Something that everyone can agree on

[MKruer]

In the Spirit of bringing the community back together after the on going ideological fights. Here is something that everyone can agree on. This is Anandtech latest and greatest, http://www.anandtech.com/weblog/showpost.aspx?i=528 I think everyone will agree that what you just read on Anandtech was a complete waste of time and lamest excuse for a review. Read the comments they are hilarious, I even added my 2 cents.

[DragonSlaya04]

Not a good sign for Intel. Using registered DIMMs on both platforms would make performance about equal. I'd like to see if Turbo mode was enabled or not.

[Karamiekos]

Yeah I read that....it's so sad, I don't even know why I let them live in my Google Reader....Anandtech used to be cool in my book. They have really taken a turn for the crapper.

[hyc]

No way, Anandtech is the paragon of perfection!



forget it, I can't play devil's advocate... I commented there too. Using Intel's MKL on Shanghai is pretty rank. Especially since Shanghai has new instructions that probably aren't supported in MKL (but *are* supported in AMD's current ACML...)

[carniver]

There's a mention of LAPACK. Someone should do benchmarks with that and see how it runs on an i7

[abinstein]

The problem is while AMD can give Shanghai to Anand for free, Intel can give ads money that Anand lives on. So the more AMD offers the more harm it will do to itself. Most of these websites, including Anandtech, are marketing tools of Intel, even though a large part of the world don't understand it.

In fact, if we ask those who don't take marketing money from Intel not to post on a forum, we'll see a lot less trolls posting here and elsewhere on the net.

[abinstein]

By the way, the test Anand did was total junk. Using the same Linpack binary for both systems is pointless. On thing about Linpack is that you want to use a binary optimized for your system. Besides, Lapack would be much better performance indicator for not-too-ancient workloads.

[Karamiekos]

Marketing tools....???

More Like Marketing Fools!

[Das Tipitz]

Looks like good results for Shanghai to me. I don't know why you have a problem with the tests.

It's disappointing that HT doesn't help i7 in LinPack but for AMD it is good news that they are very close to IPC parity with Shanghai vs. Nehalem in this one benchmark.

[AzmountAryl]

Looks like good results for Shanghai to me. I don't know why you have a problem with the tests.
It's disappointing that HT doesn't help i7 in LinPack but for AMD it is good news that they are very close to IPC parity with Shanghai vs. Nehalem in this one benchmark.

Why are you trying to distort the truth? If Anand wasn't cheating, a 2.7GHz Shanghai would have been ahead of IntEl's 2.8GHz chip in this test.
Paid IntEl viral advertisers like you should be banned from here.

[MKruer]

I come away with two things from this not-a-review
1. The legacy of Hyper Threading performance hit continues and it buyer beware as to whether it will help or hurt.
2. Judging by the jump that LINPACK had with 8384 going from 533 to 800 (a 266mhz difference) It looks like in the case of LINPACK memory bandwidth is a huge factor.
When the Opterons go from DDR2 to DDR3 I would expect that we would see roughly the same performance jump as there was between the DDR2@533 and DDR2@800. Guesstimating going from the DDR2@800 and using the same jump in Performance the Opteron 8384 using DDR3@1066 should bounce the score up to 35.7, and going to DDR3@1333 should bump the score up to 37.9 (higher then the Core i7 @ 2.66 GHz) Giving the Opteron 8384 about a 2% lean in IPC over the core, at least when it comes to this LINPACK test and potentialy more if AMD goes to quad channel at the same time.

[JulianL]

I take away two things from this as well:

1) Given the uneven memory configs (DDR3 vs DDR2) and the use of Intel-optimised code then I'm really encouraged by Shanghai raw performance. WIth the AMD-optimised libraries and DDR3 memory controllers I suspect it would have beaten i7 on LINPACK (which is still a very, very widely used measure of performance in the HPC world so is a hugely important benchmark for anyone wanting to play in the HPC market).

2) Don't forget that the 8384 has an ACP of 75W (approx TDP 95W) vs Core i7 130W TDP. Unfortunately AnandTech didn't see fit to post power consumption data for these tests but, based on the TDPs of the respective chips, I suspect that Shanghai would have trashed the Core i7 in performance per watt, even with more power-hungry DDR2.

So yes, a very lame attempt by AnandTech to do a meaningful performance comparison but ultimately quite encouraging for AMD in terms of how they did perform.

- Julian

[Das Tipitz]

Looks like good results for Shanghai to me. I don't know why you have a problem with the tests.
It's disappointing that HT doesn't help i7 in LinPack but for AMD it is good news that they are very close to IPC parity with Shanghai vs. Nehalem in this one benchmark.

Why are you trying to distort the truth? If Anand wasn't cheating, a 2.7GHz Shanghai would have been ahead of IntEl's 2.8GHz chip in this test.
Paid IntEl viral advertisers like you should be banned from here.

Intel's 2.66GHz chip.

RE: question about clock speed by JohanAnandtech, 5 hours ago

LINPACK is extremely well threaded. That means that all cores are used to their full potential. But to be sure we disabled turbo booster, as we speculate it will probably not be used on the server products. That is why I labeled the Core i7 at it's clockspeed. (and also because I absolutely hate Intel's and AMD's numbering systems)

Guesstimating going from the DDR2@800 and using the same jump in Performance the Opteron 8384 using DDR3@1066 should bounce the score up to 35.7, and going to DDR3@1333 should bump the score up to 37.9 (higher then the Core i7 @ 2.66 GHz) Giving the Opteron 8384 about a 2% lean in IPC over the core, at least when it comes to this LINPACK test and potentialy more if AMD goes to quad channel at the same time.

Additional memory bandwidth might improve performance at a substantial rate for a while, but not forever. I don't know when a 2.7GHz Shanghai will become saturated with bandwidth though so it is hard to predict how faster memory will influence performance. Certainly you see that with Nehalem in most benches, where dual vs. tri-channel makes little to no difference and a lot of the time even single channel delivers 90% of tri-channel performance.

Don't forget that the 8384 has an ACP of 75W (approx TDP 95W) vs Core i7 130W TDP. Unfortunately AnandTech didn't see fit to post power consumption data for these tests but, based on the TDPs of the respective chips, I suspect that Shanghai would have trashed the Core i7 in performance per watt, even with more power-hungry DDR2.

I wouldn't be surprised if the power consumption of these two systems would have been very similar, but when we actually see Nehalem Xeon servers vs. Shanghai servers then you will probably be right, in this benchmark Shanghai will win @ perf/watt. Testing the desktop Core i7 platform doesn't reveal the huge disadvantage of the power consumption of the FB-DIMMs that Intel has on its server platforms.

[AzmountAryl]

Intel's 2.66GHz chip.

RE: question about clock speed by JohanAnandtech, 5 hours ago
LINPACK is extremely well threaded. That means that all cores are used to their full potential. But to be sure we disabled turbo booster, as we speculate it will probably not be used on the server products. That is why I labeled the Core i7 at it's clockspeed. (and also because I absolutely hate Intel's and AMD's numbering systems)

Anand has been taking money from IntEl to make their product look good and they made bad reviews of AMD product on purpose. Why should I trust something they say they did?

Did I forget how they tried to claim that GeForce 280 is better than Radeon X2 4870 even thou Radeon was as much as 50-80% faster and was cheaper?

[Das Tipitz]

Intel's 2.66GHz chip.

RE: question about clock speed by JohanAnandtech, 5 hours ago
LINPACK is extremely well threaded. That means that all cores are used to their full potential. But to be sure we disabled turbo booster, as we speculate it will probably not be used on the server products. That is why I labeled the Core i7 at it's clockspeed. (and also because I absolutely hate Intel's and AMD's numbering systems)

Anand has been taking money from IntEl to make their product look good and they made bad reviews of AMD product on purpose. Why should I trust something they say they did?

Did I forget how they tried to claim that GeForce 280 is better than Radeon X2 4870 even thou Radeon was as much as 50-80% faster and was cheaper?

I don't remember them saying anything about the GTX 280 being better than the 4870 X2. The reviewer at Anandtech was a bit disappointed simply that AMD didn't do more to make a dual GPU card feel more like a single-GPU card and implement changes in hardware to make the 4870 X2 more interesting than just 2 x 4870 1GB on a single PCB. For example RV770's Sideport was thrown out of the boards and not utilized. The weakness of the 4870 X2 is that it relies on Crossfire scaling in the drivers, which is not always good or consistant. It's still the fastest card out there and great if you don't want multiple cards, but it simply had more potential. The HD 4870 X2 isn't cheaper than the GTX 280 though, it's more like $150 more expensive (~$400 vs. ~$550).

And if Anandtech wanted to make Intel look good, why would they publish the results of these benchmarks? They don't look particularly encouraging for Intel IMO at least, given AMD is running on a slower memory system and is still managing to acheive close to IPC parity with Nehalem.

[AussieFX]

I wouldn't be surprised if the power consumption of these two systems would have been very similar, but when we actually see Nehalem Xeon servers vs. Shanghai servers then you will probably be right, in this benchmark Shanghai will win @ perf/watt. Testing the desktop Core i7 platform doesn't reveal the huge disadvantage of the power consumption of the FB-DIMMs that Intel has on its server platforms.

Nehalem server will use even more power than the desktop so that statement is ridiculous.

We al know how efficient Shanghai is and nehalem with FB Dimms will be laughable.

There is no contest. AMD is miles in front in perf/watt.
But you sure won't see that in anands benchmarks. Hell he can't even review it. That brings me to the so called bios problem. My DFI can slot a Deneb right in, no updates necessary so I don't believe that Beta crap. Just another excuse from the shill site.

[AzmountAryl]

I don't remember them saying anything about the GTX 280 being better than the 4870 X2. The reviewer at Anandtech was a bit disappointed simply that AMD didn't do more to make a dual GPU card feel more like a single-GPU card and implement changes in hardware to make the 4870 X2 more interesting than just 2 x 4870 1GB on a single PCB. For example RV770's Sideport was thrown out of the boards and not utilized. The weakness of the 4870 X2 is that it relies on Crossfire scaling in the drivers, which is not always good or consistant. It's still the fastest card out there and great if you don't want multiple cards, but it simply had more potential. The HD 4870 X2 isn't cheaper than the GTX 280 though, it's more like $150 more expensive (~$400 vs. ~$550).

And if Anandtech wanted to make Intel look good, why would they publish the results of these benchmarks? They don't look particularly encouraging for Intel IMO at least, given AMD is running on a slower memory system and is still managing to acheive close to IPC parity with Nehalem.

You found a wrong place to defend Anand. This days even his own forums are against him.

[AussieFX]

I don't remember them saying anything about the GTX 280 being better than the 4870 X2. The reviewer at Anandtech was a bit disappointed simply that AMD didn't do more to make a dual GPU card feel more like a single-GPU card and implement changes in hardware to make the 4870 X2 more interesting than just 2 x 4870 1GB on a single PCB. For example RV770's Sideport was thrown out of the boards and not utilized. The weakness of the 4870 X2 is that it relies on Crossfire scaling in the drivers, which is not always good or consistant. It's still the fastest card out there and great if you don't want multiple cards, but it simply had more potential. The HD 4870 X2 isn't cheaper than the GTX 280 though, it's more like $150 more expensive (~$400 vs. ~$550).

And if Anandtech wanted to make Intel look good, why would they publish the results of these benchmarks? They don't look particularly encouraging for Intel IMO at least, given AMD is running on a slower memory system and is still managing to acheive close to IPC parity with Nehalem.

I don't know where you get your information from, I can buy a 4870x2 for ~$100 cheaper than a 280.

Anand published a whole quarter page with ONE benchmark. Because he had to post something... He's been sitting on that chip for probably a month or more now and that's all he can produce! Do you think that's ok while he churns out nehalem reviews one after the other? He's a shill

[piesquared]

And if there was ever any doubt, the fact that they have an Intel employee as a moderator, should seal the deal.

[The_Ghost]

some days i just love a thread like this, and the link to anands where they have to defend them selves from their own posters

i guess with the economy going like it is, their reviews might even get worse

[T800-101]

Anandtech is not even doing reviews anymore. They are nothing more than veiled adverts, it has become so painfully obvious that even the hardcore Anandtech users are questioning what is going on.

What I find even more despicable is they avoid doing reviews altogether if the results are not going to line up with their sponsors. Total garbage that site is.

[scientia]

I would have to agree. Using unbuffered memory on Nehalem and using the Intel library for Shanghai were unprofessional. It is interesting too that they completely avoided any details of cooling. Linpack has good heating (similar to and maybe worse than Prime95) so this would be an aspect of the test.

[MKruer]

Well review 2 is out, and this one is marginally relivent. The Opteron 8384 (2.7) can beat the Xeon 5472 (3 GHz) if both are using the same binary
http://www.anandtech.com/weblog/showpost.aspx?i=529

[T800-101]

^^ I really believe Anandtech's "rule" is to always show Intel at the top of the graphs. Bottom line is I don't trust a single result from them, I suspect everything they publish. That doesn't mean it is all incorrect or skewed, but they cannot be trusted in my view, and all their results must be treated with skepticism.

[carniver]

Well review 2 is out, and this one is marginally relivent. The Opteron 8384 (2.7) can beat the Xeon 5472 (3 GHz) if both are using the same binary
http://www.anandtech.com/weblog/showpost.aspx?i=529

The results with HT on are, for whatever reasons, missing