AMD Benchmarks Zacate APU, 2x Faster GPU Performance than Core i5

[kl0012]

Btw:



Means that Zacate gpu perf is something like 790GX-890GX.

[informal]

You guys are giving AMD way too much slack. It should have been obvious that something was a miss with the initial benchmark used and AMD was obviously trying to pawn off their performance or inflate it higher than it actually is. I don't think AMD is incompetent enough that they couldn't get simple driver things resolved on a test platform for a comparison.

Some of you guys piss on review websites that don't uses updated drivers in reviews. This is even worse than that because this info(driver dates) are not readily available where in most reviews, they tell you exactly what drivers they used.

Basically anandtech caught them cheating and they had to back up on their words and give anand freedom to do what he wanted.

If they didn't, Anand would have published that AMD was cheating on his website and that would have been bad press, even with AMD users.

Read the article update at AT.They(AMD) used the drivers via the the manufacturer's website.Newest intel graphics drivers can't be automatically installed due to a known issue.It was just a set of circumstances ,not cheating.

[tajoh111]

Read the article update at AT.They(AMD) used the drivers via the the manufacturer's website.Newest intel graphics drivers can't be automatically installed due to a known issue.It was just a set of circumstances ,not cheating.

They should have known that from the beginning from the performance and corrected it, as Anand saw it. They just happened to have convenient driver issues and didn't see it until Anand saw it? What the point of running a broken platform?

For the guys at AMD setting up this test, its like sending a kids town with a knife in it. They should have caught the error or in this case, they let the error slide to inflate numbers.

They should have seen the problem earlier and fixed it. These guys are not idiots. If they didn't get the problem fixed before hand, they should have used a different platform or laptop.

If Nvidia didn't this, you would be burning them at the stake and not blame driver issues.

[flippin_waffles]

And it was a publicly available part. Just highlights the uphill battle intel is going to have with driver issues. That 20 years of developing a high performance, compatable, feature rich and industry standard up to date software stack that ATi and now AMD has shouldn't be taken lightly. Intel's graphics drivers are notoriously woeful.

[kl0012]

And it was a publicly available part. Just highlights the uphill battle intel is going to have with driver issues. That 20 years of developing a high performance, compatable, feature rich and industry standard up to date software stack that ATi and now AMD has shouldn't be taken lightly. Intel's graphics drivers are notoriously woeful.

Intel's graphic drivers for the laste gen gpu are more then OK. I have not experienced any problem with them in my HTPC. Last time I updated gpu drivers was more then half year ago.

[informal]

They should have known that from the beginning from the performance and corrected it, as Anand saw it. They just happened to have convenient driver issues and didn't see it until Anand saw it? What the point of running a broken platform?

For the guys at AMD setting up this test, its like sending a kids town with a knife in it. They should have caught the error or in this case, they let the error slide to inflate numbers.

They should have seen the problem earlier and fixed it. These guys are not idiots. If they didn't get the problem fixed before hand, they should have used a different platform or laptop.

If Nvidia didn't this, you would be burning them at the stake and not blame driver issues.

As can be seen,game performance was minimally impacted(practically the same CoH numbers with the new driver).And it's actually intel's fault since it's their platform,their drivers,their responsibility.Users shouldn't have to do manual tricks and acrobatics in order to just get latest graphics drivers running on their laptops.

[ajaidev]

Btw:

[IMG]http://images.anandtech.com/graphs/5450_020310223624/21599.png[/MG]

Means that Zacate gpu perf is something like 790GX-890GX.

Hopefully not because HD4290 runs at 700mhz and 40 shaders where as the Zacate is suppose to have a 500mhz part with 80 shaders. If a 80 shader @ 500mhz is equal to 40 shader @ 700mhz it points to a lack of optimization , driver issues among others.

The 5450 is suppose to have 650mhz core clock and 104 GFLOPs theoretical performance. Now that N Body simulation on the Zacate gave 23Gflops can anyone pls test a 5450's N body bench and report the Gflop they get, i dont think it will be close to the 104 mark but most likely be in the lower double digit mark.

[-Boris-]

They should have known that from the beginning from the performance and corrected it, as Anand saw it. They just happened to have convenient driver issues and didn't see it until Anand saw it? What the point of running a broken platform?

For the guys at AMD setting up this test, its like sending a kids town with a knife in it. They should have caught the error or in this case, they let the error slide to inflate numbers.

They should have seen the problem earlier and fixed it. These guys are not idiots. If they didn't get the problem fixed before hand, they should have used a different platform or laptop.

If Nvidia didn't this, you would be burning them at the stake and not blame driver issues.

We don't even know if this benchmark was planned a month ago when they set up the systems, if it wasn't, there was no issue to detect.
And letting Anandtech play around for themselves is a good sign that AMD didn't intended to rig the test.

[Monkeywoman]

A 10.1inch display IPS with aluminum chassis...dirrty

[Lightman]

Btw:


Means that Zacate gpu perf is something like 790GX-890GX.

I expect it to be 10-25% above 890GX with sideport memory.
Obviously there will be cases where it can be 2x as fast or 50% slower than 890GX. Still for 18W TDP SoC this is great.

[informal]

Btw:



Means that Zacate gpu perf is something like 790GX-890GX.

10x7 resolution is impacted more by CPU clocks.Do we know on what platform were these tests done? If it was core2/phenom with ddr3 than these results are not directly comparable.

[flippin_waffles]

I wonder if Catalyst will be available on Zacate also, or just Llano. I doubt AMD would provide overclockability for Zacate, but it would be pretty cool.

[LightSpeed]

They should have known that from the beginning from the performance and corrected it, as Anand saw it. They just happened to have convenient driver issues and didn't see it until Anand saw it? What the point of running a broken platform?

For the guys at AMD setting up this test, its like sending a kids town with a knife in it. They should have caught the error or in this case, they let the error slide to inflate numbers.

They should have seen the problem earlier and fixed it. These guys are not idiots. If they didn't get the problem fixed before hand, they should have used a different platform or laptop.

If Nvidia didn't this, you would be burning them at the stake and not blame driver issues.

Do you really think they knew about it? Isnt it obvious that the 10x difference will make people want to verify it themselves? Why would AMD do that to themselves knowing that they will get caught? I would understand if the difference was 20 - 30 % across the board, but it was an isolated bench with a massive difference due to drivers

Its an honest mistake and the reason was plausible. Even if Nv did this, it would be understandable if the reason given was the same.

[MrMojoZ]

I bet even after this you are still going to see "Waaah AMD lied!!!" posts for the next few months.

You guys are giving AMD way too much slack. It should have been obvious that something was a miss with the initial benchmark used and AMD was obviously trying to pawn off their performance or inflate it higher than it actually is. I don't think AMD is incompetent enough that they couldn't get simple driver things resolved on a test platform for a comparison.

If they didn't, Anand would have published that AMD was cheating on his website and that would have been bad press, even with AMD users.

Haha, I totally called it. I knew someone would be ranting about this. Expect to see more.

[Florinmocanu]

They should have known that from the beginning from the performance and corrected it, as Anand saw it. They just happened to have convenient driver issues and didn't see it until Anand saw it? What the point of running a broken platform?

For the guys at AMD setting up this test, its like sending a kids town with a knife in it. They should have caught the error or in this case, they let the error slide to inflate numbers.

They should have seen the problem earlier and fixed it. These guys are not idiots. If they didn't get the problem fixed before hand, they should have used a different platform or laptop.

If Nvidia didn't this, you would be burning them at the stake and not blame driver issues.

Regarding your comment on nvidia, when they used a fake card with wooden screws () they didn't let journalists to check it in person to confirm it's real.
AMD let the guys at AT to reinstall the drivers and re-do the tests, to show how open they are. Plenty good for me.

[dartaz]

Hopefully not because HD4290 runs at 700mhz and 40 shaders where as the Zacate is suppose to have a 500mhz part with 80 shaders. If a 80 shader @ 500mhz is equal to 40 shader @ 700mhz it points to a lack of optimization , driver issues among others.

There are other things that has impact on performance like TMU, ROP, memory bandwidth etc

[madcho]

There is no more ROP in ATi HD radeon. It's now RBE.

[RaV[666]]

Btw:



Means that Zacate gpu perf is something like 790GX-890GX.

Youre comparing old data, aquired with different platform/drivers and most importantly DIFFERENT CPU.
So this comparison is invalid, with the exception if the numbers you posted are with some 1.6ghz dual core on single channel memory ;-) (which i highly doubt).

Not to mention, that its not the same walkthrough as in anands preview, diffrent parts of the game can have substantially different performance numbers.We have to wait for some comparable benchmarks.All we know for now that zacate (1.6ghz dual core 80SP) is 40% faster than 2.4ghz clardkale with intels IGP, in that particular instance of a batman run.

[SimBy]

Btw:



Means that Zacate gpu perf is something like 790GX-890GX.

Like others have pointed, you are comparing apples to oranges. Let me guess, they used some uber i7 running @3GHz for this bench?

At lower resolutions, CPU becomes an important factor and the fact that i5 was running at 2.4GHz (50% faster) than Zacate and was still 40% slower speaks volumes.

[flyck]

At lower resolution it is more cpu bound because normally the graphic cards are faster. But as the screenshot indicates, there is a performance drop for each card. This means that the game at that resolution is gpu bound and not cpu bound. (cpu bound would be that the framerates were pretty much even). If we would have the speed of the i5 in that benchmark with those settings we could see how representative it is.

[kl0012]

Youre comparing old data, aquired with different platform/drivers and most importantly DIFFERENT CPU.

Like others have pointed, you are comparing apples to oranges. Let me guess, they used some uber i7 running @3GHz for this bench?

At lower resolutions, CPU becomes an important factor and the fact that i5 was running at 2.4GHz (50% faster) than Zacate and was still 40% slower speaks volumes.

There's no reason in differentiation between cpu and gpu when looking at zacate because the cpu part is not replaceable. What is the point in fast gpu if cpu can't handle it?
40% perf advantage over i5-520M is not that great speaking about gpu which is supposed to be in range of HD5450.

[Manicdan]

AMD just made it so netbooks will have more gaming power than MOST laptops are ever getting, why, because so many people buy a laptop with super low end graphics, and this is higher than any IGP to date?

why get so mad? why try to make it seem weaker than it really is? its 80SPs and i hope it can OC with CCC like any other amd gpu can, and 80SPs is > 40, thats a fact

[kl0012]

AMD just made it so netbooks will have more gaming power than MOST laptops are ever getting, why, because so many people buy a laptop with super low end graphics, and this is higher than any IGP to date?

why get so mad? why try to make it seem weaker than it really is? its 80SPs and i hope it can OC with CCC like any other amd gpu can, and 80SPs is > 40, thats a fact

No one get mad about, but if we can speak about Intel's "unbalanced" cpu/igp configs, why can't we speak about "unblalnced" AMD configs? AMD did not demonstrated perf of the cpu part of zacate which rises some concern. Reading this thread i got impression that some ppl think that the main purpose of small and thin notebooks is to play 3d games.

[Manicdan]

No one get mad about, but if we can speak about Intel's "unbalanced" cpu/igp configs, why can't we speak about "unblalnced" AMD configs? AMD did not demonstrated perf of the cpu part of zacate which rises some concern. Reading this thread i got impression that some ppl think that the main purpose of small and thin notebooks is to play 3d games.

testing of a mobile platform is much less important on max perf, but total perf done with a fixed energy cost

since that system is not a retail product, there will be no way to get an accurate total power consumption number, and then theres no way know what kind of battery life you get with the 1.x ghz they tested. amd is pushing for alot of the heavy stuff to be done by a gpu for a reason. what kind of tasks would you be running that are better with 3ghz vs 2? however you would notice if applications are choppy because there simply isnt enough power to run them.

[flyck]

There's no reason in differentiation between cpu and gpu when looking at zacate because the cpu part is not replaceable. What is the point in fast gpu if cpu can't handle it?
40% perf advantage over i5-520M is not that great speaking about gpu which is supposed to be in range of HD5450.

The id that the cpu would be in range of the 5450 is hokum.

5450 specs:
Radeon 5450 512MB DDR3
Stream Processors 80
Graphics Core Clock Speed 650MHz
Texture Units 8 8
Texture Fill-rate 5.2 Gigatexels/sec
ROPs 4
Pixel Fill-rate 2.6 Gpixels/sec
Memory Clock Speed 800MHz
Memory Interface 64-bit
Memory Bandwidth 12.8GB/sec
Typical Board Power 19.2W

Lets asume the rumour was true that gpu in zacate ran @ 500MHz, that is 150MHz or 24% less.

Memory Bandwtih 12.8GB/s. What is the bandwidth on Zacate? (and that is shared) + the latency.

Zacate was made to be as low power as possible while still having decent graphics. Like previous poster said, they give 40% higher gaming performance with 50% of the energy consumption (on cpu level) compared to the current notebooks out there without a dedicated card. Just like SB is a revelation in ondie gpu performance for desktops and high-end notebooks, Zacate is that for the low power market.

edit: I believe Zacate is powered towards gpu enhancments for applications (flash, HD, ..) and keep the power consumption as low as possible for those. That was the primary design of the APU, having another type of calculation unit that can be used to handle certain tasks at a much higher speed. Considering their gpu supports dx11, opencl etc developers have the opportunity to optimize their applications for these things. (wether that will happen is a whole different story)