yeah, i think so too. the new integrated gpu's and chipsets seem to be outperforming their intel counterpartsQuote:
Originally Posted by Tony
Printable View
yeah, i think so too. the new integrated gpu's and chipsets seem to be outperforming their intel counterpartsQuote:
Originally Posted by Tony
With good marketing they could increase their sales
Friendly advice. Dont expect too much until its really there. Also since the core aint radically changed, its still inferiour to a Core 2 core.Quote:
Originally Posted by Rammsteiner
DDR3 wont do anything, cept synthetics and rates benches. We aint bandwidth limited today. So unless the Phenom gonna clock higher than Lynnsfield. It wont touch it. Specially not with SMT enabled either. Yorkfield it can if its something where the FSB strangles it. But again, that sure aint games....
In short, think abit conservative. Else it quickly ends like last time and Barcelona being 50% faster...
If 1 bench, then Yorkfield is also 100% faster than Kentsfield :D
I'm not surprised. SuperPi is cache-intensive, and I believe that the performance increase in SuperPi is mainly due to the increased cache in 45nm Phenom. However, most of the other applications are not as cache-intensive as SuperPi, so I highly doubt that the other applications would show the same increase as SuperPi.
No, Im not expecting it actually 'will' 100% be competitive with Lynnfield. I just dont think it's going to be impossible however it all depends mainly on Deneb's clock yields and eventual further IPC improvements.Quote:
Originally Posted by Shintai
As for DDR3, dont know really. Upcoming GPU's do require quite a bit of bandwith though, but if it would bottleneck DDR2 bandwith to justify DDR3, not sure.
But as you said, Im not setting up my mind like Deneb would become the new C2Q to say it like that. Ill wait for the release for sure.
To make my point a bit more clear thus far, I think the delta between Yorkfield and Agena will be significantly reduced compared with Lynnfield and Deneb. But still, only thoughts, no hard proof and not expecting too much;)
According to this article at Toms Hardware, a single Agena core is right around 13% faster than a single Windsor core, on average. This 13% figure represents the average difference between the cores tested in 18 games and apps, and exludes synthetic benchmarks. They didn't test Super Pi, but comparing my Phenom 9850 @ 3GHz with 3GHz Windsor scores on the internet, it appears that Agena is about 14-15% faster than Windsor in Super Pi. So with AMD processors, at least, it appears that improvements in Super Pi closely track improvements in games and apps.
Well we will just have to wait and see. I really dont trust much hype anymore after nV's "whoopass".
K10.5 vs K10 and K10 vs K8 is a whole different matter though, I'm not sure you can really come to any conclusions based on that alone - there are a lot more performance tweaks in K10 over K8 than there are in K10.5 over K10.Quote:
Originally Posted by Yokozuna
Like I said earlier, if its possible, someone should test their Phenom with the L3 disabled and see what kind of impact it has on SuperPi scores.
AFAIK,you can't just turn off the L3 cache in K10.What you can do is use B2 stepping and apply the TLB patch for it.In this way you will penalize the CPU when L3 cache is concerned (as much as possible at this moment since the L3 can't be turned off).Quote:
Originally Posted by Epsilon84
I tend to disagree.. Turning the TLB workaround on does a lot more (worse) to performance than if you were to simply 'remove' the L3..
Well since you can't remove the L3,what do you suggest instead?Quote:
Originally Posted by mAJORD
actually mAJORD is correct! You can't simulate lack of L3 by turning on TLB fix! it's just not the same thing! What I predict for K10.5 w/o L3 is really small penalization 'cos of the absence of L3, especially for the X3 and X2 versions.Quote:
Originally Posted by informal
'cos of the microarchitecture improvements of K10.5 I'd expect models w/o L3 will be in the level of Agena's performance clock-by-clock, with much higher OC headroom.
i honestly dont think the l3 cache increase can account for more than a 4% performance increase
Well not actually :DQuote:
Originally Posted by Nedjo
Less volts to start with, but at 3.5 you end up probably within throwing distance.
anybody have info how many megawatts is the tdp for deneb? if im not wrong it shouldn't exceed the mobo design w/c is around 140watts? thanks.. can't wait for deneb to get to market, im forgoing phenom due to high heat, it's currently 100degrees here and having a hot pc under the desk doesnt help.
The two most important things for AMD to do with Deneb are to lower the TDP and the L3 latencies. There isn't going to be a huge architecture change, and a huge change in IPC. What seems to be shown through the SP 1M is that the latter has been accomplished, at least to some degree. Granted, SP is cache defendant, but I don't recall the speedup from Intel's shift of 2->4Mb of cache to have improve the times by even 10%, and even then there are differences between the effects due to FSB vs. HT. As far as the TDP is concerned, I'm not so sure, and the voltages needed to get the OC'ed speeds are down-right scary. Both of these things would improve the general performance enough to tide them over until they can draw up a better architecture....
Ya'll are funny
I got the BS meter out and it's off the scale. ;)Quote:
Originally Posted by IvanAndreevich
Even at 3.5 though for a quad is more than most people would need. You know how those Johnsons are always trying to beat the Johnsons though.
It may not happen immediately but given time I'm sure it will be there. Although the brisbanes were at about all they were going to get out of them from the beginning I guess. Couple of steppings and only a couple of 100MHZ later. I think AMD knows they need to get in the game though. The newer south bridge is already showing a marked improvement on overclocking the phenoms with the same chip.
I think some of the overclocking not being so good on the phenoms is the board quality also though. I have not really seen too many really good looking phenom capable boards. Half the boards out there are barely phenom capable much less once you start pushing the 150w envelope.
Anyway readthrough has been fun!!!!
Love the super pi bs though. I do 1m in just under 27 or right over 27. The rest of my #s are close to e7200 ones which run 1m at under 20. Most my futuremark scores an Intel has to be clocked really well to take me by any kind of a margin. As in above 3.5ghz. This is just comparing the same graphics card though, X1950 pro in crossfire. I get 9000 in 06 but an Intel with the same FPS gets 12000 and the cpu mark is all that!! Thus cards are bottlenecking. The newer 4800 series cards like some cpu to feed them though. I notice nice FPS increase with CPU increase on these cards. PcMark is little strange on my scores. Almost 7000 but think it should be a little more, not sure what is going on there though. HDD scores are a little lower than they should be I believe and HDD virus scan is really low?? May be the SB600 on my board or something else but who knows???
MMMM 45nm Thx for the laughs all.
For a super PI run, though, the effect is negligble ... perhaps 5% ... if superpi is truly this cache senstive, then starving it of 2 megs of cache should have a greater impact.Quote:
Originally Posted by mAJORD
Sorry, Is this not a contradiction? Are you saying that deneb does not need to worry about nehalem because nehalem will be above mainstream but also deneb will be a success because it will allow AMD to sell cpus above the mainstream level?Quote:
Originally Posted by LOE
Also, IS there not going to be a $266 nehalem? Would this not be mainstream?
I'd lile to see the quotes from the 'sources' that say Deneb will be faster than Penryn per clock. :)Quote:
Originally Posted by LOE
As for Nehalem, the 2.66GHz part will cost $284. If a 3GHz Deneb cannot outperform it (which I think is highly likely) then the most AMD can expect to charge for it is around ~$250, or if they keep their current pricing system, $235 like the 9950BE.
I'm starting to think that Intel deliberately priced Nehalem so low to hurt AMD's ASPs. They want to keep AMD on life support, strong enough to survive so they don't become a monopoly, but not really strong enough financially to really challenge them from a R&D standpoint.
so about 20% slower per clock than the penryns... and they'll probably clock to around 3.5Ghz.
well it's a start. definitely better.
Would that be the same sources with 3Ghz Phenom and 30000 3Dmark06 and 50% faster than Kentsfield in averge? Im curious....Quote:
Originally Posted by LOE