Deneb just around the corner?

[Aussie FX]

Super pi between AMD and intel should never be compared.
Scientia disassembled superpi around 6 months ago to see why intel could run it so much faster than an otherwise equivalent AMD processor.
It turns out that it uses mainly x87 instructions so it has nothing (or very little) to do with cache at all.
It shouldn't be used as a benchmark comparison and neither should cinebench. They are both toys.
After saying that, those runs that coolaler did showed that Shanghai was a little over 20% faster than Barcelona at the same clock.

qurious63ss Thanks for those links my old memory was recalling the actual launch.
Thank god the old marketing team are now sweeping floors at the local supermarket. (hopefully)

[Eson]

I don't have a home section and I suggest you stop carrying on like AMD is your wife, it is really sad and immature.

First ignorance, then trolling and now personal insult. You will probably find yourself unable to access the forums if you don't back off.

[Chad Boga]

First ignorance, then trolling and now personal humiliation. You will probably find yourself unable to access the forums if you don't back off.

Your interpretation is ludicrous.

Did you read the post I was addressing?

Were you able to comprehend it?

[informal]

You have made some good points but on this one I will have to say you're wrong. This was a quote from Randy Allen back in 2007.


"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent," Allen said.

He used this broad term that got everyone really excited when in reallity barcelona was 40% faster but only in specfp_rate.

http://news.cnet.com/AMD-Go-to-Barce...3-6152645.html

http://www.dailytech.com/article.aspx?newsid=5863

http://www.theregister.co.uk/2007/06...arcelona_2ghz/

oh and of course: http://sharikou.blogspot.com/2007/04...e-running.html

the horses mouth: http://www.youtube.com/watch?v=G_n3wvsfq4Y

BTW I don't think Mr. Allen works for AMD anymore, I wonder why...

Look here for server benchmark results(AMD's highest clocked QC chip works at 2.5Ghz).There is really a variety of workloads where Barcelona beats higher clocked Xeons(specWeb is one,another is VM,Fluent,LS Dyna,Star CD,SPECompMbase2001).
As for Randy Allen comment,please don't talk about things you don't know.He is a senior VP now at AMD.
http://www.amd.com/us-en/Corporate/A..._15089,00.html

[Doctor Pepper]

...retail C1 stepping's HT was 2600 by default...

Forgive my ignorance , but does that imply (which I believe it does...) that also the NorthBridge/L3 will be at 2600 MHz, because that is in my view more important than the HT speed?!

And then the HT 3.1 comes and gives another boost with its 2.8-3.2 GHz HT. In other words, here we have one "sub system" (NB/L3 Cache) that recieve a healthy 50% boost (2GHz -> 3GHz) , which will give a substantial lift on overall performance of the CPU and the whole system. OK, I'm just speculating here, but it looks good so far...

---------------------

First ignorance, then trolling and now personal insult. You will probably find yourself unable to access the forums if you don't back off.

Your interpretation is ludicrous.

Did you read the post I was addressing?

Were you able to comprehend it?

Oh, and please stop the

Instead:

[Eson]

Your interpretation is ludicrous.

Did you read the post I was addressing?

Were you able to comprehend it?

I might not have English as my native language and i had to look up interpretation and ludicrous to be sure of its meaning. However you will not win any Nobel Peace Prize by continuing this sort of pointless discussion (trolling).

And if you had even the slightest thought of making a stupid comment about winning the Nobel Peace Prize was not your intention then it just proves that you are trolling.

I'll stop posting off topic now.

[Gener_AL (UK)]

I'll save everyones breath , Chad Boga, argumentative runt, please off.

[Chad Boga]

I'll save everyones breath , Chad Boga, argumentative runt, please off.

Great contribution.

[Chad Boga]

I might not have English as my native language and i had to look up interpretation and ludicrous to be sure of its meaning. However you will not win any Nobel Peace Prize by continuing this sort of pointless discussion (trolling).

And if you had even the slightest thought of making a stupid comment about winning the Nobel Peace Prize was not your intention then it just proves that you are trolling.

I'll stop posting off topic now.

What did I post that you objected to?

Did you take offence at me thinking Deneb would have only a 10% IPC improvement rather than the 15 to 20% AMD are claiming?

This forum is not AMDZone, it is not meant to be a place where one must be constantly declare their undying love for AMD, as you appear to think should be the case.

[soundood]

soundood: You and I finally agree on something!!

Who woulda thought...

equal and exact opposite?

there must always be both, to justify the others excistence

god i even made up with Nedjo in the last few days, what is the world coming too?

[B.E.E.F.]

What did I post that you objected to?

Did you take offence at me thinking Deneb would have only a 10% IPC improvement rather than the 15 to 20% AMD are claiming?

This forum is not AMDZone, it is not meant to be a place where one must be constantly declare their undying love for AMD, as you appear to think should be the case.

This is for you.



Enough of this petty bullsh*t. Let's get back to facts and figures.

[davidletterboyz]

Can someone please post the link to coolaler's benchmark? I had found the forum but the closest I got so far is only power consumption. Did he post any benchmark?

[informal]

No he didn't.He is under NDA.IIRC he works for MSI as an engineer.

[davidletterboyz]

Thanks for the clarification. No wonder I couldn't find it hehe.

[nanohead]

guys this guy has been saying the same thing for 4 pages now

only reason i can see what he sais is in your quotes, please ignore him, please

This Thread has now been escalated to a higher Risk



for use in future when an Intel Fanboy invades the forum,

Now thats funny.....

[Oliverda]

I don't have a home section and I suggest you stop carrying on like AMD is your wife, it is really sad and immature.

[demonkevy666]

The IPC is one thing, congestion in some ”sub system” is another. I truly believe that AMD have taken the opportunity to optimize cache and memory handling.

Better IPC can give a boost in certain areas (computational intensive), while yielding less in other (heavy data shuffling). I believe AMD have addressed both problem in Deneb.

In other words, even if IPC gets the 15-20% boost (which for certain IS possible), if the rest of the “chain” are unable to handle these are numbers we will only see in synthetic benchmarks *coughIntelcough* and not in our everyday computer experience…

I would like to know if also the AM2+ Deneb are going to have higher NB/HT frequencies, or if this is only for the HT3.1 ready motherboards? In theory the HT2 allows for up to 2600 MHz. I can feel, on my computer, how much better everything flows with NB/HT at 2440 MHz compared to 1800/2000 MHz, so I am certain that this will be an important point!

So, my uneducated guess, supported by some (dubious) benchmarks shown on internet is this:

1. Bigger cache

2. Optimized cache handling

3. (hopefully) Some SSE and possibly INT optimizations

4. Faster NB/HT frequencies

5. All this at lower power draw…

With most of these whishes fulfilled, there should be no problem reaching a clock for clock boost of 12-15% on an average. Weighing in substantial headroom for higher clocks, AMD have in my book a CPU that is equally fast, or faster , than Intels offering for real life computer experience. Maybe not for all synthetic SSE4 optimized benchmarks, but that is another story…



And Chad Boga, I understand your standpoint and you bring up some valid questions. However when a discussion morphs into a “meta discussion” (a discussion about the discussion…) everyone have to try to let that stop after the first round because all that is left after that is name calling (which leads to nothing good…)

thats what I was thinking.

[Bandwidth]

Forgive my ignorance , but does that imply (which I believe it does...) that also the NorthBridge/L3 will be at 2600 MHz, because that is in my view more important than the HT speed?!

And then the HT 3.1 comes and gives another boost with its 2.8-3.2 GHz HT. In other words, here we have one "sub system" (NB/L3 Cache) that recieve a healthy 50% boost (2GHz -> 3GHz) , which will give a substantial lift on overall performance of the CPU and the whole system. OK, I'm just speculating here, but it looks good so far...

I think you are right on the money there. The slow NB in Barcelona was holding it back a bit, I've been hearing AMD will increase the IMC/NB frequency for Deneb, but not sure by how much.

Tests show that for every 200MHz speed increase on the NB speed, you gain about 1-2% performance. If for example we take a Deneb with the NB running at 3GHz, then that alone can give it between 5-10% performance (over the same CPU running the NB @ 2Ghz).

Increasing the L3 cache size can give another 5-10% boost, that's why I won't be surprised to see Deneb perform anywhere between 10-20% faster than a similarly clocked Barcelona. Which puts Deneb's performance in the same league as Intel's Yorkfield.

I guess there are a few other marginal tweaks that AMD did for Deneb (faster L1/L2 cache maybe ), but probably only adding a few percent more to performance.

Can't wait to see the actual numbers now

[Oldguy932]

I don't really think that the L1 needs to be made much faster, the L2 on the other hand was a tad slow. I'm not sure about actual numbers, I will be at home from college this weekend to grab a few screens from the Q9450 system I left at home to do WCG. Both systems have the exact same ram so it should be an ok comparison at least.

Now for higher nb speeds, I used to have a 9850 setup that was on my vapo and was at 2.7ghz for core clock and for nb. In the gimps testing for prime 95, it was as fast as running my 9600 on air at around 2.93ghz with whatever the nb was at, something under 2300mhz I'm pretty sure.

Don't get to riled up in here now boys, a few arguments here and there are always good for a laugh, but its kind of ridiculous in my opinion. Its to the point where its pretty hard to find any information that people do post in the thread.

[Bandwidth]

Here is xbitlabs testing a Phenom X9850, with the NB @1.8GHz, then @2GHz. L3 cache latency went down from 8.5ms to 7.9ms. The average performance gain was about 1% (WinRar 2.8%) for the apps tested, small gain I know, but still better than nothing.

I don't think Deneb will have the NB running at 3GHz, maybe around 2.6GHz.

[Jazzman]

I think you are right on the money there. The slow NB in Barcelona was holding it back a bit, I've been hearing AMD will increase the IMC/NB frequency for Deneb, but not sure by how much.

Tests show that for every 200MHz speed increase on the NB speed, you gain about 1-2% performance. If for example we take a Deneb with the NB running at 3GHz, then that alone can give it between 5-10% performance (over the same CPU running the NB @ 2Ghz).

Increasing the L3 cache size can give another 5-10% boost, that's why I won't be surprised to see Deneb perform anywhere between 10-20% faster than a similarly clocked Barcelona. Which puts Deneb's performance in the same league as Intel's Yorkfield.

I guess there are a few other marginal tweaks that AMD did for Deneb (faster L1/L2 cache maybe ), but probably only adding a few percent more to performance.

Can't wait to see the actual numbers now

I do not presume to fully understand this relationship but I am starting to get a handle on some of it. Below is a quote from Gosh on a thread that is now closed due to a nuclear exhange between Intel advocates and AMD supporters. HERE is the thread and a comment by gosh below.

The following comment by Gosh made sense to me.

When Intel communicates with other hardware it sends data using the FSB. All traffic goes through the FSB. The Latency using the FSB is about 250 clocks. The performance penalty is rather large. If one game is reading memory and/or sends I/O to the GPU using only one thread. Then I don’t think you will see any big differences between AMD and Intel. If the video card is slow you defiantly isn't going see any differences.
Now if the game is using more than one thread the situation changes. If one thread is sending or reading data on Intel, and another thread is sending data to the video card. One of the threads needs to wait. That means that latency goes up. In worst case scenarios it would be double (about 500 clocks).
On AMD this is handled differently. AMD has hypertransport that handles I/O to the video card. If one thread is sending data to the video card it doesn’t compete with memory. On AMD latency is a bit lower too.
Running games on low res or games that isn’t that advanced then this isn’t a problem on Intel. But if you are running a game on high res, have a fast video card (or two). Then this video card is able to handle VERY MUCH I/O. Also if the game is using more memory then more data needs to travel through the FSB on Intel. This situation will lead to more conflicts in the FSB and latency is increased.
Intel is very fast when it isn’t disturbed and is able to use data in the cache. More threads, more memory, more communication with video card and maybe more synchronization between threads. Then performance will shift to favor AMD system design.

Is this why Deneb NB speed translates to better performance than Barcelona?

[Bandwidth]

I do not presume to fully understand this relationship but I am starting to get a handle on some of it. Below is a quote from Gosh on a thread that is now closed due to a nuclear exhange between Intel advocates and AMD supporters. HERE is the thread and a comment by gosh below.

It’s simple, AMD had the memory controller integrated into their CPUs for a long time now, but Intel didn’t. Integrating the memory controller lowers the overall memory latency (improves performance) because the CPU doesn’t have to go to the Northbridge (on the motherboard) to fetch data, the data is right next to it Intel made up for that by having very fast and large L2 caches on their CPUs.

Athlon 64 had the integrated memory controller (IMC) running in sync with the CPU, meaning the same as the the CPU frequency. With Barcelona AMD changed that and added L3 cache to the IMC/NB but made it’s frequency separate from the CPU, and ran it at 1.8-2GHz. They probably had to do that to lower power consumption, at the cost of a slight performance loss.

Is this why Deneb NB speed translates to better performance than Barcelona?

Deneb is similar to Barcelona in regard to it’s NB, but the L3 cache on it is increased from 2MB to 6MB. That’s where most of the performance gain will probably come from for Deneb, though some programs don’t care much for cache and may not gain much. So AMD must have done some other tweaks to Deneb, one could be what I mentioned in my previous post, running the NB frequency higher than the 2GHz that Barcelona has now.

Barcelona had such a small L3 cache, Deneb’s cache is 3x larger, so there is also a possibility it can benefit more from a faster NB/L3.

[informal]

Here is what system integrators and OEMs working with AMD have to say about now already shipping 45nm Shanghai product:

"I can see the AMD percentage of our sales increasing dramatically with Shanghai," said Philip Pokorny, chief hardware architect for Penguin Computing (Fremont, Calif.) an integrator focused on the HPC market. "Before Barcelona our business was 70 percent AMD processors and 30 percent Intel, but there was a reversal and now I am hoping we get to parity," he said, speaking at a panel of integrators gathered here by AMD.

"We were heavily skewed to AMD processors in 75 percent of our business since 2004, but [with AMD's Barcelona slip] Intel has made some recent inroads," said John Lee, vice president of advanced technology solutions at Appro International Inc. (Milpitas, Calif.), an integrator with customers including top supercomputing labs. "Our users are testing Shanghai now and seeing significant performance increases, and Shanghai will be everywhere from general purpose workstations to supercomputing solutions," he added.

"Our users are testing Shanghai now and seeing significant performance increases" coming from actual testers themselves tells us that server product (Shanghai) will indeed live up to its promise.As for Deneb(desktop) i guess we can assume a healthy bump(10-15%) with a much stronger gaming performance ,which combined,could be a winner for average desktop user if priced competitively.Deneb+790GX+Cheap DDR2 1066+HD4870(X2) = all AMD based gaming power rig!

[Gener_AL (UK)]

One thing i would like to see is an Official list of support for older boards.
In the transition of AM2 through to AM2+ and the introduction of phenom
Some AM2 boards gaind + support for the current generations of 65nm variations.

I just hope that when it finally arrives i can drop it straight iin this NF570.

[keithlm]

One thing i would like to see is an Official list of support for older boards.
In the transition of AM2 through to AM2+ and the introduction of phenom
Some AM2 boards gaind + support for the current generations of 65nm variations.

I just hope that when it finally arrives i can drop it straight iin this NF570.

The move from AM2 to AM2+ included a change to the HT, decoupling the HT from the core frequency, adding yet another voltage, and they also added a different version of cool and quiet.

All this meant there were a few important required bios changes to get these AM2+ chips to work on an AM2 motherboard. Many motherboard makers didn't bother updating their old AM2 boards on older EOL motherboards. (This made many people mad at AMD.)

Since the move from 65nm to 45nm basically does not include anything that the socket will "see" as being electrically different then any bios updates should be minor such as changing the CPU's id string.

END RESULT: If an older AM2 board supports the AM2+ 65nm Phenoms then 45nm chips might work on these boards.