Printable View
http://blogs.zdnet.com/Ou/?p=735
George Ou has happened upon some slides that are (were?) to be presented on September 10th.
For 4P, 2GHz K10 scores 160, losing to 8MB 1.86GHz Tigerton score of 166 though it bests the crippled 4MB 2.13GHz Tigerton.
For 2P, 2GHz K10 loses to 2GHz Clover by 4.2 points (88 vs 92.2)
What a disaster. Once the Seaburg chipset launches in Q4, Intel's 2P specINT score will also beat Barcelona clock/clock.
Tigerton runs on Clarksboro, which already has the updated snoop filter designed for quad core.
Seaburg will add this feature (vs. current Blackford).
reply to linked Ou blog:
Some interesting points.Quote:
Hi George,
The 4-processor SPECint_rate2006 slides you posted are actually GOOD news for AMD !
They show the Opteron 8350 platform ($1004/proc) being faster, or as fast as, the Xeon L7345 and E7320 platforms (respectively $2301 and $1177/proc) ! Then, the Xeon X7340 score is indeed higher than the Opteron 8350, but this Xeon costs twice as much: $1980. (It would have been interesting to see the score for the intermediate Xeon X7330 by the way.)
Everybody is so focused on absolute performance, but, in the end, all what matters is how much bang you get for your buck. So unless you stricly buy the highest-end processors ($2000+ proc representing the top 20% of the 4P market), these slides prove that Barcelona is faster & cheaper for everything else (the remaining 80% of the 4P market) !
I don't even pay attention to Ou, because all he seems to do is bash AMD.
Yep,Gearge Ou is like an intel-Sharikou version,only at the larger scale(since he writes for the "famous" ZDnet)
QFTQuote:
Originally Posted by informal
Yeah , too bad AMD changed their policy wrt to benchmark scores based on his input ; OTOH who gives a ***ss about Sharikou ?Quote:
Originally Posted by informal
This , if anything , shows your level of fanboyism , even when AMD gives credit you still beat the dead horse...:horse:
Yeah whatever,i couldn't expect more from you,the same story all over again.
If your this pi ssy now, it's going to be entertaining to see how you act come Monday.Quote:
Originally Posted by informal
Can we keep flame baiting out of this one? :rolleyes:
You do realize that these scores came directly from AMD, right?Quote:
Originally Posted by Zytek_Fan
well, didnt you know everything that has no amd in the link = bad ? :rofl:Quote:
Originally Posted by Proesterchen
assuming george ou's numbers are accurate:Quote:
Originally Posted by hawkeyefan
Intel can change the pricing sooner than later, but AMD NEED to recover R&D and their huge losses, Barcelona die is huge (it equals the clovertown die, but don't forget that Intels 65nm is more mature and that they just take two conroes and put them on one die - increasing yields. and penryn is launching very soon), they will have no chance when tigerton's 45nm successor arrives (when is this going to happen anyway? I don't think it's as early as Q1'08) or when Intel decides to lower their margins a little to hurt AMD. Besides we need to know performance/watt, oh and the E7330 (2,4ghz, 6mb cache) at $1391 should be a worthier competitor against the 2ghz Barcelona on price/performance.
The int numbers are really discouraging..
I thought HPC is a niche market?
How are they going to charge a price premium for HPC? Doesn't the HPC crowd use off the shelf opterons with some exotic chipsets and interconnects?
Then they can't even charge a premium for the SUPERIOR floatingpoint+bandwidth performance, because then no one will buy the opterons for normal servers (cause they are really weak on int) and if they price them according to their int performance, well...
btw
µopt was prolly right on the int performance beeing closer/worse than conroe. gj :clap:
Not only does AMD's 4P CPU has better price/performance ratio, it sits in the same mobo that customer has! For Xeon series 7300 you need new mobo. Also for Barcelona you need RDRAM and for Xeon 7300 series you need much more expensive FB-DIMM's.Quote:
Originally Posted by hawkeyefan
All in all 4P Barcelona is far more superior product.
Regarding the "niche" market... if you call auto, space, oil, mining, financial… niche (aka not worthy) markets, then OK, but targeting these market is the smartest thing that AMD could have done. Those guys have money, and they buy a lot ;)
And for the reference pricing of new Xeons, and new Opterons:
XEON
• X7350 @ 2.93GHz, 8MB L2 cache, 130W TDP at $2310
• E7340 @ 2.40GHz, 8MB L2 cache, 80W TDP at $1980
• E7330 @ 2.40GHz, 6MB L2 cache, 80W TDP at $1391
• E7320 @ 2.13GHz, 4MB L2 cache, 80W TDP at $1177
• E7310 @ 1.60GHz, 4MB L2 cache, 80W TDP at $856
OPTERON
• 8350 @ 2.0GHz, 95W TDP at $1004
• 8347 @ 1.9GHz, 95W TDP at $774
• 8346 HE @ 1.8 GHz, 68W TDP at $688
Pricing can change in a matter of days - performance on the other hand.. :rolleyes:
It's only the same if the customer has Socket F based systems. What about those who have Socket 940 systems? FB-DIMMs are essentially the same price as registered memory, while IBM's X4 based servers will use registered DDR2.Quote:
Originally Posted by Nedjo
But they haven't been buying much of AMD lately, given AMD's significant loss of market share in servers.Quote:
Regarding the "niche" market... if you call auto, space, oil, mining, financial… niche (aka not worthy) markets, then OK, but targeting these market is the smartest thing that AMD could have done. Those guys have money, and they buy a lot ;)
Performance dictates pricing.Quote:
And for the reference pricing of new Xeons, and new Opterons:
XEON
• X7350 @ 2.93GHz, 8MB L2 cache, 130W TDP at $2310
• E7340 @ 2.40GHz, 8MB L2 cache, 80W TDP at $1980
• E7330 @ 2.40GHz, 6MB L2 cache, 80W TDP at $1391
• E7320 @ 2.13GHz, 4MB L2 cache, 80W TDP at $1177
• E7310 @ 1.60GHz, 4MB L2 cache, 80W TDP at $856
OPTERON
• 8350 @ 2.0GHz, 95W TDP at $1004
• 8347 @ 1.9GHz, 95W TDP at $774
• 8346 HE @ 1.8 GHz, 68W TDP at $688
Quote:
Originally Posted by Nedjo
AMD's pricing only makes it very clear with which Intel parts they are competing ; that is 2GHz K10 is more or less equal to the E7320 in apps that customers care about.
If anything , that's worrying.You know why ? Because in that segment you're overlapping with DP , for that money you can buy a 5365 3GHz QC Xeon that would cream the K10.Few apps scale well to 16 cores , instead of 16 cores you're better off with 8 faster ones.
I'm pretty sure an 8 core 3GHz Xeon system will be cheaper and perform close enough to a 16 core 2Ghz K10 system in most commercial apps.
Quote:
Originally Posted by MR_SmartAss
Can we agree that product that cost 1004 USD wipes the floor with product that costs 1177 USD is superior?
Can we agree that product that does not require new expensive motherboard, and saves to server manufacturers big buck 'cos it uses mature and each day cheaper platform is superior?
Can we agree that product that uses far less expensive RDRAM in comparison with hot FB-DIMM is superior?
Can we agree that CPU that can independently clock its cores and logic inside of them is superior?
Can we agree that CPU that is not bottlenecked with antique FSB is superior?
;)
Since you probably backing this claim with insight to the performance of the apps that customers care about, could you be so kind and share those with us?Quote:
Originally Posted by savantu
But it doesn't, otherwise it wouldn't be priced at $1004.Quote:
Originally Posted by Nedjo
But it does require a new expensive motherboard. It's just that customers would have had to purchase it last year. But a lot didn't bother to go to Socket F and went with Intel's 2S quad-core solution instead which offered comparable performance to 4S Socket F at the time at less than half the power consumption.Quote:
Can we agree that product that does not require new expensive motherboard, and saves to server manufacturers big buck 'cos it uses mature and each day cheaper platform is superior?
They're roughly the same price. But IBM will offer registered DDR2 on its servers.Quote:
Can we agree that product that uses far less expensive RDRAM in comparison with hot FB-DIMM is superior?
No, since it's probably necessary for AMD to use every trick they can to match up with the spectacular performance/watt of the Core uarch.Quote:
Can we agree that CPU that can independently clock its cores and logic inside of them is superior?
Too bad it's bottlenecked by its lack of clockspeed, slower cache sub-system and just being slower in general.Quote:
Can we agree that CPU that is not bottlenecked with antique FSB is superior?
I'm talking TPC-C , SpecJBB , that kind of benchmarks.Quote:
Originally Posted by Nedjo
SpecFP_rate is nice , but nobody except HPC folks give a damn about it.
Oh yes it does:Quote:
Originally Posted by accord99
http://blogs.zdnet.com/Ou/images/4p-...amd-server.png
If we talk about 4P customers, we need to talk about HP, IBM, SUN, DELL, and they 4P AMD offering is SocketF, and they are very, very happy 'cos of backward compatibility.Quote:
But it does require a new expensive motherboard. It's just that customers would have had to purchase it last year. But a lot didn't bother to go to Socket F and went with Intel's 2S quad-core solution instead which offered comparable performance to 4S Socket F at the time at less than half the power consumption.
RDRAM (DDR-667): http://www.newegg.com/Product/Produc...82E16820134329Quote:
They're roughly the same price. But IBM will offer registered DDR2 on its servers.
FB-DIMM (DDR2-667): http://www.newegg.com/Product/Produc...82E16820134327
So, $50 price diference for one 4 GB kit. And since you need at least two of them that's 100 USD pricce diference. if you can call that "roughly same price" then its fine with me!
pff with all that spectacularity that you talk about, AMD still leads when it comes to perf/wat on server market: http://www.worlds-fastest.com/wfz991.htmlQuote:
No, since it's probably necessary for AMD to use every trick they can to match up with the spectacular performance/watt of the Core uarch.
if this is slower, then there is no point to this discusion:Quote:
Too bad it's bottlenecked by its lack of clockspeed, slower cache sub-system and just being slower in general.
http://blogs.zdnet.com/Ou/images/2p-...ptember-fp.png