Please register or login. There are 0 registered and 1165 anonymous users currently online. Current bandwidth usage: 326.30 kbit/s December 15 - 10:37pm EST 
Hardware Analysis
      
Forums Product Prices
  Contents 
 
 

  Latest Topics 
 

More >>
 

    
 
 

  You Are Here: 
 
/ Forums / AMD's AM2 socket, lowering performance?
 

  Re: AMD’s M2 socket, lowering performance? 
 
 Author 
 Date Written 
 Tools 
Continue Reading on Page: 1, 2, 3, 4, Next >>
RI Redneck Mar 14, 2006, 11:31am EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List Replies: 74 - Views: 6092
"Early reports of the performance of CPUs sporting this new socket indicate that their performance is quite a bit less than what we’re used to from a similarly configured socket-939. This is surprising since DDR2 provides more bandwidth, so should in theory also be able to offer better performance. However some report a reduction in performance by as much as 50%, that’s no small margin to be honest."

I would greatly appreciate a link to this "Early Report" you are referring to. The ONLY review/report, that contains actual benchmarks, available on the web is by Tom's Hardware ( http://www.tomshardware.com/2006/02/21/a_look_at_amds_socket_am2_platform/ ). It shows the AM2 platform performing fairly evenly when equipped with high latency DDR2. DDR2 latency has been an issue since it first arrived on the market and it can and does affect performance when compared to low latency DDR (even at slower speeds). However, I've NEVER seen it cause a 50% performance drop with ANY processor. Considering the volatility of this market, I would appreciate unbiased numbers to back up any reports whether positive or negative.

Thanks,

BT


Want to enjoy fewer advertisements and more features? Click here to become a Hardware Analysis registered user.
Sander Sassen Mar 14, 2006, 11:37am EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
Hi Brian,

Of course I'm able to back these up, although most of these come from direct communication with my peers, I think only Anand bothered to write up his experience.

http://www.anandtech.com/weblog/default.aspx?bid=279

Best regards,

Sander Sassen
Editor in Chief - Hardware Analysis
ssassen@hardwareanalysis.com
Mathew Everett Mar 14, 2006, 12:04pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List

Edited: Mar 14, 2006, 12:10pm EST

 
>> Re: Re: AMD’s M2 socket, lowering performance?
SS has no lies in his pockets :)

A bite from Anandtech
---------------------------------------------------------------------
Needless to say my excitement vanished after I ran the first performance numbers and it offered about half the memory bandwidth of an average Socket-939 platform. Remember that the major change with AM2 is the migration to DDR2-667 (and DDR2-800) over DDR-400, so memory bandwidth should go up - not be cut in half.

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

I think the AM2 is going to have a few suprises up it's sleeve, I also think AMD will jump to high cache levels on their chips as I think they could benifit from that. "we have seen early reports of this also around the net" Maybe AMD has their Quad core chips already... kind of like when they pushed dual core out first :) it is all up in the air as of now & AMD has a few secrets we don't know about... But I will have to say I am very happy with my 4800+ X2 & my FX 57 and see no reason to jump ship on the 939 for atleast a year or so. Unless of course Quad core is rolled out by AMD...

Thanks

Mat Everett
http://www.bfmetournaments.com
http://www.bf2tournaments.com


Beavis Khan Mar 14, 2006, 12:18pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List

Edited: Mar 14, 2006, 12:19pm EST

 
>> Re: Re: AMD’s M2 socket, lowering performance?
The difference in performance between 512KB and 1MB level 2 cache seems to be, at best, 10% or so, and more typically 2-5% (Edit - on 939, to be specific. I suppose it's possible M2 is different in this regard...). Throwing more cache at new processors isn't going to solve the problem if M2 ships with the issues Sander and others are describing...

____
"For every complex problem, there is a solution that is simple, neat, and wrong."

- H.L. Mencken
Lou Bot Mar 14, 2006, 12:37pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
This may not be comparing apples to apples but the results of an intel mobo w/ DDR2 memory was tested against an Intel mobo w/DDR1. The results were sub-par.

MSI K8N NEO4 SLI(MS7100) BIOS3.A
AMD Opteron185 @2.86
2GB PC4000 2.5-3-3-8 2T
2xBFG 7800GTX 256MB (91.31)
2xRaptor74RAID0
Enermax EG651P-VE 24P & CS-718
"Thee BEST chip is a Dorrito Chip
Wayne Bradford Mar 14, 2006, 02:07pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
I have always thought high latencies coupled with processors containing on board memory controllers to be counter productive. Obviously off chip controllers (north bridge) create latency based on design and location, so the only way AMD can overcome this inherent problem is to increase the on die L2, however I feel this alone is not enough to allow the performance necessary to overcome the loss. Like it or not we are moving toward DDR-2 and AMD is on board with this. I have for some time been curious as to how they intend to accomplish this feat.
WB

Radomir Jordanovic Mar 14, 2006, 02:10pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
The GeForce 7900GTX has a memory bandwidth of over 50GB/s with its 256-bit DDR3 ram. I'm guessing it's the "quad channel" memory and speed combined that provide this amazing transfer rate. When are we going to have such bandwidth for desktops?

Silverstone 750W, Asus P5KC, C2 Q6600, 2GB OCZ DDR3 1600
ATI HD 4850 512MB, 1TB Caviar. 1TB Seagate
Shadow_Ops_Airman1 Mar 14, 2006, 02:12pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
probably Current DDR2 is still at fault, probably after AM2 releases, we will start seeing even lower latencies for DDR2.

AMD Athlon XP-M 2500+ (133x14= 1867MHz) (209x11= 2299MHz)
DFI LP NF2 Ultra-B (Hellfire 3EG Rev2)
Antec SX800, Neo HE 500, 4 Antec 8CM Fans
Thermalright SI-97 1 Antec Tricool 12CM Fan
CL SB XFi Xtreme Music
2x Barracuda HDs (250/400)
2x Samsung Write
Wayne Bradford Mar 14, 2006, 02:14pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
We are at 32 BIT now with some 64 bit applications available.
256 bit is unfortunately far down the road.

Radomir Jordanovic Mar 14, 2006, 02:20pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List

Edited: Mar 14, 2006, 02:20pm EST

 
>> Re: Re: AMD’s M2 socket, lowering performance?
I didn't mean 32-bit and 64-bit computing. DDR is 64-bit ram. That 64-bit ram has always been as such, even on 32-bit operating systems. When DDR is run in dual channel mode, it's at 128-bit. DDR2 is also at 128-bit with dual channel mode, but GDDR3 for graphics cards has been available for a while as 256-bit ram. That 256-bit ram ran on video cards that ran on 32-bit and 64-bit operating systems that had either 64-bit or 128-bit ram. This is possible because the operating system has nothing to do with the ram except that it uses it.

Silverstone 750W, Asus P5KC, C2 Q6600, 2GB OCZ DDR3 1600
ATI HD 4850 512MB, 1TB Caviar. 1TB Seagate
DarK_SlayeR Mar 14, 2006, 02:22pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
I believe he was talking about the next level of a duel channel memory interface greater then 128bit. I myslef have thought about this aswell, however, the 7xxx series card , like my X800, has GDDR3, which is a bandwidth king. It's unlikely we'll see this level of bandwidth untill we addopt something simular on the core system. I do think that a 256bit memory interface would help greatly in some areas, but if you think about it, 128bit DDR500 isn't being utalized to it's fullest, to my understanding. 7000Mb/s+ is a lot of room per second, just like the AGP bus.

I still say we need better hardrives; This is one thing Super XP and I agree on.

-------------------
Intel Core 2 Duo e8400 @ stock
Asus P5Q Pro P45
2GB x 2 Kingston HyperX DDR2 1066
eVGA GTX 550 Ti 1024mb Superclocked
Windows Vista Home Premium x64
DarK_SlayeR Mar 14, 2006, 02:22pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
HAHAHAHA, yeah, what he said ;). (a little late I was)

-------------------
Intel Core 2 Duo e8400 @ stock
Asus P5Q Pro P45
2GB x 2 Kingston HyperX DDR2 1066
eVGA GTX 550 Ti 1024mb Superclocked
Windows Vista Home Premium x64
Radomir Jordanovic Mar 14, 2006, 02:28pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
With an i865PE chipset and a 3GHz P4, I ran a 500MHz FSB (1GHz quad-pumped) and got about 5800MB/s. I was running at 3-4-4-8 latencies with my DDR500 Patriot ram. However, a friend of mine ran his Athlon64 3500 with a 220MHz ram speed and got 6600MB/s with SiSoft Sandra. He was at 2-3-3-5 latencies. I know it has a lot to do with how the ram is accessed by the processor, so I'm not comparing apples to apples. This is good, though, because since AM2 processors will have the same type of communication with the ram, as long as the latencies are generally low, it should be fine.

If you run 2-2-2-5 latencies at DDR400 with an Athlon64, you should get about 6GB/s or so. Therefore, 4-4-4-10 latencies at DDR2800 should provide nearly twice the bandwidth of the Athlon64's DDR. With DDR3 running at 1.6GHz with even higher latencies, you could approach over 20GB/s. I don't know what the latencies are for the 7900GTX ram, but they must be awesome to hit about 52GB/s.

Silverstone 750W, Asus P5KC, C2 Q6600, 2GB OCZ DDR3 1600
ATI HD 4850 512MB, 1TB Caviar. 1TB Seagate
Wayne Bradford Mar 14, 2006, 05:30pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
"We are at 32 BIT now with some 64 bit applications available.
256 bit is unfortunately far down the road."

I should have been a little clearer on this, Sorry

Radomir Jordanovic Mar 14, 2006, 05:41pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
32-bit operating systems run 32-bit applications, and this is done with 64-bit or 128-bit ram. I'd like to have 256-bit DDR3 ram for my 32-bit operating system to have greater bandwidth.

Silverstone 750W, Asus P5KC, C2 Q6600, 2GB OCZ DDR3 1600
ATI HD 4850 512MB, 1TB Caviar. 1TB Seagate
Julian Innerhofer Mar 14, 2006, 06:40pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List

Edited: Mar 14, 2006, 06:41pm EST

 
>> Re: Re: AMD’s M2 socket, lowering performance?
That the DDR2-latencies are so much higher than DDR1 is only a myth. DDR2 uses more clock cycles, but the latencie, that is relevant for performance is not measured in clock cylces, it is measured in time units (ns). Wehn the clock i i.e. 50% higher, than you have the same latencie w/ cl3, that you would have w/ cl2 on the lower clocked RAM. This time is about the same for DDR1 and DDR2. At the beginning DDR2 was a little bit slower, but that changed.

I think, the reason, why Socket M2 is so slow no is just, because the architecture is not really finished.



"This is possible because the operating system has nothing to do with the ram except that it uses it."

i dont think, that this is the only reason. Another reason is, that high performance Graphics Cards use less RAM, but they are very expensive, so they can use more expensive RAM than for Mainboards (normally 512MB RAM for main memory cost about 50$).

Another reason may be, that the chips on graphic cards are diectly on the board and dont use a slot interface, which could make it more difficult to raise the bus speed.

Also, on Graphics Cards bandwight is much more improtant than latencies, so the RAM is made for high bandwidth, but it also has high latencies.

But I dont think, that the OS has an influence on the maximum theoretical throughput or at the latencies.

SuPeR Xp Mar 14, 2006, 06:48pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List

Edited: Mar 14, 2006, 06:50pm EST

 
>> Re: Re: AMD’s M2 socket, lowering performance?
Havent you guys read anything? There is a reason why AMD postponed there AM2 Socket CPU's. Read Up.

And any testing with there samples of AM2 CPU's floating around is not credible at all. They do not reflect the true.

FACT OR FICTION 2 - AMD vs. Intel
http://www.hardwareanalysis.com/content/topic/56550/

-------------------------------------------------
Custom AMD HAF 932 Red Dragon GAMING MOD!!!
http://www.techpowerup.com/gallery/2442.html
Everett Williams Mar 14, 2006, 07:40pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
DDR2 is a dead end, even for the consumer market. I'm seldom fond of Intel's solutions to problems, but they have the best for this until such time as a RAM can be found that operates fast enough to eliminate most cache, and that would be FB-DIMM. What is spent on the one extra chip is easily gained back in the costs of the MB, and the width of access is then much more easily extended. The latency is no worse than DDR2, and the amount of RAM is almost limitless. What's not to like. DDR2 should have been aborted in the womb. In the meantime, large cache along with compilers that control locality of reference make for a guaranteed improvement in performance. AMD's big problem right now is that their cache coherence routines create massive overhead. They are aware of the problem and apparently have a solution in the stream for 2007, but Intel's next set of DP's are going to rock AMD for the 3-6 months that they will exist before AMD gets there with improved cache coherence routines. Until Intel gets out that next group, AMD is way ahead of the curve on them.

SuPeR Xp Mar 14, 2006, 08:05pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
Well, it is a good thing that the AM2 will support DDR3 also.

-------------------------------------------------
Custom AMD HAF 932 Red Dragon GAMING MOD!!!
http://www.techpowerup.com/gallery/2442.html
SuPeR Xp Mar 14, 2006, 10:24pm EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
"Until Intel gets out that next group, AMD is way ahead of the curve on them."

Nice comments Bro and I would agree in terms of technology, AMD is superior IMO. Unless Intel finally rids themselves of the old FSB & comes out with there new SCI design for inter connectivity. Intel's SCI seems to be postponed further back to like 2008 now. It looks like they are still having problems or it is just taking longer then they thought. I mean, ATI's HTT was not fun & games you know ;)

-------------------------------------------------
Custom AMD HAF 932 Red Dragon GAMING MOD!!!
http://www.techpowerup.com/gallery/2442.html
Sander Sassen Mar 15, 2006, 03:16am EST Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Re: Re: AMD’s M2 socket, lowering performance?
Well, the way I see it Intel demo'd their Conroe system and basically said to AMD to show their cards. The question everybody is interested in is can AMD beat them or not. The answers are basically already given by Henri Richard, AMD's executive vice president and chief officer for marketing and sales in the below noted interview. They won't be able to. The way AMD has been operating these past few years is that if they have the goods they're not in the least part shy about it and if they don't they say they 'can't comment for competitive reasons'.

http://www.digitimes.com/bits_chips/a20060313PR201.html

Best regards,

Sander Sassen
Editor in Chief - Hardware Analysis
ssassen@hardwareanalysis.com

Write a Reply >>

Continue Reading on Page: 1, 2, 3, 4, Next >>

 

    
 
 

  Topic Tools 
 
RSS UpdatesRSS Updates
 

  Related Articles 
 
 

  Newsletter 
 
A weekly newsletter featuring an editorial and a roundup of the latest articles, news and other interesting topics.

Please enter your email address below and click Subscribe.