Found another one: Ukraine store accidentally ships FX-8120 and it gets tested!

[LondonBurning]

I'm waiting for official reviews come the 12th. I'm hearing too many conflicting sides of the story. Is there a chance these 'leaked' benches have been somehow caused by AMD limiting the CPU in some way before the truth about its performance is revealed?

I'm still betting on an Nvidia/AMD style neck-and-neck battle (at the given price point).
The added twist is what can Catalyst driver people do when a system runs a BD and Radeon in tandem (this has been somewhat hinted at in the recent AMD blog, but it could just be marketing).

 

[Keysplayr]

Yet another false leak campaign from AMD? I hope not and just rather they keep it shut until official NDA expiration.

 

[Schmide]

Why does the chip string say ES (engineering sample) on it???

 

[Ryun]

Higher IPC than Core 2 Duo? Not really setting the bar high there, lol.

And yet, from the initial looks of it, I set it too high!

I have a C2D E6600 right now OC'ed to 3.0GHz. Clock for clock, Phenom II is either slightly ahead or slightly behind Penryn so it's not a real clear win for Phenom II over the similar Conroe.

I'm not really willing to give up the IPC I have now, though I could use some more cores and a better board for my next upgrade. I would prefer to give AMD my money over Intel but afterall, I don't want to lose performance either. I'm not planning a platform upgrade until mid Q2 2012 so really it's kind of a moot point on my part.

 

[Hauk]

Then you want a 2600k. At stock it beats an OC'd Bulldozer in both single and multi-threaded apps.

An OC'd 2600k would downright embarrass it. And this is coming from an AMD owner*

(*assuming these benchmarks and all the others that have been popping up now that the press kits are out are legitimate)

Done..

 

[Tsavo]

:|

I wonder if he's using W7 SP1 and LinX with the libraries updated to use AVX.

My Core i5-2410M @ 2.7GHz manages 32 GFlops and it's only 2C/4T.

Using LinX with non-AVX libraries and the same CPU manages 17 GFlops in LinX.

...so I hope for AMD's sake that he's either using W7 without SP1 or LinX with outdated libraries.

:|

 

[Tsavo]

:|

...so I hope for AMD's sake that he's either using W7 without SP1 or LinX with outdated libraries.

:|

Even then, it doesn't beat a stock clocked Core i5 750. :'(

 

[sequoia464]

Found this link at overclock.net - it is purported to be a live feed of the AMD event going on today ... if anyone is still interested

Not sure which of the BD threads to post in, hate to start another. They (AMD) have confirmed release and reviews next week - (from the chat on this event).

Link ...http://www.twitch.tv/amd

 

[alyarb]

Done..

We are halfway through the 2600k's life, so anyone who is mulling over making a move may as well wait out the other half. I've been officially bored with both this node and SNB since they tied up the BCLK.

32nm was fun a year ago because you had $100 clarkdales doing 4.6 GHz on a $70 board, and though all those parts are still available online, there is no member of the SNB family that corresponds to the same value or anywhere near it.

So now I hear that IVB has clock ratios in the system agent such that for certain increases in BCLK, the GPU, PCIe and so forth are given a smaller ratio so they will run within spec. Probably the most unnecessarily kind thing intel has done for people like us since I can't imagine how it benefits them to include this capability, but IF it's true I will donate to IVB if only to reward their behavior. Plus the whole paradigm shift in xtor manufacturing thing.

 

[Accord99]

Could it possibly be a bad windows\scheduler\BIOS issue, in other words, would a better driver or perhaps a software patch improve BD's performance?

Many of the tests done by the various leaks scale to 8 or more threads, each thread being CPU intensive. In such situations I would assume that the scheduler would not matter so much as every core is running a thread that is doing a lot of work. I'd only expect the scheduler to matter when the number of CPU intensive threads are less than the number of available cores.

 

[frostedflakes]

There was an AMD slide that showed some pretty solid improvements with the Windows 8 scheduler over the one in Windows 7, from a few percent up to 10% better performance with 8-core Bulldozer I believe.

Of course an improved scheduler would benefit Intel's HyperThreaded CPUs just as much probably. So that may help Bulldozer, but it's not like Intel's performance is going to stay static and not be increasing as well.

 

[sm625]

My iPoop is having a hell of a time rendering this page. I think this thread should be stickied and used for benchmarking.

Edit: This thing is seriously hot... its as if I'm playing infinity blade. I think its having a meltdown.

 

[formulav8]

We are halfway through the 2600k's life, so anyone who is mulling over making a move may as well wait out the other half. I've been officially bored with both this node and SNB since they tied up the BCLK.

32nm was fun a year ago because you had $100 clarkdales doing 4.6 GHz on a $70 board.

Definitely This! I can't stand that Intel was so greedy to even hinder bus style ocing. (or Reference clock if you prefer). That is why ocing is nearly dead and boring to me when it comes to Intel. You have to buy their overpriced K chips to do any real ocing (If $200-$300 is Not over-priced to You then Great for You!). I liked the ocing days when taking a lower cost chip, for example say an $80 chip and transform it to near a $160 or more chip. But thanks to Intel its not currently possible for SB. Blah I say. Greedy Guts :thumbsdown:

 

[Atreidin]

Why does the chip string say ES (engineering sample) on it???

I was wondering the same thing. Yet another fake "leak" to get people all riled up?

 

[classy]

I was wondering the same thing. Yet another fake "leak" to get people all riled up?

I don't know what to believe, but these seem awfully suspect. This guy is not under any NDA and he has a pic up that would give you the impression he is some kind of reviewer. But yet not one game benchmark and the same benchmark stuff we have been seeing now for the last 2 months and many of them are confirmed to be fake benchmarks. Not one encoding benchmark, nothing. What are the chances any of us would have something like this fall in our laps and we not run full on scale benches, especially games.

 

[Black96ws6]

I was wondering the same thing. Yet another fake "leak" to get people all riled up?

It's because he was using CPU-Z 1.58 at first, which doesn't read BD right. If you look at the later screenshots he has upgraded to 1.58.6 or .7 and it reads it properly...

 

[Blue Shift]

he numbers that look the most ridiculous are the TDP numbers... 124 watts for an 8120, but 211 watts for a 2130P at the same clocks? Not likely.

Actually, what IS an 8130P? Why does CPU-z say "8120... (ES)" in the specification field for it? This doesn't look any more convincing than the benchmarks we saw last month.

 

[Black96ws6]

he numbers that look the most ridiculous are the TDP numbers... 124 watts for an 8120, but 211 watts for a 2130P at the same clocks? Not likely.

Actually, what IS an 8130P? Why does CPU-z say "8120... (ES)" in the specification field for it? This doesn't look any more convincing than the benchmarks we saw last month.

See my post above.

 

[alyarb]

Definitely This! I can't stand that Intel was so greedy to even hinder bus style ocing. (or Reference clock if you prefer). That is why ocing is nearly dead and boring to me when it comes to Intel. You have to buy their overpriced K chips to do any real ocing (If $200-$300 is Not over-priced to You then Great for You!). I liked the ocing days when taking a lower cost chip, for example say an $80 chip and transform it to near a $160 or more chip. But thanks to Intel its not currently possible for SB. Blah I say. Greedy Guts :thumbsdown:

They were just too occupied with giving newbs a (more costly, marketable) way to get the "speed" they thought they wanted without it being too hard for them, and inadvertently took all the sport out of it for the rest of us. Not only that but lot of people now think that having your cores at 5 GHz is all that counts, but it's such a waste of power because your L3 and IMC are still stuck at their crappy locked multiplier x 100 MHz. having uber fast cores and 2133 RAM is almost meaningless when the pipe connecting them is only intended to be wide enough for a 3 GHz part. They shorted us on the robustness of the IMC as well as their process technology, at least in this sense. And it's kind of an insult to people who know what they're doing.

Like I said, if this is corrected in IVB then I will be buying.

 

[Accord99]

Not only that but lot of people now think that having your cores at 5 GHz is all that counts, but it's such a waste of power because your L3 and IMC are still stuck at their crappy locked multiplier x 100 MHz. having uber fast cores and 2133 RAM is almost meaningless when the pipe connecting them is only intended to be wide enough for a 3 GHz part.

S1155 Sandy Bridge's L3 runs at core clock. I'm not sure about the IMC but given how well memory bandwidth and latency scales with faster memory, it probably doesn't matter whether it is or not.

 

[996GT2]

:|

I wonder if he's using W7 SP1 and LinX with the libraries updated to use AVX.

My Core i5-2410M @ 2.7GHz manages 32 GFlops and it's only 2C/4T.

Using LinX with non-AVX libraries and the same CPU manages 17 GFlops in LinX.

...so I hope for AMD's sake that he's either using W7 without SP1 or LinX with outdated libraries.

:|

Regardless of whether he was using AVX or not, those results are extremely poor.

For reference, my 2500K @ 4.7 GHz, using the same problem size in Linpack as the one used in the leaked benches.

Without AVX: 66.5 GFlops

And with AVX: 126 GFlops

And this is from a 4-core processor without hyperthreading!

 

[Black96ws6]

I found out something else interesting.

Remember that Dutch PC site?

Well, technically, that page did not get taken down. It just got restricted to non-authorized users! Here, see for yourself:

http://www.pcmweb.nl/hardware/componenten/processoren/fx-8150

But look what shows up at the bottom after the "no access" warning (Click the Specifications tab, not quite in English ?:

Socket
Socket AM3 +
Number of cores
8
Core
Bulldozer
Manufacturing Process
32 nm
Clock frequency (GHz)
3.6
Memory controller
DDR3-1800
Level-2 cache (KB)
8192
Level 3 cache (KB)
8192
64-bit Extensions
yes
Hyperthreading
No
Virtualization
yes
TDP (Watts)
125

That review is technically still there! My guess (just GUESSING here) - that Dutch page gets unrestricted access as soon as the NDA lifts, meaning, again, it might be legit...

 

[Tsavo]

Regardless of whether he was using AVX or not, those results are extremely poor.

Well, that's what I was saying. 40 GFlops w/out AVX is a disaster because that's what a 95-watt Core i5-750 does at stock speeds. BD needs near as makes no difference uses 2x the power and 2 more GHz than a 750 to equal it in this.

BUT. Going to have to wait for real benchmarks before we roll this lifeless corpse into a shallow grave.

 

[bronxzv]

Well, that's what I was saying. 40 GFlops w/out AVX

note that on paper the AVX-256 throughput (peak FLOP/s with 2 fp intensive threads per module) is the same than with SSE on Bulldozer so we can't expect a big gain from AVX (if any) on Bulldozer, unlike on Sandy Bridge where AVX-256 has twice the throughput of SSE / AVX-128

peak FLOP/clock (FP32) with balanced add/mul :

                  SNB core        BD module
 
SSE/AVX-128          8                 8
AVX-256             16                 8
FMA4                 -                16

 

[AdamK47]

Quick! Anyone with a 2600K do the Winrar bench please at stock 3.4GHz.

I know my 2500K @ 4.3GHz is ~3800 KB/s.

I had forgotten WinRAR had a built-in benchmark mode. Here's my 980X @ 4.2GHz.