Fudzilla: Bulldozer performance figures are in

[Abwx]

My french is pretty rusty, but I believe they are saying something to the effect that benches with FX are all over the place and they removed some that lead too much.

Lead being extremely heavy , the french terms mean in fact
the benches that are too much on the bottom side ;i.e , the
ones that seems to not perform well enough since the guy
that did the tests did try to recompile some softs as well
as using AVX , wich he said didnt work with some softs
such as Linpack...

There have been much infos given on a french site forum
Hardware.fr by this article s author.

 

[Despoiler]

Hmmm, did they trim off the ones that seemed too low as well? I stopped taking French after Year 11 in High School.

"Plombaient" is the operative word. I take it to mean "leads", but that is literal.

 

[drizek]

So, the conclusion from those benchmarks is that Bulldozer sucks.

At best, it is 7% faster than a 2600K. That's already suck territory. Then consider the possibility of it being up to 35% slower than a 2500K...

Edit: actually, I think it is highly likely that the 100% bar is the 2500K, and the CPU that consistently scores above it is, of course, the 2600K, which puts Bulldozer... somewhere below it, except for maybe that first benchmark where there are two CPUs above 100%.

 

[Vesku]

How is 7% faster than a 2600K suck territory? Taking the details known about BD then even if it has Nehalem level IPC it would only be roughly equivalent to 5.5ish non-HT SB cores. Personally, I think it would be a pretty big win for AMD if the 8150 was overall faster than the 2600K and the rumored pricing is accurate.

 

[frozentundra123456]

How is 7% faster than a 2600K suck territory? Taking the details known about BD then even if it has Nehalem level IPC it would only be roughly equivalent to 5.5ish non-HT SB cores. Personally, I think it would be a pretty big win for AMD if the 8150 was overall faster than the 2600K and the rumored pricing is accurate.

If it is overall 7% faster than 2600K that would be impressive. But if it is only faster on one or 2 cherry picked benchmarks (very highly multithreaded) and overall quite a bit slower, that is less impressive, and what I expect.

 

[drizek]

How is 7% faster than a 2600K suck territory? Taking the details known about BD then even if it has Nehalem level IPC it would only be roughly equivalent to 5.5ish non-HT SB cores. Personally, I think it would be a pretty big win for AMD if the 8150 was overall faster than the 2600K and the rumored pricing is accurate.

The 2600K came out 9 months ago. Intel is going to release 22nm chips at CES, and Bulldozer is still a no show. Moreover, thats only an advantage in content creation. It is inferior in games and productivity.

The price of BD is not particularly aggressive. I almost purchased a P68-Deluxe with a 2600K combo for $450 before deciding to wait for BD, but seeing as how I have spent $200 for an equivalent 990FX board, spending $250 for an 8150P 6 months later for only a single-digit advantage in only a handful of apps isn't really particularly compelling.

AMD can't win on price with this one. IB is likely going to be priced even more aggressively than SB, and once people start hitting 6GHz overclocks on a chip with even better IPC than SB that they got for $150 at microcenter... AMD is going to be in trouble.

 

[Vesku]

If it is overall 7% faster than 2600K that would be impressive. But if it is only faster on one or 2 cherry picked benchmarks (very highly multithreaded) and overall quite a bit slower, that is less impressive, and what I expect.

Yes, my current best guess is multi-threaded between a 2500K and a 2600K and single threaded being just a bit behind Nehalem on an IPC basis which is why clocks and max stable 24/7 OC will be important.

The 2600K came out 9 months ago. Intel is going to release 22nm chips at CES, and Bulldozer is still a no show. Moreover, thats only an advantage in content creation. It is inferior in games and productivity.

Considering the difference in resources between the two companies and that AMD is currently selling a Core 2 equivalent architecture versus Intel's 2nd iteration past the Core 2, I think winning some and losing some is a decent change. Even before they sold off their fabs AMD was taking longer and longer to catch up to Intel in terms of fabrication. Granted, this doesn't mean the consumer should do anything other than make purchasing decisions based on their personal needs. But it should tone down expectations of them being the tech leader in their industry.

AMD can't win on price with this one. IB is likely going to be priced even more aggressively than SB, and once people start hitting 6GHz overclocks on a chip with even better IPC than SB that they got for $150 at microcenter... AMD is going to be in trouble.

Last rumors had IB out no sooner than April '12. What the delays have done though is put the BD launch awfully close to Intel's SB-E release.

 

[drizek]

OCd clockspeed will have to approach 6ghz for it to keep up with an OCd 2600K.

 

[frozentundra123456]

Trinity is supposed to be launched in Q1 2012, so we'll see. Maybe it's delayed to Q2 and is finally launched in Q3.

It should bring higher CPU and IGP performance along with lower power consumption than Llano, naturally.

But isnt Trinity based on the Bulldozer core?? Bulldozer itself may not be out until 2012 in any significant numbers. But yes, Trinity might be interesting. Llano for the desk top had an outdated CPU and not quite good enough graphics to make up for it.

If trinity can improve both it could be a good solution for me. I am currently running an E4500 and 9800GT. If trinity could get graphics performance near that level with a better CPU, I might like to upgrade.

 

[Tuna-Fish]

For your information, SB has 3 ALU ports and 2 "AGU" ports.


K10 has symmetric ALU/AGU organization with 3 ALU and 3 AGU units and separate FP coprocessor unit. It can execute up to 9 uops but retire only 3 macro ops .

Yes, I know all that. Also, executing more than 3 ops is typically limited by register read ports, not instruction retiring.

 

[LOL_Wut_Axel]

Yes, my current best guess is multi-threaded between a 2500K and a 2600K and single threaded being just a bit behind Nehalem on an IPC basis which is why clocks and max stable 24/7 OC will be important.

If it had only 5% lower IPC than Nehalem it would trash the 2600K in multi-threaded, being comparable to the 990X. Obviously that won't happen, so you can't expect a higher-than-10% IPC improvement in comparison to Phenom II. For the FX-8150, comparable multi-threaded and much lower single-threaded than the 2600K. For the FX-8120, somewhat higher multi-threaded and much lower single-threaded than the 2500K. That's what I think we'll see.

 

[bronxzv]

"Plombaient" is the operative word. I take it to mean "leads", but that is literal.

as a native French speaker I'll say that "plombaient" must be translated by "taken down"

FYI in French "plomb" is the name for "lead" the heavy chemical element (thus the Pb symbol)

 

[AdamK47]

Eh, thats debatable, but even so, gaming is irrelevant to CPU's in this day and age for several reasons.

1. Games in their nature arent very CPU intensive, and are usually bottlenecked by the GPU or other parts.
2. The games coming out now either are console ports and therefore any modern CPU with a gaming grade GPU can get 100FPS+, or a few select games that arent console ports like BF3 will be able to use some insane amount of cores (16 or 64 cores, I forget to be honest).
3. Heres a perfect example:

http://www.tomshardware.com/reviews/phenom-ii-x6-1100t-thuban-amd,2810-6.html

Even though the 1100T scores 19.1s in SuperPI, it doesnt show at all in gaming. Furthermore, the BD with a score of 19.5s will surely be closer to around 16.5s when the ES is brought up to faster clock speed. Only real world gaming benchmarks would show, however gaming for CPU's is almost irrelevant, so no matter.

Your perspective on CPUs for gaming is interesting and may even be unique. Could your perspective be GPU limited?

 

[Despoiler]

as a native French speaker I'll say that "plombaient" must be translated by "taken down"

FYI in French "plomb" is the name for "lead" the heavy chemical element (thus the Pb symbol)

Oh god, lead like the element. That is a really odd use of the word when translated. Sadface for Bulldozer

Sent from my DROIDX using Tapatalk

 

[LOL_Wut_Axel]

Your perspective on CPUs for gaming is interesting and may even be unique. Could your perspective be GPU limited?

I do find it interesting. Clearly, CPU performance is irrelevant in gaming.

This clearly demonstrates that.

 

[Vesku]

Ok, so 5% under Nehalem IPC might look something like this (8150 actually being almost equal to the 2600K in multi-threaded and maybe a little behind if HyperThreading gets good returns):

8 (clusters) * .8 (BD to SB IPC ratio) * .8 (performance hit for clustering) = 5.12 (SB equiv MT)

4 (SB cores) * 1.3 (Gains from Hyperthreading) = 5.2 (SB equiv MT)

As reproduced from my earlier post -

http://forums.anandtech.com/showpost.php?p=32333187&postcount=1856

So maybe it is roughly 5% under Nehalem IPC, that would fit with drizek's guess that 2500K is the 100% reference. 8150 would most likely be the ~102% bar and 2600K the 112% bar. Not sure if I'm going to throw in with drizek just yet but I would love for more people to guess at what the reference CPU is in those CanardPC charts.

If it had only 5% lower IPC than Nehalem it would trash the 2600K in multi-threaded, being comparable to the 990X. Obviously that won't happen, so you can't expect a higher-than-10% IPC improvement in comparison to Phenom II. For the FX-8150, comparable multi-threaded and much lower single-threaded than the 2600K. For the FX-8120, somewhat higher multi-threaded and much lower single-threaded than the 2500K. That's what I think we'll see.

 

[LOL_Wut_Axel]

Ok, so 5% under Nehalem IPC might look something like this (8150 actually being almost equal to the 2600K in multi-threaded and maybe a little behind if HyperThreading gets good returns):

8 (clusters) * .8 (BD to SB IPC ratio) * .8 (performance hit for clustering) = 5.12 (SB equiv MT)

4 (SB cores) * 1.3 (Gains from Hyperthreading) = 5.2 (SB equiv MT)

As reproduced from my earlier post -

http://forums.anandtech.com/showpost.php?p=32333187&postcount=1856

So maybe it is roughly 5% under Nehalem IPC, that would fit with drizek's guess that 2500K is the 100% reference. 8150 would most likely be the ~102% bar and 2600K the 112% bar. Not sure if I'm going to throw in with drizek just yet but I would love for more people to guess at what the reference CPU is in those CanardPC charts.

Integer performance won't suffer from the module design, but floating point will (in comparison to having everything dedicated). Given that, it's not that clear.

If it's only a 5-10% IPC improvement, which looks more like what it'll probably be, what I said will pan out. The Phenom II X6 1100T is currently 20-25% slower than the 2600K in multi-threaded, and given two more cores, a bit higher clock speed, and a bit higher IPC, you'd be at comparable performance, sometimes a tiny bit better.

 

[Vesku]

Never heard AMD claim no penalty on integer performance versus 2 theoretical monolithic BD class cores. They did say it was closer to 90% than 80% but I chose to not be too rosy when making the comparison, hence the 30% hyperthreading gain for 2600K.

 

[LOL_Wut_Axel]

Never heard AMD claim no penalty on integer performance versus 2 theoretical monolithic BD class cores. They did say it was closer to 90% than 80% but I chose to not be too rosy when making the comparison, hence the 30% hyperthreading gain for 2600K.

Only the FPU shares important resources. When only one SSE/256-bit AVX FP thread is being run on it, it'll have the same performance as dedicated. When two threads are being run on the module, you get a loss of performance from them having to share resources, and you can no longer run 256-bit AVX--mostly relevant for enterprise users and people being looked at by the FBI, but still.

Shared L2 cache won't result in any performance loss for integer. It's the same concept as Core 2 where all cores have access to a single large L2 cache. Unlike Core 2, they also have access to a large L3 cache.

 

[Vesku]

Most code is still SSE and there are 2 128 bit units for each module. The integer cores share the whole front end I thought, prefetcher and such. Pretty sure there will be some hit even on pure integer code, more power to AMD if they can make a shared front end cost no performance.

 

[SimplyComplex]

I do find it interesting. Clearly, CPU performance is irrelevant in gaming.

[snip 4 charts showing 35fps vs 90 based on cpu]

This clearly demonstrates that.

You sir, are made of win.

 

[LOL_Wut_Axel]

Most code is still SSE and there are 2 128 bit units for each module. The integer cores share the whole front end I thought, prefetcher and such. Pretty sure there will be some hit even on pure integer code, more power to AMD if they can make a shared front end cost no performance.

Found some info regarding that, and yes, you're right.

The decoder is now 4-wide an increase from the 3-wide front end that AMD has had since the K7 all the way up to Phenom II. AMD can now fuse x86 branch instructions, similar to Intel’s macro-ops fusion to increase the effective width of the machine as well. At a high level, AMD’s front end has finally caught up to Intel, but here’s where AMD moves into the passing lane.
The 4-wide decode engine feeds three independent schedulers: two for the integer cores and one for the shared floating point hardware.


Each integer scheduler is now unified. In the Phenom II and previous architectures AMD had individual schedulers for math and address operations, but with Bulldozer it’s all treated as one.

Each scheduler has four ports that feed a pair of ALUs and a pair of AGUs. This is down one ALU/AGU from Phenom II (it had 3 ALUs and 3 AGUs respectively and could do any mix of 3). AMD insists that the 3rd address generation unit wasn’t necessary in Phenom II and was only kept around for symmetry with the ALUs and to avoid redesigning that part of the chip - the integer execution core is something AMD has kept around since the K8. The 3rd ALU does have some performance benefits, and AMD canned it to reduce die size, but AMD mentioned that the 4-wide front end, fusion and other enhancements more than make up for this reduction. In other words, while there’s fewer single thread integer execution resources in Bulldozer than Phenom II, single threaded integer performance should still be higher.

There should be a loss compared to having it dedicated with another ALU, and many design elements are now integrated. However, I doubt this would be anywhere near the same performance penalty as from the shared FPU.

 

[drizek]

How terrible would it be if the two CPU's in the content creation benchmark above the hypothesized 2500K are the 2600K and... the 1100T?

 

[mosox]

I do find it interesting. Clearly, CPU performance is irrelevant in gaming.
This clearly demonstrates that.

What resolution for the Anand benches? Yeah low ones. What game on Tom's, what GPU? Why a single game?

10% between the Athlon II X4 and the 2500K in full HD.

 

[LOL_Wut_Axel]

What resolution for the Anand benches? Yeah low ones. What game on Tom's, what GPU?

10% between the Athlon II X4 and 2500K in full HD.

Because clearly all games are GPU-bound, and the only game to ever exist is Crysis.

And the tests were at 1680x1050, so not a low res either. At 2560x1440 is where most games typically don't care much about the CPU.