't fake and the voltage is read out correctly.

 

[Homeles]

And netburst was brilliant... Northwood at least .

The L2 went from 256KB to 512KB and the die shrunk...

 

[inf64]

I didn't say that... obvious straw man is obvious.

Actually that guy (is he trolling you ? ) is clueless or didn't even read the article since THG did run the tests at same clock . They have ran 2 distinctively different benchmarks at same clock speed with Turbo and power management options disabled on both Trinity and FX8150. To quote the article:

We took the A10-5800K, set it to 3.8 GHz, turned off Turbo Core and any power-saving feature that’d spin the chip down. Then, we took FX-8150, overclocked it to 3.8 GHz, and disabled all of the same features. By running a single-threaded workload like iTunes, we could neutralize the difference in core count (though, if anything, FX could have benefited from its 8 MB L3). Nevertheless, Piledriver clearly completes our workload much faster, yielding a 15% improvement, per clock cycle, over Bulldozer.

Turning off two of FX-8150's Bulldozer modules gives us the opportunity to run a threaded workload like 3ds Max without slanting the result toward Bulldozer. And once again, the Piledriver-based APU wins by roughly 15%.
Ivy Bridge was only about 4% faster at a given clock rate than Sandy Bridge. So, while we’re fairly certain that a Piledriver-based FX wouldn’t overtake the newest Core i7s, it should be more competitive than today’s Bulldozer-based CPUs. Where does the speed-up come from? Doesn't appear to be cache latency; Sandra shows the same results for Bulldozer and Piledriver.

It's obvious that in these two very different workloads we see the same story. Both chips hard locked @ 3.8Ghz,both with all power/Turbo options disabled. Similar results in both cases,Piledriver core without L3 cache is around 15% faster than 2 module FX8150 with 8MBs of L3 at the same clock.

 

[Olikan]

That is assuming the screenshot isn't fake and the voltage is read out correctly.

i doubt it's fake....Tom's Hardware would be shooting theyr own foot doing that

reading again, it's a voltage read problem... trinity uses 1.3V at stock clocks
it's a heavy undervolt and overclock, to be realistic

 

[Riek]

The L2 went from 256KB to 512KB and the die shrunk...

And they eventually enabled hyperthreading, they reached 3GHz and were doing pretty ok.
The changes might be minor, but the clockspeed could be upped tremendously, as
was its design choice and it performed great during that time. It was only when AMD came with A64 and Intel went prescott style the netburst design failed to be competitive.

doubt it's fake....Tom's Hardware would be shooting theyr own foot doing that

reading again, it's a voltage read problem... trinity uses 1.3V at stock clocks
it's a heavy undervolt and overclock, to be realistic

Ow its from tomshw. I remember i read he needed 1.5V for it?

 

[sm625]

What really boggles me is that the FCH is pretty much unchanged, yet it is still a separate package. Why not integrate it into the cpu? Or at least put something on there. I understadn that it would be a lot of pins, but there has to be a decent compromise. They could give me 4 SATA, 6 USB, audio, LAN, TPM, and a couple pcie lanes at least. I just have to bang my head against to wall trying to understand why they need this chip. Why add an extra $10 or $15 to the cost of the platform for what appears to be no reason? Who is going to use 8 sata ports??? I swear these guys are just hellbent on going bankrupt.

 

[pelov]

Enough to be considered "new" I'd guess. It's done to please OEMs so they can sell more motherboards and isn't exactly restricted to motherboards either. Think about the different chips that Intel and AMD sell that are essentially just different bins with an incremental 100mhz bump. Why in the world would you need like 10 different chips? You don't, but it provides OEMs/retailers with a wide variety of options and incentive for upgrade

 

[ShintaiDK]

In terms of the last chipset features to move to the CPU. We will have to wait abit more. Haswell/Broadwell will somewhat do it for mobile. So realisticly I say its something that gonna be standard for AMD and Intel in around 2015. Then chipsets will be gone for good.

 

[frozentundra123456]

Getting mighty close to high settings playable at 1080p with no discrete card.

1 more generation and I think we might see the end of all GPUs under $100 for good. 1080p at 30FPS + and high settings? Yesplz

I think I might be close to tossing my desktop in favor of a small APU HTPC. It looks like I may get my wish in 2013. Couple that with a nice Asus tablet and dock and I'm all set

Well, there is quite a ways to go for high settings at 1080P.

Skyrim: <30FPS at medium
Arkham City: the only game that makes the cut at high settings
WOW: Cant say because only tested at good quality.
Diablo 3: barely beats 30 FPS at low quality

I only have a 900p monitor, but my 5 year old desktop with an E4500 and 9800GT easily will play D3 and Skyrim at high quality.

I still dont see the point of an APU on the desktop (for gaming) when it is so easy to add a discrete card and get much, much better performance.

 

[Olikan]

I only have a 900p monitor, but my 5 year old desktop with an E4500 and 9800GT easily will play D3 and Skyrim at high quality.

I still dont see the point of an APU on the desktop (for gaming) when it is so easy to add a discrete card and get much, much better performance.

if you add a GPU for crossfire...you end up having a better cpu and gpu performance

trinity looks good...it's just a matter of price now

 

[Soulkeeper]

Here are some benchmarks comparing my A8-3870 @ 3.4 against bulldozer and such at similar clocks

 

[inf64]

Unless we get 5800K's Phoronix benchmark results I don't see how is that relevant. We already have Llano's perf. numbers in THG preview (win7).

 

[Soulkeeper]

Unless we get 5800K's Phoronix benchmark results I don't see how is that relevant. We already have Llano's perf. numbers in THG preview (win7).

Relevant is subjective.

Took me 2hrs to run those benchmarks just now in the hope they'd be usefull ...

Regardless, if you see trinity as having a 15% increase in performance compared to bulldozer modules at the same clock just imagine that FX-4100 chip 15% higher ...

Yes, I too am eagerly waiting for phoronix leaks so I can compare

 

[happysmiles]

I still dont see the point of an APU on the desktop (for gaming) when it is so easy to add a discrete card and get much, much better performance.

Its for those on a budget and if the driver updates are any indication, they're trying to crossfire everything so apu+discrete = ^_^

 

[KompuKare]

In terms of the last chipset features to move to the CPU. We will have to wait abit more. Haswell/Broadwell will somewhat do it for mobile. So realisticly I say its something that gonna be standard for AMD and Intel in around 2015. Then chipsets will be gone for good.

Intel's revenue won't like that though will they?

Seems to me Intel are running a nice racket with the price of their chipsets now that their questionable patents on interfacing to their CPUs has eliminated all other chipset makers (I think those patents are up there with printer manufacturers patenting their ink/toner counting chip in cartridges). I mean Intel charge quiet a bit for those 64nm chipsets:

Intel will formally announce its Z77, H77, Z75 and B75 chipsets on April 8 with wholesale prices of $48, $43, $40 and $37 USD respectively, and will subsequently launch Q77 and Q75 chipsets at $44 and $40 USD next month on Friday, May 13th.

(http://fudzilla.com/home/item/26609-intel-z77-chipset-lga-1155-boards-launching-april-8th)

I know, I know I'm a cheapskate and the 3770K is a massive bargain and all that, but if Intel integrate the chipset what will they do with their old fabs before they are upgraded. Even if they want to start doing contract work with their fabs, who would be that interested in 65nm atm?

 

[ShintaiDK]

Intel's revenue won't like that though will they?

People said the same when we got Lynnfield/Clarkdale. Intel responded with record profits and revenue. Personally I just think CPUs will cost those 25-45$ more, and boards those 25-45$ less.

 

[KompuKare]

Its for those on a budget and if the driver updates are any indication, they're trying to crossfire everything so apu+discrete = ^_^

Maybe the reason the GCN driver team is behind in crossfire support is because lots of people are busy polishing hybrid crossfire...

Unlikely knowing AMD and drivers (although the current AFDS seems to better than past AMD shows), but if they were to get hybrid crossfire working well, anyone who bought a desktop based on an AMD APU would lose quite a fair bit of performance if they later upgraded to a Nvidia card.

What I would be more interested in is teaming a discrete GPU with Trinity+ which supports ZeroCore and shuts off totally when not being used. The cheapskate CPU upgrades (buy value CPU and overclock like mad) I was used to have come to an end, so my next upgrade is likely to focus on power saving at idle and noise.

 

[inf64]

Relevant is subjective.

Took me 2hrs to run those benchmarks just now in the hope they'd be usefull ...

Regardless, if you see trinity as having a 15% increase in performance compared to bulldozer modules at the same clock just imagine that FX-4100 chip 15% higher ...

Yes, I too am eagerly waiting for phoronix leaks so I can compare

Ok maybe I didn't word it correctly,sorry about that. I appreciate your time and effort ,it's just the fact that Piledriver Vs Bulldozer is very tricky comparison. In THG test we see 2 tests that show 15% improvement and this is with PD core that has no 3rd level cache. Now check this one out,a Passmark result(3 samples in database so it's not a glitch) of Trinity 4600M( 2.3Ghz QC with 3.2Ghz Turbo for single core;passmark stresses all CPU's resources and has single and multithread subtests;it's safe to say that FX4170 is running at 4.2Ghz versus average clock of 2.7Ghz for 4600M):

http://www.cpubenchmark.net/high_end_cpus.html

AMD A10-4600M APU 5,365
AMD FX-8150 Eight-Core 8,273
AMD FX-4170 Quad-Core 4,574
AMD FX-4100 Quad-Core 3,979
AMD A8-3870 APU 4,794

As you can see above ,Piledriver core performs MILES ahead of Bulldozer Version1 core in Passmark. A score of 5365pts on a mobile chip that runs at 2.3-3.2Ghz is massively better than say FX4100 @ 3.6ghz which scores 3979pts. If you take a closer look at clock scaling and core scaling,for example a 16% higher clock(FX4107 vs FX4100) results in 4574/3979=1.15 or 15% better score in Passmark. So clock scaling applies. Further more,if you look at core scaling, 2x more cores/resources results in almost exactly 2x more points in Passmark: 8273pts(FX8150@3.6Ghz)/3979(FX4100@3.6Ghz)=2.07 or 2x better performance. The scaling is a bit better than two times since FX8150 has higher all core Turbo (3.9Ghz) versus FX4100(3.7Ghz).

Now if we look at the Passmark score for "lowly" AMD 4600M which is 2M/4C part with stock 2.3Ghz clock and take a geometric mean of stock and turbo clock(2.7Ghz) and then apply the clock uplift of 5800K (4Ghz "mean" clock;geometric mean of 3.8 and 4.2Ghz stock/Turbo) we get : 5365pts*4Ghz/2.7Ghz=7950pts. To put this number in perspective,this is just a hair slower than 8core FX score. At equal to FX4170 clock,at worst(running the whole time at highest Turbo of 3.2Ghz which is not possible) the 4600M would score 5365*4.2/3.2=7040pts which is 54% faster . I don't even want to speculate how high Vishera 8C @ 4Ghz may score if mobile quad core 2.3Ghz 4600M APU scores 5365pts.

Now,it is true that some subtests in Passmark are running much better due to fixes in Piledriver (like integer divider fix) but these alone are not enough to inflate the score so much. PD core indeed runs better than Bulldozer,but determining whys and hows in particular workloads is very hard. That's why Phoronix test suite of FX4100 is not much helpful when it comes to Piledriver. Do we even know if FX4100 was using optimized compiler switches for those rendering benchmarks? There is a whole article @ Phoronix about how much compiler affect Bulldozer's performance. The conclusion is: there is a massive difference between optimized binaries and non-optimized ones. As much as 100% in some cases(using FMA/XOP instructions vs legacy SSE for example).

Edit:
"AMD's FX-8150 Bulldozer Benefits From New Compilers, Tuning"

On the last page you can see the effect of proper Bulldozer support in C-ray. Your link states 51s for FX8150 with whatever compiler it was used. In the article above ,on the last page,a proper bulldozer compiler support results in a total time of 26.92s which is roughly 2x better(!). Take a look at all other subtests from both links and you can see that in many subtests FX8150 practically doubles its scores with proper compiler support. That is an amazing speedup.

 

[Nemesis 1]

I didn't say that... obvious straw man is obvious.

No straw man . here is what you said .

The funny thing is, I agree with you. We do need a more thorough review. I was just destroying your ridiculous claim that there wasn't anything showing IPC increases in this review, when Tom's specifically addressed that issue.

Your a new member here. cough. You come in here tring to confuse people with your IPC is same as clocks . Lol . Look every member here understands that only X% of people O/C O/Cing benchmarks are not important only to people like us. Everyone else runs at stock .
AT when doing compares between the last generation and the new . Basicly tries to normalize the clocks as he should . But You AMD guys always want to change the rules of past practice inorder to overlook short comings. Its been that way now for 6 years and likely the next 6 years if amd manages to stay in the game.

Amd has a hard time competing with intel on x86. Now they are going to try to align with other companies . That in time AMD will turn around and try to sue for unfair business practicies . Its AMds way.That market has many players . Once intel comes with 22nm atom . Those companies will slowly die off. Just as they did in the early years of x86. Same story differant song. Apple is the company that has my interest . Will they switch or elect to die as they did many years back. Fablesss companies are in danger of losing everthing. Intel is helping acer out in many ways with tablets. Is this unfair to others . Hell no its not. But AMD isn't happy about . and will likely file yet another suit against intel . After Windows 8 NO more X86 monoply all cpus will run on same operating system. Than watch what intel does. Medfield clearly shows intel is in the game . 2nd generation atom puts them ontop of the heap. In all things

 

[frozentundra123456]

if you add a GPU for crossfire...you end up having a better cpu and gpu performance

trinity looks good...it's just a matter of price now

If you add a discrete card to an Intel CPU, you get better CPU performance and are not limited to AMD cards. Also does anyone know what AMD cards support crossfire with the IGP? I think you are limited to low/midrange cards.

I still dont think the APU is of any great benefit for gaming on the desktop. Just add a discrete card. Matter of fact, I would even prefer that Intel CPUs for the desktop were more focused on the CPU and even came without the IGP.

Laptops, where space and power savings are critical, yes an IGP is useful, but for the desktop, not so much, at least for anything above very low end, non-gaming, and perhaps HTPC uses.

 

[Riek]

Your a new member here. cough. You come in here tring to confuse people with your IPC is same as clocks . Lol . Look every member here understands that only X% of people O/C O/Cing benchmarks are not important only to people like us. Everyone else runs at stock .
AT when doing compares between the last generation and the new . Basicly tries to normalize the clocks as he should . But You AMD guys always want to change the rules of past practice inorder to overlook short comings. Its been that way now for 6 years and likely the next 6 years if amd manages to stay in the game.

ipc is not the same as clocks... ipc is the performance devided by the clocks. Mosly used in single threaded enviroment. (although the concept ipc is hugely flawed since its dependant on the application, the codepath and the code itself)

But lets assume ipc is a valid expression
--> performance/clock = ipc
this means that
ipc != performance
clock != performance.

If you normalize clocks, you measure something that might by ipc (if the same code path is used that is). However unles that ipc value = 0 it is just as irrelevant as knowing the frequency in terms of performance.

And since bulldozer and derivatives runs at higher clock speeds they should be tested at higher clock speeds. The only reason why tomshw ipc test is relevant is because bulldozer and trinity will run at similar clockspeeds, thus making extrapolation possible wih that data.

 

[Nemesis 1]

Yes that is a valid. I agee testing Intel against AMD @ same clocks is invalid . But AMD Vs. AMD next generation . Clock for clock is a valid means of checking IPC increase clocks just make things happen faster for the most part . Now when you start talking code paths in 2012 . Again we all see that point. As trinity is apu the CL code path the Avx code path. As of right now tho those code paths as of now leave much to be desired . Will trinity APU matter in 3 years . Hell no. Will IVB matter in 3 years hell no . Many like AMDs direction . Thats great . I like intels direction . Once X86 has been all recompiled AVX(old programms ) X86 will be a thing of the past . sooner than ya think. When conroe appeared did AT in his review show P4 last generation against the nextgeneration . Its always been done this way. Back befor Conroe AT didn't compare AMD intel at same clocks . But comparring same brand against same brand on new arch or Shrink the last generation and the present need clocks normalized

 

[pelov]

Laptops, where space and power savings are critical, yes an IGP is useful, but for the desktop, not so much, at least for anything above very low end, non-gaming, and perhaps HTPC uses.

A lot of gaming is still done at resolutions lower than 1080p and I'm talking about desktop.
http://store.steampowered.com/hwsurvey
1080p represents only 25% of their sample with laptop res at 17%. Keep in mind laptops outsell desktops 2:1 as well.

I think with the next generation, Kaveri, AMD will have surpassed that 30FPS+ mark at 1080p with high settings hump. For someone like myself who only does some gaming, a couple hours a week at most, it's a perfect budget solution This isn't an AMD only thing, though. Intel is focusing most of their architectural improvements and performance gains to the on-die GPU as well.

They've both got a lot of ground to make up, Intel on GPU and AMD on the CPU, but they're both headed in the same direction: making a processor that can do it all.