AMD Processors
Decrease font size
Increase font size
Topic Title: X2 vs. Opteron 939
Topic Summary:
Created On: 11/05/2005 03:57 PM
Status: Read Only
Linear : Threading : Single : Branch
1 2 3 Next Last unread
Search Topic Search Topic
Topic Tools Topic Tools
View similar topics View similar topics
View topic in raw text format. Print this topic.
 11/05/2005 03:57 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

Ok so I have been discussing this topic with some people for a while now and there seems to be great confusion over the differences of the two processors.

Let us use an example: 2 IDENTICAL 939 systems with a X2 and opteron that are clocked IDENTICALLY and have IDENTICAL amounts of cache.

Which one will perform better?

Someone seems to think the following:

QUOTEIt IS [operton] optimised for massive floating-point-throughput however
He claims the a64 doesn't have these same identical optimizations.

QUOTEIt [operton] really is not optimised for gaming/geometric-through-GPU-specific instruction sets.
Which means the the a64 has the optimizations but the opteron doesn't.

QUOTEAny dual core x2 or ANY FX CPU will beat down an Opteron in gaming metrics, no matter what mobo one uses, compared side by side.
So again, given my example above with 100% otherwise identical systems does his statement hold any truth?

QUOTEAs I explained before - the Opteron is designed for raw floating point performance - NOT geometric processing or BUS optimised instruction pushing to a GPU.


Now from my understanding the opteron is just tested more rigorously. I have also heard (from unreliable sources) that the opteron uses better silicon. Even if both of these were true is there ANY shred of evidence that the a64 is specifically tweaked for gaming while the opteron doesn't have those optimizations but is optimized for FP? I can understand the first two things I touched on simply because a company would gladly pay for better reliability, isn't this the whole point of opteron?

I know other things differ between them such as non-coherent links. But on a basic core level here is the opteron the same as an a64?
 11/05/2005 05:26 PM
User is offline View Users Profile Print this message

Author Icon
Opteron
Senior Member

Posts: 220
Joined: 10/08/2003

Are you trying to make us laugh ? Have you discussed that topics with 10 year olds ? Sorry if that is a real questions, but I thought that everybody should know nowadays, that Opteron S939 are *identical* to its Athlon64 brethren.

The only thruth your statment includes is, that Opterons are picked exspecially for stability/reliability, so the quality tests are somehow stricter. That results in a better overclockability of most Opterons. So you might say (in an odd way) that the Opteron has a "better" silicon.

But at same speed there are NO performance differences whatsoever !

For S939 parts there is no difference with the coherent links either.

I advise to read some Opteron / Athlon64 articels first. Google is your friend /smile.gif' border='0' style='vertical-align:middle' alt='smile.gif' />

byebye

Opteron
 11/05/2005 11:44 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

QUOTEAre you trying to make us laugh ?
No.

QUOTEThe only thruth your statment includes is

Could you be more specific? And it wasn't stating that it *did*, I simply read it on a couple of sites. Even wiki IIRC says the silicon could be different. That is why I said unreliable sources. I asked a lot of questions for myself and on the behalf of the "person" I mentioned earlier and I wanted to cover all of the bases, as in EVEN IF it had "better silicon" would it effect anything performance wise.

QUOTEFor S939 parts there is no difference with the coherent links either

Uh yeah there is. 939 opterons have 3 non-coherent links. 939 athlons have 1. The only reason I mentioned it is so I could get at the heart of the performance questions. I know there are some things that make a opteron different but they are moot when it comes to performance. I just wanted to make it clear I knew what they were.

QUOTEI advise to read some Opteron / Athlon64 articels first. Google is your friend

I have googled quite a bit. I came here because this "person" refuses to believe anything *I* have said. He wants reliable sources, who better than amd? That is why I am here. Google does me no good when the person only wants information from the horse’s mouth.

If anyone can confirm or deny anything please feel free. I apparently need all the sources I can get (mods included if you aren't busy).
 11/06/2005 01:50 AM
User is offline View Users Profile Print this message

Author Icon
functional-pc
Senior Member

Posts: 1558
Joined: 01/10/2005

yes... show us a link to information dictating that 939 opterons have one ht link and not the three.
three would mean higher bandwidth through the I/Os.
unless you can show me that they dont have the traditional three ht links, i will have to stick with believing what i have already learned.

"Although the AMD Opteron 100 Series processors with unbuffered memory support will share the same core technology as the AMD Athlon 64 and AMD Athlon 64 FX processors and will be socket and thermally compatible; there are some distinct differences:

* AMD Opteron 100 Series processors with ECC unbuffered memory all have 1MB of L2 cache.
* AMD Opteron 100 Series processors with ECC unbuffered memory are produced on AMD Opteron processor die material and follow the same AMD Opteron processor manufacturing process as do the 800 Series and 200 Series.
* AMD Opteron 100 Series processors with ECC unbuffered memory undergo the same AMD Opteron processor-level testing and validation as do the 800 Series and 200 Series..."
different die material sounds like its not exactly the same...


-------------------------
Latest workstation:
Lian li PC-A05B, KN3-SLi, 4400 Brisbane, Freezer 64 Pro, 2 gigs Super-Talent T800UX2GC4, 36 raptor system drive, Caviar RE 320 storage drive, Hiper 580, Ati 1300 Pro, running Vista Home Premium 64-bit with Symantec 10.2.224
 11/06/2005 07:12 AM
User is offline View Users Profile Print this message

Author Icon
zir_blazer
Code Warrior

Posts: 2840
Joined: 03/19/2004

The Architecture for both Opterons and Athlons 64 (And derivates like Turions 64 and Socket 754 Semprons) is the K8. How the Opteron may have higher FPU performance than the Athlon 64 if they're both using the same Execution Unit (We aren't comparing a Pentium M against a Pentium 4)? The diferent presentations of the K8 Architecture may got slighty extra features depending on the market that it is aimed to (For example, Athlons 64 FX is aimed to enthusiast, and because most of them are overclockers, it favours them with an Unlocked Multiplier). So, contrary to what most people says, the Athlon 64/64 FX aren't "Designed for gaming" because not only that all those Processors uses the same Execution Unit, but more importantly because the K8 Architecture was made with everything that would be needed to have great scalability in Server enviroment in mind (That is what Hyper Transport, x86-64 and the Integrated Memory Controller are for), and it was the Opteron the first K8 based Processor to reach the market while the Athlon 64 arrived months later with the same identical architecture that was made and selled for Servers, so if anything, games just liked the Server oriented architecture's new features. You could enjoy practicaly the same performance with Socket 940 Opterons before Athlons 64 were released (They are slighty slower because the platform (Chipset and Buffered Memory Modules and probabily with ECC capabilities too) is slower, not they by themselves).
I can't exactly tell you if Socket 939 Opterons with either San Diego or Toledo Cores are the ones of higher qualities that comes from a Wafer or got its own, dedicated production line that from start uses higher quality silicon, but because the Core technically is the same, the only good resume that you can make about it is that the San Diego Cores used in Socket 939 Opterons seems to be of higher quality than the average ones, what seems to be accurate though is the fact that they are more heavily tested.
On the other hand, all Socket 754 and Socket 939 got only one Uncoherent Hyper Transport Link that is used to communicate with the Northbridge. Why would someone want an Opteron 1xx with three Uncoherent HTLs? Its makes little sense that you had three Northbridges to take care with a single Processor. Socket 940 Opterons 2xx and 8xx got Multiprocessor support and that is when the extra Hyper Transport Links becomes useful instead of giving extra Bandwidth that you will never need. Opterons 2xx got an extra Coherent Hyper Transport Link and Opterons 8xx got a total of three Coherent HTLs that are used for communicate with other Processors in the system. Having a Opteron 2xx in an Uniprocessor configuration won't have any advantage over an old Socket 940 Opteron 1xx because that extra Coherent HTL isn't connecting it with anything, with the exeption that you can later add another identical Opteron 2xx and get instantly a Dual Processor system.
 11/06/2005 08:06 AM
User is offline View Users Profile Print this message

Author Icon
Opteron
Senior Member

Posts: 220
Joined: 10/08/2003

QUOTE(atomicFan @ Nov 5 2005, 07:44 PM)(...)
Uh yeah there is.  939 opterons have 3 non-coherent links.  939 athlons have 1.  The only reason I mentioned it is so I could get at the heart of the performance questions.  I know there are some things that make a opteron different but they are moot when it comes to performance.  I just wanted to make it clear I knew what they were.
I have googled quite a bit.  I came here because this "person" refuses to believe anything *I* have said.  He wants reliable sources, who better than amd?  That is why I am here.  Google does me no good when the person only wants information from the horse’s mouth.

Oh man, that's going to be fun. S939 Opterons have 3non-coherent links... great stuff.
Ok but seriously, first I was not sure if you are trying to kid with us, but that statment ... looks like you are really just very, very uninformed. So I am trying to help.

Luckily zir_blazer already mentioned all important facts, there are no Opterons or Athlon64s there is just the one and only K8 Core.
For this K8 Core, 3 different sockets exists

S940: dual channel, registered memory interface, ECC Optional.
Three HTr Links, (Zero coherent links there for the old 100 Opterons, 1 link is coherent for the 200 Series and all 3 are coherent for the 800 Series)
Marketing names: Opteron and partially AthlonFX
Old code name: Sledgehammer

S939: dual channel, no need for registered memory, *ONE* 16bit HTr Link.
Marketing names: Opteron 100, Athlon64, AthlonFX, Athlon64 X2, soon Sempron

S754: only single channel meomry interface, 2x8bit HTr Links, most often used as one 16bit link (reason for the 2x8 bit links was, that AMD intended first (around 2000) the S754 as Opteron 200 platform.
Marketing names: Athlon64, Sempron, Turion
Old code name: Clawhammer
So once again, with S939 it does not matter if the CPU is called Athlon or Opteron or whatsoever. AMD is a little bit lazy with updating its material, but you should understand, that the Opteron and Athlon64s have the same Pin out, hence you can plug in an Opteron in every S939 mainbaord. You maybe get some PowerNow! problems, but it will run.
So, it is clear, that it does not feature another pin layout with 3 Links, because then Opterons wont work.
The proof that S939 is 1 16bit link is here:

http://www.amd.com/us-en/Proce...chni...39_7203,00.html' ">http://www.amd.com/us-en/Proce...0_182_739_7203,00.html

There are a lot of documents (Wonder if you did not yet found it). What should be of interest for you is the

AMD Functional Data Sheet, 939-Pin Package and
AMD Functional Data Sheet, 940-Pin Package

just guess, why there is no references to Athlon64 or Opteron ... I leave the math to you /wink.gif' border='0' style='vertical-align:middle' alt='wink.gif' />

Also the Revision Guide for AMD Athlon™ 64 and AMD Opteron™ Processors might be of the interest for you. There on p.7/8 you can find all revisions of the K8 core and its different CPU-ID values for the various flavours of it, like Athlon64, Sempron, Opteron, etc ....

I hope that is enough proof, that there is just one core ...

To the "better" silicon issue. There is no official proof of that. Only "proof" is that all new S939 are very good overclockers. Again, as long as you do not "exploit" these advantage, e.g. by overclocking and/or lowering the vcore the performance is not different from any other K8 CPU with the same specifications (MHz, cache, RAM speed or interface)

I hope all your questions are answered now, if not please try to read the documentations first, e.g. the "BIOS and Kernel Developer's Guide for AMD Athlon™ 64 and AMD Opteron™ Processors".

Yes a lot to read, but you wanted proof, so so dont bother /wink.gif' border='0' style='vertical-align:middle' alt='wink.gif' />

byebye

Opteron
 11/06/2005 02:39 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

You guys are doing good.

About the 3 non-coherent links. I emailed amd already about this issue. Quite a few tech sites believed that they did indeed have 3 non-coherent links like their 940 brothers. In fact before you posted I hadn't heard anything to the contrary. Amd tech support seems to agree with those sites (not with me because well I wasn't sure to begin with) that they have 3 non-coherent links.

That isn't to say you are wrong. AMD tech support could have gotten it mixed up or confused and thought I was talking about 940 opterons (although I did specifically say 939).

I looked over the documents you linked to but I am not really seeing what you mentioned. They do talk about coherent links but the pinout confuses me so yeah..
 11/06/2005 04:36 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

http://www.amd.com/us-en/Proce...oduc...ml?redir=CPOS14' ">http://www.amd.com/us-en/Proce...0,00.html?redir=CPOS14

It says it has 3 non-coherent for the 1xx dual core solutions. Then once you head over to:

http://www.amdcompare.com/us%2Den/opteron/Default.aspx' ">http://www.amdcompare.com/us%2Den/opteron/Default.aspx

Click the drop down box for "Processor" and select "Dual-core opteron processor". Scroll down and you will see the entire 1xx dual core opteron line listed as 939. So there isn’t any confusion on what they reference in my first link, they are socket 939 and have 3 non-coherent links.

QUOTE(Opteron)Oh man, that's going to be fun. S939 Opterons have 3non-coherent links... great stuff.
Ok but seriously, first I was not sure if you are trying to kid with us, but that statment ... looks like you are really just very, very uninformed. So I am trying to help.
Who is the very, very uninformed poster now?
 11/06/2005 04:50 PM
User is offline View Users Profile Print this message

Author Icon
functional-pc
Senior Member

Posts: 1558
Joined: 01/10/2005

also... if they do in fact retain the three links, they are not usless as eluded to above. in multi- socket applications, yes the links get utilized with smp communication. but it is a known fact that in 1xx opterons, the 3 links are utilized improving I/O communication... very important.

-------------------------
Latest workstation:
Lian li PC-A05B, KN3-SLi, 4400 Brisbane, Freezer 64 Pro, 2 gigs Super-Talent T800UX2GC4, 36 raptor system drive, Caviar RE 320 storage drive, Hiper 580, Ati 1300 Pro, running Vista Home Premium 64-bit with Symantec 10.2.224
 11/06/2005 05:06 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

Actually they are useless unless you take advantage of them. Nvidia's nforce pro for example can hang two chips off of a single processor which leads to things such as true 16x sli. The 2200 hangs off of one while the 2050 bypasses the 2200 and is connected directly to the processor.

Coherent links are the ones that get utilized for multi socket configurations. Coherent links are used for tasks specific to cache coherency between different processors.
 11/06/2005 08:43 PM
User is offline View Users Profile Print this message

Author Icon
Opteron
Senior Member

Posts: 220
Joined: 10/08/2003

QUOTE(atomicFan @ Nov 6 2005, 01:36 PM)Who is the very, very uninformed poster now?
[right][snapback]539706[/snapback][/right]


Sorry to tell you, but it is still you.
But you got some credit, because that AMD webpage, you have posted, is inconsistent, so not your fault.

You are right, that there are no S940 dual core Opterons 100 (I know that quiet well, I guess I own the only dual core compatible single nforce150 S940 mainboard, maybe I will buy a 265 later), but there are neither single core S939 Opteron 100 HE or EE, 140 or 142, that are mentioned on the same webpage.

Conclusion: They are mixing up S939 and S940 on that page.

Just try to use your brain and do not look for fuzzy documentations.

HTr Links needs Pins. How much can be looked up on a whitepaper at hypertransport.org, I look it up once 2 years ago, I do not have the link now.
Anyways, it should be clear, that you need some pins ... /rolleyes.gif' border='0' style='vertical-align:middle' alt='rolleyes.gif' />

S939 has one single 16 bit HTr link.

If you now state, that the S939 Opteron has 3 HTr Links I ask you, where are the pins for the other two links ? Even if they would be there somehow, the pin layout would be different from the Athlon64 CPUs. If so, do you then think that such an Opteron S939 CPU with a different pin layout would be compatible with the current S939 mainboards ?

Come on ...

A lot of HTr have furthermore no purpose for single CPU Systems. There was not any S940 mainboard that even used the 2nd HTr link, not to mention all 3.
Nowadays it would be a little bit more interesting because of your mentioned NVidia SLi stuff, but again, that does not matter, because of HTr's Tunnel function. So you just chain the 2 chips together in a row, the "Southbridge" to at the end, the CPU on the top no problem, works very well, and gives you a clean mainboard design with short traces. The result is the A8N32 SLI:



Good luck, finding the truth /smile.gif' border='0' style='vertical-align:middle' alt='smile.gif' />

Opteron

P.S: Good job with your last comment, that the 3 links are useless if you dont use them /smile.gif' border='0' style='vertical-align:middle' alt='smile.gif' />
 11/06/2005 10:44 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

To which link are you referring to? The first link doesn't mention what socket they are using and makes no mention of the 142EE/HE. On the second page all EE/HE processors are listed as 940's and again makes no mention of the 142EE/HE, only the 140EE.
 11/07/2005 09:11 AM
User is offline View Users Profile Print this message

Author Icon
Opteron
Senior Member

Posts: 220
Joined: 10/08/2003

QUOTE(atomicFan @ Nov 6 2005, 07:44 PM)To which link are you referring to?  The first link doesn't mention what socket they are using and makes no mention of the 142EE/HE.  On the second page all EE/HE processors are listed as 940's and again makes no mention of the 142EE/HE, only the 140EE.
[right][snapback]540005[/snapback][/right]


First one:

http://www.amd.com/us-en/Proce...oduc...ml?redir=CPOS14' ">http://www.amd.com/us-en/Proce...0,00.html?redir=CPOS14

AMD lists there 140, 142 and several EE and HE Opterons. All of them are not available for S939. Well maybe AMD is introducing HE / EE types for S939 in the future, too, but anyways there wont be any 140 or 142, that are mentioned on the page. Still as you found out correctly all He/EE CPUs are S940 CPUs according to your 2nd source, so it is clear that the page mentioned above mixes S939 and S940 Opteron CPUs, hence the confusion about the Hypertransport declaration "3/0" which is only valid for S940 CPUs.

Everything is sooo clear, you still do not want to draw the obvious conclusion, do u ?

byebye

Opteron

P.S: I guess there was a misunderstanding, I havent meant a specific "142EE" Opteron, but the whole HE/EE family and normal 140/142 Opterons.




 11/07/2005 01:24 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

Of course the first page is mixing the sockets, there are no dual core 1xx opterons for socket 940. That doesn’t mean they mislabeled the non-coherent links. Nothing in that chart suggests they mislabeled the dual core 939's when it came to non-coherent links (or mislabeled anything else for that matter). Not by the chart itself anyways.

Again, that isn't to say you are wrong. I contacted amd once again (by phone) and they assured me they are using 3 non-coherent links on the 1xx socket 939 opterons. I even brought up your point about them using a different pin out, he asked his boss this time and came back letting me know that they do indeed use 3 non-coherent links.

Anyone that works at amd want to clear this up? I am not discrediting the white documents you linked to (thanks for them by the way), I just need some assurance.

Although the tech support guy did say the opterons were the same exact processor as their athlon 64 brothers. So I guess the call wasn't completely useless, this should provide my college some more proof.
 11/07/2005 02:07 PM
User is offline View Users Profile Print this message

Author Icon
J-trader
Member

Posts: 50
Joined: 08/14/2004

Interesting stuff guys, thks.

But just tell me, is the opteron 165 more bang for the buck then the X2 3800?

Which one has more OC'ing potential?
 11/07/2005 03:01 PM
User is offline View Users Profile Print this message

Author Icon
functional-pc
Senior Member

Posts: 1558
Joined: 01/10/2005

QUOTE(J-trader @ Nov 7 2005, 11:07 AM)Interesting stuff guys, thks.

But just tell me, is the opteron 165 more bang for the buck then the X2 3800?

Which one has more OC'ing potential?
[right][snapback]540383[/snapback][/right]


yes it is...
twice the cache will be noticable... and just the fact that the denmark core is more refined is reason enough to go that way. the three 'non-coherent' links are supposed to greatly improve all I/O traffic... including DDR, PCI-X... everything that matters.

-------------------------
Latest workstation:
Lian li PC-A05B, KN3-SLi, 4400 Brisbane, Freezer 64 Pro, 2 gigs Super-Talent T800UX2GC4, 36 raptor system drive, Caviar RE 320 storage drive, Hiper 580, Ati 1300 Pro, running Vista Home Premium 64-bit with Symantec 10.2.224
 11/07/2005 09:15 PM
User is offline View Users Profile Print this message

Author Icon
atomicFan
Junior Member

Posts: 8
Joined: 11/05/2005

If the 939 opterons even have the non-coherent links they are going to lay dormant until he uses the nforce pro 939 version chipset or something similar. Think of it as dual channel ddr, if you are only using one stick the performance increase is non existent.
 11/07/2005 09:46 PM
User is offline View Users Profile Print this message

Author Icon
Bitey
Elite

Posts: 1492
Joined: 10/07/2003

Simple fact for most situations a single HT link to your chipset for IO has way more bandwidth than is required be most computers.

I have not heard of a single motherboard that uses more than 1 HT link to connect to the chipset/mobo except some dual tyan boards which is use 1 HT link on each CPU to to connect to the chipsets..still one per cpu.

just has not happend yet folks, though please let me know when it has /wink.gif' border='0' style='vertical-align:middle' alt='wink.gif' />
 11/08/2005 09:32 AM
User is offline View Users Profile Print this message

Author Icon
iq100
Junior Member

Posts: 9
Joined: 09/12/2005

QUOTE(functional-pc @ Nov 6 2005, 01:50 AM)    ... * AMD Opteron 100 Series processors with ECC unbuffered memory all have 1MB of L2 cache. ...
    [right][snapback]539429[/snapback][/right]



Can I use unregisterd ECC memory.
My motherboard is the AsROCK939dualSata2.

One guy said that since the memory controller is inside the Opt 165, and since the memory controller supports unreg-ECC, that I should be able to use ECC.

But another guy said that ECC needs to be turned on in the BIOS, and no AsROCK BIOS does this??

Yet another guy said use A64Tweaker to turn on ECC, after Windows starts?

But another guy said that when ECC is turned on ALL memory is cleared, so system will crash?

What do you think?
I would like to use XP-Professional and the AsRock939dual and Opt 165 with non-reg ECC?

Anyplace to get the definitive answer?

Hope that people (and AMD tech) can and will respond.
Thanks,
 11/19/2005 03:26 PM
User is offline View Users Profile Print this message

Author Icon
1R4dD4d
Junior Member

Posts: 1
Joined: 11/19/2005

I also am in the market for a dual core AMD processor, for my Home Theatre, Multi-Media PC, I recently built. My system is currently a AMD 64 3000+ running at HTT4 250 fsb, 2x 512Mb Corsair Pro pc 3200 LL ram, (Timings 2.5-3-3-7. 1T command rate), on an MSI K8N Neo 4 Plat SLI board, Latest Bios, I am mainly wanting to know the differences between the X2 series & the opteron 170 & if the differences in FSB make a difference i.e. X2 64=2000 FSB while Opteron 170=1000 is there a performance difference or is it the same fsb just listed different? Also does the Opteron 170 also support SSE3 & the No Execute extensions as well?

Thanks in Advance=)
Statistics
112018 users are registered to the AMD Processors forum.
There are currently 0 users logged in.

FuseTalk Hosting Executive Plan v3.2 - © 1999-2014 FuseTalk Inc. All rights reserved.



Contact AMD Terms and Conditions ©2007 Advanced Micro Devices, Inc. Privacy Trademark information