The Ultimate AMD Trinity Notebook List

I'm surprised to see how week gamer is A10-4655M in U38N, of course the not real dual-channel memory can be limiting factor and we do not know what kind of driver he is using. It also seemed like something was holding back his laptop, maybe too many programs in the background... He also did not noticed, a lot of details can be set on high on Trinity without loosing FPS, of course where to set higher if he do not has the minimal 20FPS on low?...

A10-4600M is level with GT630M, so yes than A10-4655M should be level at least with GT620M, but we did not see performance like that.

I was also showing you a grapf, that CPU ain't limiting if you use the benefit of 1080p screen and you give work to 7970M. Not sure why you cannot understand? I actually hate the CPU bottleneck crap talk, since the Pentium 4 ages; I was told by forum members do not put X850xt near that, because CPU will bottleneck rather buy new motherboard and CPU... I did not listen and I put the x850xt, than I put x1950pro, than HD3850 pro and I always enjoyed the higher gaming experiences. I just did not put higher GPU because did not came out higher AGP card than HD3850... So I saved bunch of money on motherboards, CPUs and memories just denying CPU bottleneck .

 

I did the same thing. Tested it, saw the stats, made the purchase, got the results I expected. Was happy with that.

Read later on some hardware review site that what I had bought was pure e and couldn't /possibru/ perform to within 2% of the "true intel" system. Even though it demonstrably did on my system, because the running of the game wasn't cpu-bound in any way.

But that was wrong, according to this tech-site. Because the other platform was "more efficient per Mhz". I swear I'm not making this up - a guy with an education used that phrase.

Actually I saw someone else use the same phrase not too long ago on a different site. The idea being that for each clock cycle, an intel processor does more work than all the rest. When that's.. you know.. technically false. It's the opposite that actually is the case.

Ironically we had the same when Intel launched it's own pentium processor line. These were cisc based (with larger instruction words, reuse of cache for smaller common operations, etc) -- and the "knowledgeable" people, oh, they knew that a "true" 486 system was faster at the same Mhz. Didn't matter what the numbers said.

Also - if for some reason the ever helpful moderators actually scrubbed a post again without telling me, the tests they're using at Anandtech are well-known for being cpu-bound, and they are specifically cpu-bound. Anandtech knows this on beforehand. And they deliberately avoid offering the stats of normal gameplay on high detail, just to make their point.

Of course, no one is going to play these games in benchmark mode. But they imply that you're going to get half the performance out of the amd processor when you play the game on the same graphics card. Which is of course patently false.

Less kind people would say that they are deliberately misleading people. As usual.

 

I too am surprised to see the U38N gaming performance so low. Maybe it was old video drivers? Maybe immature BIOS/UEFI?? With AMD OverDrive the igpu should be able to overclock some like HTWingNut's IdeaPad S405 and with PSCheck it can be undervolted and locked at a certain frequency.

Did anyone see the 3DMark11 P1020 score? That's similar to the P1030 score we have in the OP (page1) for an Samsung A10-4655M.

 

^perfectly good score, no doubt about that. But he's running the test-unit(and said he would not do any tweaking), and as far as I know the standard stock setup doesn't go very high or low on the clock-speeds, and focuses more on "average" performance than peak speeds. If it's anything like the Llano setups, it's possible to go up more than 40% compared to stock, as well as underclock safely down to 300Mhz with much lower volt, with no issues.

And note that the cpu/combined score in the 3dmark test draws down the total score compared to an intel test, specially when the processor isn't being pushed. I.e., the graphics card score is more than twice as high when compared to an intel hd4000. It's a huge difference.

Have a bit of a problem with the idea that the ulw processors from intel draw less power as well. It's not actually the case when you include the graphics card. The apu system also draws less power on moderate loads, and idle. As well as peak loads, obviously. So I don't know... outside the 17tdw number -- questionable how true that one is.

 

From the gaming review he showed it seems that there is a lot of driver or BIOS problems, with the games not occupying the full window for example.
Really weird. This is a pre-production unit anyway so maybe we should wait for more reviews. It sucks though.

 

I would vote we move all the gaming argument out of this thread. it will end as they all do.

that said, what IS pertinent to the point of this thread:
I'm going from memory of every page in this thread, but it seems that at this point we don't have any proof whatsoever that a Trinity-based-anything will, bang-for-buck, outperform or even match a comparable IVY+Radeon-GPU rig. So whatever the rationale is for nice laptops based on Trinity, it ain't gaming value.

 

I really hope this is the case. Moreover we have no clue about pricing... Asus or AMD better fix this, because just yesterday there was a good offer for the UX32VD for 1080 USD (i7 version with IPS panel). Competition has become serious and of course between the two - fir the same or very similar price- I would get the i7 and the GT620 anyday (with this gaming performance at least...). Disappointment, no doubt, but let's see if a driver or BIOS update fixes this...

 

I did this already with Llano and there's no reason to believe it would be any different with Trinity: http://forum.notebookreview.com/gam...amd-radeon-hd-6750m-benchmarking-results.html

Look at the green bar (2x4GB @ CAS 9 1600), red bar (2x2GB @ CAS 7 1066), and dark blue bar (1x4GB @ CAS 9 1600). Pretty much ZERO difference between them for any game or benchmark when using the dedicated card, like within 1%. I could repeat this with any CPU and results would be the same. I did a handful of test with my Ivy Bridge and results were the same.

 

^so, did you run the benchmarks with the same bus-speeds? That's what is significant. Not the specs on the ram itself. And you're not going to deny that you can pull the bus-speeds higher on a 1333Mhz ram setup compared to a 1066Mhz setup, or that the performance while in crossfire increases when you do that. There's any amount of tests and benchmarks you can find that proves that. Whether it is tomshardware, or any random overclocking guide.

Again, I'm sincerely having difficulties believing you didn't know that before you ran the tests.

Translation:
.."let's not discuss what I'm now going to argue in my own post, while establishing something that flies in the face of evidence I just read less than three posts over mine. And which there really was no way to turn around even with the most impossibly tortured logic. So just let me end the discussion by declaring something flatly false - which is now true because I say so. And if you disagree I'm going to report your posts to the moderators to get the thread locked!".

Seriously, though -- come up with something that can be tested. Create some sort of reasoning. Don't just flatly deny evidence and say the opposite is true. Whether you work for Intel and are astroturfing or not (and I don't believe you do) - you're not going to convince anyone that way - so why try? Why spend time coming up with reasoning that anyone can pick apart, and anyone will see instantly is false, regardless of motivation? It's the missionary equivalent of "I believe in SANDALS WITH SPIKES THROUGH THE SOLES THAT MAKE IT IMPOSSIBLE TO WALK! AND SO SHOULD YOU! HEY, WAIT UP! DON'T RUN AWAY!".

Not compelling.

So "I believe that 1+1 is 3" isn't an argument for anything other than that you believe 1+1 is 3. So why insist it is? You could say: "Well, from what I see, which is such and such - 1+1 /could/ be 3, /if/ 1 was worth something else than 1!". That's fine. But you're not going to say: "Well, I believe 1+1 is 3, so let's not discuss it! And if you contradict me you're a jerk!".

Doesn't work like that, when we're looking at data where 1 and 3 already have reasonably close and accurate values.

 

No, not same bus speeds, but when you run 2x4GB 1600 CAS 9, 1x4GB 1600 CAS 9 and then 2x2GB 1066 CAS 7 and the results are the same, it's pretty compelling. The BUS speed is far from the limiting factor when using the dedicated GPU. If you can run at 1066 and 1600 and have same results.

Here's the other thread with the IGP that I showed images of the RAM and their respective CPU-Z screenshots: http://forum.notebookreview.com/gam...g-benchmarked-various-ram-configurations.html

 

I believe it's these problems because there are a lot of evidence from people with A10-4655M sleekbooks that have better gaming benchmark numbers than that.
So lets cross our fingers for Asus not sending more pre-production units for reviews without correcting all the kinks.

 

..of course the bus-speed and ram timing would affect the apu runs and the crossfire runs.

Look.. Time out. Lecture time. This may take ... 45 mins. Be seated.

The reason why an overclock on a traditionally set up dedicated graphics card works in the way it does depends on three things.
-The bus interface and speed to the graphics card buffer.
-The graphics card's processing power and simd unit execution.
-The graphics card's ram interface to the simd units/chipset.

In other words, the graphics card sits on the bus, transfers data back and forth. The amount of data that can be transmitted usually is higher than what the simd units can do any processing on. But resubmits of entire scenes tend to take a very long time. Therefore this bus-speed can be a bottleneck in games, since typically parts of the operations need to be submitted and run on cpu logic before they are reduced again on the gpu. So if the bus transport is slow, this can be a bottleneck.

That part is especially interesting when looking at crossfire, because the two graphics card units typically share resources.

Next it's the simd units on the graphics cards. These will perform various operations on the memory blocks and place completed work units to be transferred through the bus. It can be as simple as to one front-buffer that's just displayed via the graphics card logic, or that buffer might be transferred elsewhere.

When you raise the core speed the graphics card operates on, the fixed number of arithmetic units or the simd units can make exactly as many operations per cycle as before. But since there are many work units and the operations are assigned internally, raising the frequency of the operations increases the amount of work units that can be started as long as there are free memory units.

Next it's the ram. Works really in exactly the same way as main memory, except the simd units on a graphics card can perform parallel tasks. If the ram speed is increased, the operations assigned can possibly be completed sooner.

But. If you raise one variable without raising the other, the benefits aren't that great. So if you for example raise the core speed but lower the ram-speed, the increase might not turn up at all. In fact, the actual work being done on the chip could very well be the same as before.

In the same way, if the bus is a bottleneck, completing work-units on the graphics card sooner will be a complete waste.

Or, if you were able to raise the bus-speed, you can still not force the core and the memory on the graphics card to complete operations sooner.

-----

So what do we have on an apu setup? We have the graphics card connected to the memory bus in the same way as the cpu (or at least that's the plan somewhere up the line - at the moment there's a "turbo pipeline" of some sort to facilitate reads and writes). This means the transfer speeds between gpu operations (as in simd type operations in parallel - "shader logic") to memory and over to cpu operations in memory (as in complex mathematical functions that shader logic isn't complex enough to complete quickly) is extremely high. Massively higher than on any dedicated card attached to the pci bus.

So there's no bottleneck here. But the speed of operations between main ram and gpu(and cpu) directly depend on the bus speed and the speed of the memory controller (the difference between 800Mhz X 2 and 400Mhz X 4 is large for smaller operations, that typically is what the gpu will run). And there is a bottleneck if the card is running in crossfire, since assigned single work units might depend on the execution on the other "device".

In other words the bus speed is significant when running either apu graphics or apu/crossfire graphics.

But it's not as significant when running only the dedicated card, specially in practical tests. Though stress-tests that are cpu and bandwidth bound, will very likely take a hit because of too little processor power, and probably to a lesser degree the bus speed and multipler. This is something you notice on nvidia card tweaking as well - in many benchmarks you can underclock the processor and get exactly the same gpu results, simply because running the game doesn't depend on preparation of work-units in the program-logic running in main ram, or on the bandwidth between the ram and the graphics card via the pci-bus. (Until you run "fast forward" benchmarks - because these rely on how fast the processor can prepare game-data -- this is how Haven benchmarks run, or the Crysis mark runs, among others).

-------

So what we have in practice here is that there is going to be potentially a bottleneck, but mostly a possible performance gain involved with the bus-speed when running apu graphics and crossfire/dual graphics. If you look in PScheck, you can raise the multipler without raising the total clock-speed, for example. Or you increase the bus and lower the multipler to get higher timing. This isn't there by default - they choose "safe" timing, even on "boost" states. So for peak performance, that's going to have an impact. The question so far is if the cores are as overclock safe as they were on Llano. Not if the performance increases or not when you raise the bus-speed. It clearly does.

So this is the equivalent of the core speed in the overclock example above.

The memory timing is significant as well, because the "graphics card" operations are done on RAM. And these are dynamically set, if you look in PSCheck, for clock speed on the core of the cpu/gpu part with the same variable - the bus-speed x the multipler.

The dedicated card has it's own timing variables, and you can reach those if the dedicated card has been enabled, is running, and is present on the system.

---------

A couple of other things as well. What this means is that the actual ram timing is ignored. Was pretty interesting. That you can pick the different busXmultipler states and push the ram speed up and down. This is why it's so significant for the overclock, and also the cpu speed.. But also why perhaps Trinity has lower overclock potential - that the cpu is closer to the limit even if the gpu part has more to go on.

And we just don't know if latency timing is significant for running the simd operations. It might be. It seems from the data I had when I had the Llano laptop that OpenCL operations (that will shift contexts often) might run faster if the latency timings are quicker. I don't think it has anything to do with the actual graphics performance of the cores, though, because they are so much slower than the ram anyway. It's possible you might be able to pick some optimized timings and get a good raise in things like 3dmark11, because it depends on other factors as well. ..of course, we're not ever going to do that because Asus of course locks their bios, and all that, just like everyone else.

And it's also very much possible that if you don't follow up an apu overclock with an overclock of the dedicated card in a crossfire setup - that the stress-tests specially are going to suffer. Because, like explained, the two "graphics devices" share work-areas, and will likely wait on the slowest one.

-------

Ok? Was just a couple of things. But sort of think we should have at least some picture of how these things hang together before reading benchmark data....

 

didn't read LOL

 

All this "debate" is so pointless. I don't know how much piledriver benefits from CAS timings or speed, but it shouldn't be compared to ivy bridge.
The memory controllers on piledriver and ivy bridge are on completely different levels and not to mention that AMD cores in general are more bandwidth hungry than Intel cores. So there might be slight difference with higher speed RAM, but until AMD's memory controller is improved I don't see any significant difference for Trinity on the CPU side.

 

Your 6750 thread is not relevant to the latency because you didnt vary it at the same frequency and didnt go over 9, but there you go, your 6620 thread has some validity. And you note some increase using CL9 vs CL11, but youre only doing graphic benches in there. And thats with LLano, Trinity is different, so your results arent even completely comparable. In your own thread you mentioned others got improvement using CL8 1600 over CL11, and Meaker posted that AMD themselves said it was latency sensitive.

Go back to GX60 thread, actual user recommends 8GB with tighest timings over 16GB of RAM for best performance. That is based on his use and benches of Trinity using CL11 vs CL8. Its worth revisiting.

 

Posts like this are the real problem, The GX60 puts several Ivy + Radeon/Nvidia GPU notebooks to shame, at a lower price. That may be bacause it uses a better GPU, but at a lower price point who cares. And as has been shown even an i7 gets GPU limited, so that the A10 peforms the same when settings are turned up high.

 

epic!!
this could be post of the year, except its a little tangental to the thread... Still an awesome flyby

 

I would like to appeal to our unofficial Potentate davidricardo, for a summary shopping list, since it is that time of year ... at least for some of us

I suggest two categories only: "High Perf", and "Budget Bang" ... understood that the latter is a matter of one's own definition.
High-P
At this point, what is the known shortlist of candidate purchases with these qualifications?:
- Trinity A8-xxxx or higher [is any regular attender of this meeting seriously interested in A6/A4 ??? not I]
- minimum 1600X900 high quality display panel
- Under 5 lbs? or is that too restrictive?
- Min RAM capacity 8GB Dual

BB:
- Best candidate under $400 USD?
- Best candidate under $500 USD?
- Best "highly portable" sub-4lb or sub-2kg

??
I guess to be global we'd need this by region... or at the very least, Eurozone, Americas

 

Wall of text...

Thank you Mr. Wikipedia.

You bring up CrossfireX and APU and that's not what I was referring to AT ALL. I was referring to the fact that with the dedicated card the timings of the RAM do not make a lick of difference with gaming with a dedicated card, period. Nothing more.

And EXCUSE ME for not using the RAM and timings you think I should. I used the RAM that was available to me, and the timings they offered. I had no control over their timings. I used MY machine and MY RAM and MY TOOLS available to me. Feel free to offer supporting evidence using your own resources.

All I hear from you and TommyB0y and a couple others offer a lot of lip service with zero support from your own trials or even anything from a reliable source.

Your input has completely taken anything that was originally discussed OFF COURSE... DERAILED the original topic of discussion. So please kindly take the time to read and understand the argument before you go off on a tangent about OPENCL which has nothing to do with gaming and about generic operation of how PC's work. You guys seem to think Trinity is working COMPETELY different from the any x86/x64 CPU has operated in history. That it somehow is magically affected by things that have not affected any other CPU. That CAS latency from 11 to 9 or 7 even is going to improve performance by some significant amount. It won't. It can't.

Read same above... well here you go:
And EXCUSE ME for not using the RAM and timings you think I should. I used the RAM that was available to me, and the timings they offered. I had no control over their timings. I used MY machine and MY RAM and MY TOOLS available to me. Feel free to offer supporting evidence using your own resources.

And I trust Meaker with his results. He backs up his statements with real data and sources. However, he said it "would perform better" but there's nothing more to substantiate what "better" is or what is actually better.

 

aside that a 2600ish to 2900ish score on physics aint nothing to blab about

atom ant, you provide me with a bench of only one game, and calls the matter settled? really? the results that I posted are meaningless? And interestingly enough the same fps that the bench showed is the same fps of a 7970m suffering underutilization.

 

I dont care what tests you do with your stuff. I was just pointing out that what you thought relevant wasnt at all, becuase you just dont get it. Thats fine, no need to discuss it. I will just believe people that actually use this stuff and are happy with the performance. And maybe I did not mention it, but AMD engineer layed out all the timing, and the theoretical math on paper says CL11 will cause processing delays. What happens on paper and in real life can be different, but I have seen more than one bench improved with tighter timings, some not.

 

AMD has admitted to Enduro issues in driver fixes, so yeah, some of the GPU limiting is from drivers, so once its all hashed out might get closer to CPU limiting.

 

thing is you guys specially atom ant, dont recognize that there are enduro issues with this machine. I dont believe it has it as well. Again your benchie for saying to me that there isnt a bottleneck has a bottleneck and it really aint enduro.

what AMD engineer said that CL11 gives performance problems?

 

Well, both i7 with 680M, no enduro issues, and A10 with HD7970M get the same FPS. I dont know why you cannot see that the CPU is not the bottleneck, unless you think the A10 can hang with an i7. The 680 and 7970 are close enough to do that common sense math as you guys like to do.

 

really? havent you saw the results of the sli 680m in that game? they are just a little bit better. Again this is a AMD favored game, heavily one might I add, however its still a fact that the i7+7970m+enduro issues has the same results as a10+7970m, thus still a bottleneck.

Until you get one and compare the results of that machine against a i7+7970m unit without enduro issues you arent going to be able to see the cpu bottleneck, because so far the hide in hole exercise has been pretty well daunting.

 

So we wont really know until you compare the right two things? Sounds like another excuse, but I can agree with that! Now apply that same logic elsewhere and not make statements based on old testing on other platforms.

 

MSI GX60 owners are not complaining about Enduro issue since Catalyst driver 12.11, neither about CPU bottleneck problems .

 

And this is a surprise when the graphics card doesn't use the system's ram... how?

Right. So I pointed out you are probably testing with identical bus/ram speeds, while the bios locks the latency timing. edit: and if you're testing an external card that doesn't actually use the system ram in the first place, then..

No magic involved. What we wanted to know about was the 3d performance of a "graphics card" that's fused to the laptop's memory controller, uses the system ram for operations, and is placed on the cpu die. So the system ram speed is important.

And the reason why I'm saying it's /likely/ that latency timing affects performance, is that you have more frequent context shifts in apu workloads than when comparing it to a graphics card. Like I said, on a graphics card, low latency couldn't matter less, because the simd units are comparatively slow, they execute work units that take several cycles.

But on an apu that might not be how it works, because the work units will be scheduled on the memory controller like any other operation.

So yes, it's different, and yes, we don't really know. I'm just explaining why it's not out of the realm of probability to look for that as a factor in the 3d performance.. and stuff.

 

really you have trouble seeing things and understanding whats being said?

there is a cpu bottleneck. all your benchies show that. I compared all results except 2 with a 7970m without enduro issues, which meant all of those were done on fairly old drivers with a aw m17x or m18x, except as mentioned on those 2 using clevos, which also had fairly old drivers to boot.


so there is a cpu bottleneck them.

I dont understand how can that be more clear than game benchies, not 3dmark crap or unique engine, its games, where the performance should be since that is a gaming laptop. No number seeking just plain old fps.

and that shows on every single benchie that has been shown here or compared.

I havent used a single number from a system using 680m, just 7970m. Im really going to ask someone to do a bench with a 7970m + i7 on sleeping dogs just to satify my curiosity on how much bottleneck that apu gives, so we can bury your last standing number that suddenly disproves all the others provided here, because somewhat in your heads comparing i7+7970m to a10+7970m using the same games at the same settings is unfair to the a10 because of the ram used, because the cat pooed in the garden, because the bird flew over the cuckoos nest.

 

Why don't you guys take all this back and forth discussion to the GX60 thread? It has hardly anything to do with Trinity notebooks anymore and you guys have been involved in circular arguments for a looong time.

 

No, the "problem" was very specific. You can repeat the experiment with an i7 with "boost" and an i3 without boost as well.

On normally running the game, below 60-70fps or so. (In Crysis warhead, the fps-limit is higher as well).. When you run the game normally and measure for example how close the game is to a v-sync at 60fps -- then you get similar performance results. On an i7 and an i3, the result here is within 3-10%. Both systems keep the v-sync, and you can pile on detail and AA and maintain the same results.

It's easy to explain why that is. It's because then the game is gpu-bound. The cpu can serve, with large margins, more than enough for the gpu to handle.

When you do the "fast forward" benchmarks, or the stress-tests that produce as many frames as possible. Such as the Heaven benchmark and the Crysis benchmark that anandtech likes to use. Same with the Civ 5 run and the Dirt run. Then the results are the other way around. Now the i7 (at 3.2-3.3Ghz for example) scores at least twice as high as the i3 (on 1.8-2.1Ghz).

And it's again very easy to explain why that is. It's because now the benchmark is cpu-bound. The game-engine's logic can't create frames fast enough to max out the gpu at all times. If you actually looked carefully at the results, you would see that the score is inflated by running extremely fast in the low gpu density parts as well. The parts of the run where the average fps would be very high.

So. When looking at that experiment, you always will get the following: i3/low Mhz, no boost, small cache -- is in 99% of the time more than sufficient to serve a 60fps game. It's still extremely rare to see a game that's actually cpu bound. But you would still recommend that people pick the processor with the higher clock-speed (i5/i7) to raise potential lower fps-spikes, or to lessen loading times, or gaps with badly programmed node-generation, that sort of thing.

Meanwhile, what does the cpu-bound benchmark tell us when it comes to gpu utilisation? Back in the long-long ago, we used those stress-benchmarks to get an indication of whether the gpu was at all capable of getting a particular level detail to run smoothly over 30fps.

See where it comes from now? How this hangs together?

Basically, the A10 will "max out" at, you know... 300fps in some benchmarks, while going for a lower fps-limit of 150. The i7 will have the same lower limit, where the test dips down - if it does that at all anymore - where the test is gpu bound. While it's going to maybe have a higher limit at 400000fps, or whatever, where the gpu can produce that many frames when the test only displays the wireframe of a model, or during some sprite being rolled across the screen.

But does that mean you get better "gaming performance" on the i7? Of course not. Like Atom Ant says, this stuff is old. As old as the first 3d cards. Except no one uses those "fast forward" benchmarks any more to see if there is a lower fps-dip that has to be fixed or smoothed over. Because you don't have to. Because the lower fps-limit tends to be over 30fps anyway, or else it drops down because of some game-logic fault, for example. So now those benchmarks are only used to get a huge total score number to show off. Which is completely irrelevant to anything. Or at least it used to be.

Basically it also was irrelevant when those tests were used in for example quake2 to do the run-through of the first level. Because what mattered was the lower fps-limit, because that indicated if your 3d card at all was capable of running the game without framedrops.

..

..so that's part 2 of the lecture, I guess.. And now you know why the Anandtech benchmarks are so completely disingenuous.

 

No it doesn't. I posted the CPU-z screen shots that I linked to. The timings were different. You obviously did not read my post or point, again. It was all about whether dedicated card is affected by latency of the RAM. Bottom line is NO. Now you're talking about APU and integrated GPU which yes of course it's affected. Minorly, but yes it is, I agree 1000%, but that wasn't my point. The point was brought up that the APU is sensitive to RAM latency even with dedicated card (by TommyB0y). I just refuted that argument saying that it wasn't

You're right. I was a bit harsh, and it is taking this thread too off topic from its original intent. A lot of it is people only listening to themselves though, and not reading what other people are saying. I'll move over to the GX60 thread. Maybe we can get some definitive testing from someone with a GX60 (may be me soon )

 

I dont respond much of your posts anymore because, lets be fair, you either post information with mistakes that suit your points, like in your previous post, and usually with a tangent that is actually not relevant.

For example here we have cpu lowers level loading times, they dont. Unless you are dealing with some seriously low level cpu that aint really sold anymore to consumers that aint a factor. and if you get that your system will be so low performing by today standards that it wont matter. Why? because either the levels are loaded by the cpu, or from the HDD, most of the times both happen. The cpu would just do the coordinating.

The cpu is relevant due to the lowest fps possible, not because on a spike it can reach 300fps, I dont like avg fps, but that is usually whats provided, a graph with a normal distribution would be much more interesting. However thats also problematic since games vary the load that they put on the system, depending on how many objects it needs to render, and that varies fast.

I really cant see anything relevant or enlightning that you wrote above or the other post.

Fast forward benchmarks? picking one game that is cpu bound to devoid that games that are cpu bound are so meaningless that you can say that they dont matter or that because of that they cant be used to measure gaming performance? Discard in built game benchmarks because?? well you didnt provide a reason for that. Since civ 5 has a gpu bound benchie.

Also if you see serious gamers they arent concerned with avg fps, they are concerned about the fluidity, which means if the fps are going to dip lower than what they establish as good enough performance for them, some settle that for 30, some 60 and some are going for 120

So I cant really see anything bad about how anandtech does anything that makes them so unreliable so discredited.

you are not probably aware how that makes you a **** in the eyes of people that own the system or some other stuff, its always best to talk about the model in another thread not the owners lounge, but I agree lets stop with this, its getting really tiresome.

 

No, we know for certain the ram latency and bus-speed has no impact when the dedicated card does not use the system ram to store data.

But when the dedicated card is used in crossfire, or you have the modules that don't have dedicated ram on the card itself - then the total score might be dependent on it. On my Llano rig, that's definitely the case. Upping the bus-speeds raised the graphics score almost to the double, even when not raising the total clock-speed. In other words - when a 7670m/apu combo uses system ram (instead of dedicated ram on the graphics card), then bus-speed and very likely ram-timing affects the score.

And.. we've also seen already that ram in dual-channel increases the performance much more than you would expect from increased bandwidth total. I.e., response time is more important than just bandwidth. So there's a lot to suggest ram-timing is significant for the performance on a combo like that.

But I pointed out two problems with your test. 1. That they might not have been testing graphics cards that use system ram in the first place - which we must be doing for the test to show something that has any connection to the system ram's latency timing. And 2. that bus-speed is controlled by the p-states, and might have been put to the same speeds on all the tests.

And if we're then not actually testing something that is gpu-bound anyway... what does it tell us? ...Nothing.

..yeah.. that was obviously central to the entire treatise, of course. But no, it might be dependent on that. Rebuilding textures and models, pre-generation, node generation for AI, creation of static lookup tables, streaming assets - this both happens at the beginning of a level, and during pre-loads in the levels. So if you have cpu-cycles to spare, reducing the run-time of these things can make things snappier and levels load faster.

This is something that directly translates into increased scores on benchmark runs, since it shaves off pauses that can be fairly long - that of course wouldn't actually have been a problem if the game ran at 60fps instead of 180.

Right. The lower fps has to be above that for as much of the run-time as possible. Being able to produce more cheap frames at higher speeds -- doesn't help with that.

..Because they imply that amd-processors will give you half the 3d performance on a dedicated card? They say, directly, that you should pick "something else" for gaming. They don't suggest something in a more narrow context, or single out some "enthusiast benchmark society" score, or something like that. They specifically say that unless you have the highest clocked intel processor, then your gaming performance will suffer. Down to, according to objective benchmarks - half of the actual "potential" of the 3d card.

Which is completely bonk. As you also point out - the point is the fluidity at 60 or 120fps. That's what we're looking for.

And as you can see - anandtech's test is not making some point that their intel test, for example, was able to run one of the tests with a minimum fps-limit at 121fps, while the amd rig only reached 110. If they did that, I'd have no problem with it. It wouldn't be relevant to notebook gaming and igp graphics, and everyone would see that.

But, like I said - they're not doing that. They're laying out a case for how games are gpu-bound at any framerate, and that amd's apu offerings don't match up when it comes to 3d performance in general.

That's also what several people in this thread said, and used that specific page to "prove". But of course it doesn't prove that at all.

 

Here's some good news:

Playing Starcraft II on ASUS U38N: gameplay performance on various settings and resolutions - YouTube

The performance in Starcraft is rather excellent with playable frame rates at FullHD AND high setting, alhtough low shader details. This is quite impressive from such a device.

I think we can hope there will be soon a driver update that will also fix the issues in the other games.

Personally, I'm looking forward to the notebookcheck review of the device.

Let's see how this goes...
Asus... Please price this guy lower than the UX32VD

P.S. Don't you guys think the color of the U38N chasis looks better than the plain metallic of the other U series?

 

The hilarious thing is that you think we are stubborn, but you have the blinders on, there is no getting through to you

 

while I dont know why I keep posting, obvisouly because Im a stubborn ******, you guys cant see that there is a bottleneck to the cpu.

I really wont give anything to a reviewer that gets loading times in a review, be it before or even during mid level loading, its simple to skew those things, put a SSD. Benchies dont account that be it game or synth. So it doesnt matter.

the point of fluidity still stands. When the cpu cant feed enough data thus making the gpu not perform as it can this is called a bottleneck. Its simple, you dont dip the fps beneath what you established as the base minimum. When the fps varies from 35-60 its pretty noticeable.

the apu doesnt provide enough power for a gpu from 2010. Why would it matter? its not something that can be recommended for gaming. Its simple as that. And given that the 7850 is not something less powerful than the 5870, the same is true for the a10.

 

I don't mind the side conversations going on here so as long as it pertains to or relates to AMD CPUs/GPUs and related hardware (preferably Trinity). Some of the things you guys are talking about is out of my knowledge base and at the same time I'm learning something new. I just don't want the thread to derail too much (or more than it has to).

I will try and put something like this together but you'll have to give me a little bit of time. This week's been busy with the Thanksgiving, my birthday and other such family events. This is a good idea.

I agree, that was pretty impressive for 1080p resolution and high details. Let's see an Intel-based Ultrabook (with only integrated gpu) pull this off.

I do have to say I like the exterior design, that lovely "frameless" IPS display and the color. Its certainly a very attractive ultraportable.

The U38 is available here for 800 Euros, I wonder if anyone's bought one yet?
U38N C4010H
U38DT-R3001H

What is the price in Euros for the UX32VD?


Here's a Chinese review of the Lenovo S405 with 7450M, looks like ULV Trinity DOES SUPPORT Dual Graphics. I've been told it doesn't but the photos in this review indicate otherwise.
http://www.beareyes.com.cn/2/lib/201211/22/20121122096_1.htm

 

Imagine that, some people thought dual graphics on Trinity impossible, suprising what it can do that people thought otherwise.

 

Happy birthday davidricardo.

That's a really great price for that notebook. the Asus UX32VD is 200€ more expensive.

 

We know it can because desktop trinity can, so if someone wants to create a machine with that they could.

Also lower latency ram has helped my results a bit, not a huge amount, with my GX60.

With this the CPU is certainly a bottleneck, an unlocked or higher rated model would help very much. When paired with an SSD loading times are noticeably slower compared to my old I7-3740qm @ 3.9ghz but that's a no brainer I think.

 

davidricardo if its your birthday than congratulations and happy birthday!

Your posts are great so looking forward to reading more of them in the future.

 

Guys I ordered the ASUS N56DP-DH11 from XOTICPC everything stock except I added 6X Blu-Ray Writer/DVD burner. Also bought Samsung 840 Pro from Newegg, and it comes with a free copy of Assassin's Creed III (good deal for $269 +tax + no shipping) so looking forward to running benchmarks on my first laptop!

So far had two PCs. 1st was an Athlon 64 (man that was fast back at the time) in Armenia, current one is Athlon X2 4400+ GeForce 6150SE (crappy integrated nVidia graphics on nForce 430 chipset) which is really old now and hasn't even been upgraded (I'm s surprised is still works) other than adding .5GB of ram to make it 2.5GB and runs Vista, its the first computer I owned since I came to US but I guess I can't complain since It was a birthday gift. So if there are no review's of the laptop I can ether create a new thread or can just post things in here.

 

The UX32VD is starting at 929 euros for the 768p model in France, but it's a limited offer, it's mostly around 1000 euros+++, with a 24GB ssd but no backlit keyboard.

The U38 seems really nice priced indeed.

Love this lap. It looks great.

 

On the first one - the question was not, "can mobile Trinity do CrossfireX?" but centered specifically on the ULV APUs, which were speculated to have had that capability removed.

Your personal experiences with the GX60 are a big boon to this thread - one of the first really relevant things that's been said in a few days now.

 

It looks awesome, nothing comes close to the N series, and the design is very unique. Should get the laptop in about a week so need to free up some time to dedicate to benchmarks. The 840 Pro 256GB should arrive in a couple of days, but I won't have a way of testing it until I get the laptop.

 

While the CPU may be slower at opening things, the big questions have been how much does it limit FPS in games.

Your 3Dmark physics score imporved significantly, would you contribute that to the RAM timings? Starting around 2500 and ending up over 2900 on that metric. Wonder how that translates to improved FPS.

 

A mini review would be awesome. That notebook has a dedicated GPU with no Crossfire with the iGPU of the Trinity APU so gaming benchmarks won't show so much the superiority of the APUs in that aspect. But you could give some benchmarks and overall impression of the notebook. How much will you pay for it?.

 

I already did, waiting for it to arrive. I don't expecting it to take less than a week since I added a Blu-Ray writer on it which the laptop does not support by default, so they have to keep the bay cover and replace the drive only. I ended up paying $938 + No TAX for the laptop and the optical drive upgrade.

 

AMD Radeon HD 7970M video card benchmark result - Intel Core i7-3720QM Processor,Alienware M17xR4 score: P7609 3DMarks

here is a score of a m17x r4 with a 3720qm and a oced 7970m, very similar level to what meaker used. Look at the physics score

btw the 3720qm is also oced. otherwise the normal would be around 7400

http://www.3dmark.com/3dm11/4726972

this guy cant go past 950mhz on the gpu core otherwise the gpu will give artifacts.

Im still looking for that bench of sleeping dogs

but I hopelessly thought that what meaker said should suffice.