Nintendo Maniac 64 Wrote:The easiest reference is that all three systems are fast enough to run YouTube 720p WebM via VLC, but not YouTube 720p MP4 via VLC. By comparison, my Brisbane can do YouTube 720p WebM @ 1GHz via VLC but YouTube 720p MP4 via VLC requires 1.3GHz. This is a performance gap of only 33% which is pretty small considering that's only the difference between a 2GHz Northwood vs a 2.66GHz Northwood.
Remember that the IPC difference will be different for different applications. Video encoding/decoding tends to favor the netburst architecture far more than other types of applications.
Nintendo Maniac 64 Wrote:So therefore I think it is safe to say that Atom has an IPC extremely similar to K7.
It does. Although it depends on the application and which generation of K7 you're referring to.
Nintendo Maniac 64 Wrote:Also, Brazos in my experience has an IPC that's about 75% of K8, which means the upcoming Jaguar core should have K8-levels of IPC.
1. Bobcat, not brazos
2. I really doubt that considering its IPC is ~22% higher than bonnell (atom) on average. I would expect the IPC on bobcat to be similar to K8 and jaguar to be slightly higher. But I could be wrong. I haven't crunched the numbers yet.
3. Jaguar is only going to boost IPC by 10-15% over bobcat according to amd (so likely 10% since AMD has a habit of inflating these numbers).
Atom IPC relative to P4 northwood:
3DS Max 2010
1.33
7-Zip COmpression
1.33
Adobe Acrobat 9 Pro PDF Creation
1.03
Adobe Photoshop CS4 Image Processing
1.63
WinRAR compression
1.22
WinZIP Compression
1
Fritz Chess game
1.3
iTunes Audio Encoding
0.65
Lame Audio Encoding
0.66
Mainconcept Reference v2.0 Video Encoding
1.26
Handbrake Video Encoding
1.26
Source:
http://www.tomshardware.com/reviews/atom-d510-pentium-4-nettop,2649-7.html
Taken from tomshardware comparing single core with HT in both cases (turning off HT would yield more accurate results). As you can see it depends on the application. For some reason the P4 had a 50% higher IPC in the audio encoding tests while having a lower or similar (usually lower) IPC in everything else. This is particularly odd given that the P4 should benefit more from HT than bonnell and both audio encoders were single threaded.
(03-15-2013, 09:57 AM)NaturalViolence Wrote: [ -> ]Remember that the IPC difference will be different for different applications. Video encoding/decoding tends to favor the netburst architecture far more than other types of applications.
Nintendo Maniac 64 Wrote:So therefore I think it is safe to say that Atom has an IPC extremely similar to K7.
It does. Although it depends on the application and which generation of K7 you're referring to.
Indeed, I just chose it because it was the easiest and most repeatable real-world test that I had done. I would have done Cinebench but it requires SSE2 which K7 lacks so I never got hard numbers, and currently the K7 and Northwood PCs are disassembled so it'd be difficult to run benchmarks. As for the K7 generation, mine is a Thoroughbred B - fun fact, I JUST discovered that the top-right number-code on my Thoroughbred's CPU label ends in 1337
.
(03-15-2013, 09:57 AM)NaturalViolence Wrote: [ -> ]Nintendo Maniac 64 Wrote:Also, Brazos in my experience has an IPC that's about 75% of K8, which means the upcoming Jaguar core should have K8-levels of IPC.
1. Bobcat, not brazos
Oh come on, Brazos uses Bobcat-based CPU cores, so it's not like I'm wrong.
(03-15-2013, 09:57 AM)NaturalViolence Wrote: [ -> ]2. I really doubt that considering its IPC is ~22% higher than bonnell (atom) on average. I would expect the IPC on bobcat to be similar to K8 and jaguar to be slightly higher. But I could be wrong. I haven't crunched the numbers yet.
Cinebench disagrees with you:
http://www.legitreviews.com/article/1470/6/
(03-15-2013, 09:57 AM)NaturalViolence Wrote: [ -> ]3. Jaguar is only going to boost IPC by 10-15% over bobcat according to amd (so likely 10% since AMD has a habit of inflating these numbers).
My estimate was based off of the supposed Cinebench results posted here:
http://www.overclock.net/t/1364086/sweclockers-amd-temash-apu-a6-1450-in-cinebench-r11-5
Also note the post I made:
http://www.overclock.net/t/1364086/sweclockers-amd-temash-apu-a6-1450-in-cinebench-r11-5/30#post_19371626
(03-15-2013, 09:57 AM)NaturalViolence Wrote: [ -> ]For some reason the P4 had a 50% higher IPC in the audio encoding tests while having a lower or similar (usually lower) IPC in everything else.
This is because media tasks benefit from a longer instruction pipeline, though I don't know why this is. Note that Bulldozer and Piledriver also have a similar but not as dramatic performance boost in media-type tasks as well, supposedly due to the longer pipeline over K10.
Nintendo Maniac 64 Wrote:Cinebench disagrees with you:
http://www.legitreviews.com/article/1470/6/
1. Cinebench isn't the only application on the planet. I don't know why it has become the "ultimate benchmark" on this forum. Cinebench is a good way to judge SSE performance in complex control flow scenarios.
2. The link you posted doesn't even mention the results of any other chip. This makes it difficult for me to make a comparison since I would need to look up the results of other chips and then adjust for clock rate.
Please try to make your case next time if the site you're linking to doesn't provide all the necessary data. And/or show me your math so I can confirm it.
Nintendo Maniac 64 Wrote:My estimate was based off of the supposed Cinebench results posted here:
http://www.overclock.net/t/1364086/swecl...ench-r11-5
Also note the post I made:
http://www.overclock.net/t/1364086/swecl...t_19371626
This is based off an unverified leak. Take it with a grain of salt. Your math is solid though so if it is true I can't argue with that.
Nintendo Maniac 64 Wrote:This is because media tasks benefit from a longer instruction pipeline, though I don't know why this is. Note that Bulldozer and Piledriver also have a similar but not as dramatic performance boost in media-type tasks as well, supposedly due to the longer pipeline over K10.
1. No data dependencies
2. Large data set made of small components
3. Simple sequential storage of data structures/elements
4. Functions tend to be long, simple, and repetitive
5. Almost all basic arithmetic operations. Very few changes in control flow, state changes, or other complicated control operations.
This makes pipelining extremely efficient and high ILP and TLP laughably easy to achieve. Pipeline width, clock rate, cache size, register file size, SMT implementation, and core count (and probably some other stuff) become more important and other factors like branch prediction and out-of-order execution control logic (scheduling) become less important.
But regardless of this fact all of the other media tasks (video encoding, image processing, 3D rendering) all have significantly higher IPC on bonnell. Yet the audio encoders have much higher IPC on netburst. Clearly this is more complex then "it's a multimedia application".
(03-15-2013, 11:26 AM)NaturalViolence Wrote: [ -> ]1. Cinebench isn't the only application on the planet. I don't know why it has become the "ultimate benchmark" on this forum. Cinebench is a good way to judge SSE performance in complex control flow scenarios.
Meant to only be a reference point, but if I'm remembering correctly (which I very well may not be) from informal testing I did a year or so ago I found an E-350 to be considerably slower than 75% IPC performance of K8 in the Fritz Chess benchmark. I believe the 75% was an average I had calculated from a verity of tests.
(03-15-2013, 11:26 AM)NaturalViolence Wrote: [ -> ]2. The link you posted doesn't even mention the results of any other chip. This makes it difficult for me to make a comparison since I would need to look up the results of other chips and then adjust for clock rate.
Please try to make your case next time if the site you're linking to doesn't provide all the necessary data. And/or show me your math so I can confirm it.
I was expecting you to refer to the Cinebench results for my Brisbane that I linked to farther down in my post...I guess I should have spelled it out for you (normally I do, but I figured you were sharp enough to pick it up yourself).
(03-15-2013, 11:26 AM)NaturalViolence Wrote: [ -> ]This is based off an unverified leak. Take it with a grain of salt.
We don't have 100% legit numbers yet, so it's the only thing we have to work with in terms of predicting Jaguar's performance.
Nintendo Maniac 64 Wrote:I was expecting you to refer to the Cinebench results for my Brisbane that I linked to farther down in my post...I guess I should have spelled it out for you (normally I do, but I figured you were sharp enough to pick it up yourself).
I'm not that stupid. The post that you linked was referring to Jaguar vs. K8. The line that you quoted was referring to atom vs. bobcat.
I said that bobcat had a 22% higher IPC than bonnell on average. You said that cinebench disagreed with me. I asked you to back it up and you responded with that legitreviews link. The link shows bobcat performance but does not show bonnell performance in cinebench. And for the record neither does the thread that you linked me from that forum. So I then asked you add more data (and math) when making your case. You still haven't made that case for why I'm wrong about bobcats IPC relative to bonnell which is what I asked you to do.
Isn't your "I really doubt that" referring to my estimate of Bobcat's IPC being 75% of K8? That's how I took it.
It is. Although that's now what I was asking for when I asked for clarification later on (I quoted the legitreviews link so it should have been obvious that I was responding to that). Anyways you haven't established the IPC difference between bobcat and jaguar either so that still doesn't work. Your math would only work if there was a big difference in IPC (+33%) between Jaguar and bobcat which there doesn't appear to be. Either that or my bonnell vs. bobcat IPC comparison is flawed. I don't think it is but you could certainly try to prove me wrong by posting some bobcat vs. bonnell benchmarks.
Looking at the benchmark chart, why is there a 2500k 4.5ghz chip there destroying newer and faster clocked cpus like that 3570k @ 4.7ghz? I feel bad inside all of a sudden for ordering a 3570k for a new build lol (my old computer had a 2500k @ 4.5ghz but I couldn't bring it with me when I moved to the US early this year)
Are you seriously upgrading to a 3570K from a 2500K?
Ah I see. Still...
You should've waited man. Haswell is right around the corner.
That 2500K result is mine. Oddly enough,
a similar clocked 3570K still doesn't outdo my result.
(03-17-2013, 07:12 AM)Garteal Wrote: [ -> ]Are you seriously upgrading to a 3570K from a 2500K?
Ah I see. Still...
You should've waited man. Haswell is right around the corner.
That 2500K result is mine. Oddly enough, a similar clocked 3570K still doesn't outdo my result.
Margin of error maybe, but that is weird. Anyway yeah, I have to build a new rig. I used a pc building thread over at NeoGAF for tips and it looks like haswell won't be a big upgrade speed-wise but it'll mostly contain improvements to power management or something like that. I just need something up and running asap and I found a sweet deal at microcenter where you can get $40 off a motherboard paired with a 3570k (which also already has a sweet discounted price there - 189.99).
I guess my real problem would be heat issues with ivybridge. Hopefully I get lucky and can still pull off 4.5 with a decent hsf