AMD Bulldozer Dual-Interlagos Benchmarks On Linux

[Arkadrel]

http://www.phoronix.com/scan.php?page=article&item=amd_bulldozer_linux&num=1


^ dual socket old motherboard + 2x 16 "core" engineering sample bulldozers.


Check out the Raytraceing Benchmarks for C-Ray v1.1

http://openbenchmarking.org/result/...IV-AMDWORKST94,1102261-IV-1102243IV79&compare

to put it into perspective:

While there are other software/hardware differences in play too, the 32-core 1.80GHz Bulldozer system's 25 seconds compared to the Intel Core i5 2500K (quad-core + Hyper Threading; 3.3GHz + 3.7GHz Intel Turbo Boost) at 61 seconds or the dual quad-core AMD Opteron 2384 system of ours taking 127 seconds to complete. The Intel Core i7 970 (six cores + Hyper Threading; 3.2GHz Base Frequency + 3.46GHz Turbo Boost) comes in at about 61 seconds too.

I just wish they would have shown some 4ghz bulldozers instead of 1.8ghz cores.

Anyways...

lets say that 32-core 1.8ghz at 25secounds, was like ~50 secounds with "just" 16 "core"s... which would be 1 cpu (@1.8ghz). It would still be faster than the Core i5 2500k (that runs at 3.7ghz with turbo) in this benchmark.

also its a engineering sample, its bound to get better performance when its released, and undoubtably see much faster speeds (than just 1.8ghz).

 

[Martimus]

The low clock speed that was benchmarked does not give me a warm and fuzzy feeling about the upcoming release.

 

[busydude]

The low clock speed that was benchmarked does not give me a warm and fuzzy feeling about the upcoming release.

What is with 1.8GHz? Even Llano's demo was with a 1.8GHz ES..

 

[podspi]

The low clock speed that was benchmarked does not give me a warm and fuzzy feeling about the upcoming release.

I'm not too worried about that. Some of these benchmarks look good, but some of them make me cringe. Like this one.

http://openbenchmarking.org/result/1103223-IV-HIMENUBUL97,1103164-IV-NEWINTELC72&compare


D:

Obviously the clocks are going to be much higher but that benchmarks implies IPC is lower than K10.5! Is anybody more knowledgeable about these benchmarks and what can affect their results? Of course, this could be fake but it 'smells' to me like our first real leak.

 

[Idontcare]

podspi, your link seems to be borked, what benches are you looking at?

 

[podspi]

Himeno Benchmark v3.0 (Poisson Pressure Solver).

Bulldozer ES #1.8ghz vs PhII 710, a bunch of i3s, and of course the 2500K. Assuming the results scale linearly (often a poor assumption, I know) even at the same clock speed as the 710, 32-core BD would appear to lose...

 

[busydude]

podspi, your link seems to be borked, what benches are you looking at?

That link works for me.. anyway its himenu_bulldozer, you can find it in the latest results section.

 

[RobertPters77]

I don't understand what the benchmarks are saying.

Is their indication of BD IGP clock for clock compared to sandy?

 

[Dresdenboy]

I don't understand what the benchmarks are saying.

Is their indication of BD IGP clock for clock compared to sandy?

The mem performance is 1/10 of older AMD CPUs. So nobody should go to the shops and buy a BD ES

 

[RobertPters77]

@dresdenboy

Ok I won't.

Does anyone have any comparable benchmarks with opterons?

 

[RobertPters77]

What is with 1.8GHz? Even Llano's demo was with a 1.8GHz ES..

The original A64's and Opterons launched at 1.8ghz. It must be symbolic ofcourse.

 

[podspi]

The mem performance is 1/10 of older AMD CPUs. So nobody should go to the shops and buy a BD ES

That's actually a relief. Could be a reason why the results are so... :\

 

[Arkadrel]

I guess it depends on alot of things, but Im sure their just testing things and working out kinks with an engineering sample.

The C-ray (raytraceing benmark) seems to look okay.
It looks like 16 "core" 1.8ghz bulldozer > i5 2500k 3.7ghz (turbo on) in that one.

But yeah not all of them look that impressive.... probably just those kinks they need to work out. I still wish theyd show a 4ghz bulldozer even if they had to use a 8 "core" version or so..... also... these benchmarks.. how many of them use up to 32 threads?

 

[IntelUser2000]

Explaining the results:

Himeno is a benchmark that's extremely limited by memory bandwidth. C-ray hits the FPU without hitting the memory subsystem.

C-ray:
-Linear clock speed scaling
-Near linear scaling with amount of cores
-No thread count limit observed
-Benefits with SMT

A quad socket setup based on the 2500K clocked at 2.4GHz would score similarly.

 

[podspi]

Explaining the results:

Himeno is a benchmark that's extremely limited by memory bandwidth. C-ray hits the FPU without hitting the memory subsystem.

C-ray:
-Linear clock speed scaling
-Near linear scaling with amount of cores
-No thread count limit observed
-Benefits with SMT

A quad socket setup based on the 2500K clocked at 2.4GHz would score similarly.

Quad socket 2500K They compare it to a quad-socket Intel...

 

[Zstream]

If bulldozer stays at 1.8ghz, it will come between a i7 920 and a i7 950 in terms of gaming performance.

If bulldozer is at 2.4ghz, it will nip at the heals of the 2500k in gaming performance and beat it barely on multi-threaded applications.

If bulldozer is at 2.8ghz, it will beat the 2500k by 5-10% in gaming and 15% in heavily threaded applications.

8 core bulldozer I mean.

 

[Arkadrel]

bulldozer is supposed to be a high speed designed cpu.

The only reason you would see a 1.8ghz is because its a low energy server part with 16 cores. Also engineering samples probably arnt clocked that high at first?

Im guessing we ll see the 8 core versions at 3.5ghz atleast, with 500mhz turbo meaning stock 4ghz for most things.

 

[RaistlinZ]

32 Core @ 3.5GHz would be yummy.

 

[IntelUser2000]

Base clock w/o Turbo speeds probably isn't that far away from 1.8GHz for a server part with 16 cores like Interlagos.

 

[Idontcare]

If bulldozer stays at 1.8ghz, it will come between a i7 920 and a i7 950 in terms of gaming performance.

If bulldozer is at 2.4ghz, it will nip at the heals of the 2500k in gaming performance and beat it barely on multi-threaded applications.

If bulldozer is at 2.8ghz, it will beat the 2500k by 5-10% in gaming and 15% in heavily threaded applications.

8 core bulldozer I mean.

Z, I believe you, can you tell me what GHz you'd project are needed to but an 8-core on par with a 2600K? I haven't really looked much into the performance difference between a 2500K and 2600K.

 

[Riek]

Explaining the results:

Himeno is a benchmark that's extremely limited by memory bandwidth. C-ray hits the FPU without hitting the memory subsystem.

C-ray:
-Linear clock speed scaling
-Near linear scaling with amount of cores
-No thread count limit observed
-Benefits with SMT

A quad socket setup based on the 2500K clocked at 2.4GHz would score similarly.

something is missing in the C-ray list of scaling.

4 8core intel cpu's (32core) running at 2GHz score half of the 32core 1.8GHz ES interlagos.

This intel system scales above lineair compared to the desktop part. e.g.6core/12T scoring 61seconds. So there is memory limitiation or HD limitation. Interlagos streaming result really sucked and with a desktop HD of TB one might expect this to be limiting factor also.

Edit an 48core 1.9GHz (quad MC) scores far better then this dual interlagos 1.8Ghz. using the effect of lineair scaling in clock and cores would make a quad Mc with 8cores running at 1.8GHz 3-4 seconds faster than this 2p interlagos @ 1.8GHz

 

[Borealis7]

Explaining the results:

Himeno is a benchmark that's extremely limited by memory bandwidth. C-ray hits the FPU without hitting the memory subsystem.

C-ray:
-Linear clock speed scaling
-Near linear scaling with amount of cores
-No thread count limit observed
-Benefits with SMT

A quad socket setup based on the 2500K clocked at 2.4GHz would score similarly.

then it isnt surprising DB does well in Ray-Tracing because, according to AMD, DB is a FP "monster".

perhaps AMD wants to usher in a new era of graphics processing with chips that specialize in Ray Tracing and then they'll be up against Intel in the graphics department (LRB) and not against nVidia.

 

[Zstream]

Z, I believe you, can you tell me what GHz you'd project are needed to but an 8-core on par with a 2600K? I haven't really looked much into the performance difference between a 2500K and 2600K.

Well the extra cache helps the 2600k in heavy multi-threaded applications but HT is a hit or miss type thing. I'd wager 2.9 to 3ghz would reach it.

I don't know how turbo works on the AMD side so *shrug*

Whomever thinks that AMD is going to be releasing 3.5ghz parts needs to take a step back and rationally think about this. I mean, Bulldozer can be a high MHZ player but the micro-architecture will stall after a certain point. They will refine the architecture/process and bin some 3.4ghz+ cpu's eight to twelve months down the line.

I'll wager the top of the line cpu will be 3ghz and at 3.4ghz turbo.

 

[Martimus]

Whomever thinks that AMD is going to be releasing 3.5ghz parts needs to take a step back and rationally think about this. I mean, Bulldozer can be a high MHZ player but the micro-architecture will stall after a certain point. They will refine the architecture/process and bin some 3.4ghz+ cpu's eight to twelve months down the line.

I'll wager the top of the line cpu will be 3ghz and at 3.4ghz turbo.

The current 45nm X6 1100T runs at 3.3GHz. It doesn't seem unreasonable for a new 32nm (with HKMG) X8 to run faster than that, especially considering that the pipeline and cache latency all was designed for faster clock speeds than the current X6.

 

[Tuna-Fish]

I'll wager the top of the line cpu will be 3ghz and at 3.4ghz turbo.

They have CPU's out at 3.4GHz right now, and in BD the longest path in the CPU was reportedly shortened from 22 in K10.5 to 17, so working on the same process, you'd expect to gain ~30% more clocks from architecture alone. If BD doesn't ship at above 3.5GHz and single core turbo well above 4GHz (nearing on 5), there is something horribly wrong with the GF 32nm SOI process.

Just what analysis did you base your 3.0GHz number on?

(edit) Note that it's not all roses -- they pay a high price for hunting higher clocks, with many instruction latencies going up. Based on released compiler latency numbers, they've ditched many pieces of special-purpose hardware, for example the floating point divider, and cache access became slower.