here is a PDf of 45nm...the do mention of High-k down the road, and I believe it might be on 45nm AM3..which might be why the FX might be back...any thoughts..
Printable View
here is a PDf of 45nm...the do mention of High-k down the road, and I believe it might be on 45nm AM3..which might be why the FX might be back...any thoughts..
hhhhmmm let see all of deneb's new features
1. HT3.1 3200mhz (like Icore7 i 3200mhz too) helps memopery mostly, maybe a little with GPU.
2. L3 cache speed is linked to HT speed(aka HT can't be higher then NB) (1 and 2 is more for overclocking)
3. agena had 32way L3 cache and deneb will have 48 way (maybe we can get revision with 64 way please wishful thinking lol)
4. deneb 15% faster at same speeds as agena this means nb/L3 cache speeds may have been the same. increasing the nb/L3cache speed helps as much as 2-5%.
5. clocking is better due to removed resistor in the cache. (no idea why that was in there in the first place maybe an engineer over looked it)
6 K10 was already 15% faster then K8. which AMD stated it would be
7. NO TLB buggs actually if you look even intel had one of them too.
8 Penryn was IPC boost and shrink too, it wans't 15% it was only 5-10% and had more cache dual 6mb vs 4mb and Quad 12mb vs 8mb. that's only 25% more cache and came out to be and average 5-7% faster and like maybe 7% less watts.
9. That IPC is L2 cache for penryn to conroe (16 way for conroe and 24 way for penryn) L3 is whole different story on phenom. When you consider phenom is only 15%(rarely is that high it's more 5-7%) with 50% less cache.
10. given all details including the SuperPI score have we needed a 4.0ghz 65nm AGENA beat a 3.2ghz DENEB
11. so yeah it can beat york field in my opinion by at lest 2-5%
12. I can not see why people are comparing Deneb to nehalem. I see it more of a conroe vs york field not nehalem (AGENA v DENEB)
13. the architecture is basically still K10. not all new like Icore7 nehalem is.
( last few are a couple things Iv'e been reading around here)
14. I still think AMD motherboards are getting the screw job in the Power phase area (where our digital PWM ? )
15. Icore7 is looking Hotter then R600, phenom may have been hot but no that hot.
16. at least where not in a TDP OC with phenom.
wow demon save some of the "speculating" for others. :p: you just about hit everything. and #15 on ur list about the nehalem temperatures, yea that just confuses me. i believe their tdp is DOUBLE of what shanghai's is.
wonder what the prices of the fx chips are gonna be next year. most likely $1000+
I wonder what the comparison of actual draw numbers will be.
I agree he covered most everything...and his point's all point to deneb being a very big improvement over agena.
The biggest reason will be the fact it will overclock to q6600's 3.6 level's without tdp going crazy like on nehalem.
This combined with the ipc of 15-20% that's been quoted many lead deneb to be a winner....specially for the non rich gamer crowd.
Down with DDR3 :p:
I'm going to guess that Deneb is going to be around 10% faster overall, with anywhere from 0-15% improvement depending on the application.
I'm guessing pricing will be a bit higher than current Phenoms but not much; the 20550 @ 3.0GHz will likely come in a bit lower than the Q9550, so under $300.
For overclocking I am thinking 3.6-3.8GHz stable on SB750 mobos with the launch revision. Maybe 4GHz possible with a future revision in 2009.
so is deneb going to be as fast or slightly faster than the first gen c2d (conroe)?
My speculation is that Deneb will not at first have an unlocked multiplier. We will only be able to overclock HTT. The clocks will "only" be 3.4 to 3.6.
AMD will realize that they need a BE version and it will be released in late January. Clocks will hit 3.8 to 4.0.
In late June, an FX version will come out with High-K metal gates. Overclocks will hit 4.4 - 4.6. The FX chip will run a lot cooler and take a ton of voltage. Most of us will kill our AM2+ boards by blowing numerous mosfets and will have to buy AM3 mobo versions, probably DFI because of their digital power handling.
Nehalem buyers will have remorse when they discover that even slight overclocks result in between 80C and 100C temps. We will see a huge migration of upset Intel customers to AMD Deneb.
My speculation is that Deneb will be a nice little bump in improvement for anyone with supporting AM2+ boards. Nothing over the top amazing but a nice upgrade prolly at a decent price. AM3 on the other hand they can start fresh I think will be interesting or else they are going to be in trouble.
sounds too good to be true...not that i would know.Quote:
In late June....Overclocks will hit 4.4 - 4.6
if you had said i would be hitting 3 gig on a quad core from AMD, this time last year (maybe little more) i would have found that hard to believe.
but here we are, hitting 3.5 gig some of us, this time next year i fully expect a 24 hour clock speed of 4 gig + :up:
Im shure well see 4ghz easy by the First of the year using water/high end air 24/7....Wisky ,,Beer and other Overclocking additives will come into play this Holiday for any one who can get thier hands on One:D.
i got to admit, im getting excited now, my goal has allways been 4 gig on an amd chip, god when i made 3 gig on this 9850BE i was ecstatic.
pity i am in the UK, i wont see any of this till march probably, god were just getting the Asus Asus M3A79-T here.
if the 'vaporware' Nvidia ACC enabled boards dont turn up, i might just drag it out till AM3.
he sais, lying
LN2 might get you to 4.4-4.5, but don't expect more.
3.8-4.0 will be the max sweet spot I reckon, but it would be an extremely nice speed for most of us already. And unlike the i7, it doesn't get up to 70+ degrees at that.
Nehalem looks to be the best and easiest cpu to oc ever. Before you reply with the predictable "But anandtech found that if you oc the lowest binned part by 1.3ghz on the stock cooler it might overheat" Think what you will say if deneb runs hot when you oc the lowest binned one by 1.3ghz on a stock cooler.
My prediction, Deneb will be a success, It will enjoy a 10-15% ipc improvement on average some things higher and some things lower and will have very good power usage below 3ghz, Over 3ghz it will be a lot worse due to current leakage with no highk metal gates.
It has a good chance of taking the performance per $ crown and the performance per watt below 3ghz crown.
I'm guessing we'll see something more sane--in the neighborhood of 3.3-3.4 GHz or so stable OCs in 64-bit being common. Even so, with the higher IPC, it's going to be nice.
I agree it will be a lil hoter under higher clocks.. but I still feel it will not be as bad as Agena.. did the guys over at Coolaler ever mention temps on the OC they had on air? I believe it was a huge heatsink..but anyways.. untill Highk, the OCing attepmt will most likely still need the v's to get there..which will result in higher temps and more power draw of course. I still agree they will do a much better job at OCing tho. I am no engineer so correct me if I am off base :p:
Is 4.0Ghz at 53C on air while under 100% 4x load hot? :cool: It's a new day in AMD land!
http://i35.tinypic.com/25fiwj4.jpg
http://enthusiast.hardocp.com/articl...50aHVzaWFzdA==
83C on a TRUE heat sink.
the chips will operate at 80C but there life span will be shortened.
some lower phenom have a higher max temp then higher clocked one like 9550 72C vs 9850 66C
higher bins need less volts there i probably another reason why. but if it's above 70C for most chips that's hot.
Deneb 45nm lithography uses water immersion form IBM if you look at the PDF posted in the thread you will see it's has more precise etching now.
Will deneb work in AM2 motherboards? My prediction is that it will not. :(
I don´t remember where i read it about this but the info was that AMD doesn´t recommend to give support for deneb in motherboards without separated power planes.
Something is not right with this, There is no way that 1.475v on a non highk process will max out at 53c on air, It just is not possible.
What temp do you expect deneb to run at with a 1.2ghz oc? Also, I believe water immersion will not help with leakage with high volts on a non highk 45nm process.
Well, I think it is within the realm of possibility. Informal was kind enough to post some comments that help illuminate AMD's recent effort.
AMD knows the stakes are high. They have to defend the server market and improve their desktop performance.
With no bearing on what it will perform like, Unless AMD have done something radical they will experience severe leakage with high volts on a non high k process. It could have 90% higher ipc but it will not change the fact that a deneb running at 4ghz with 1.474v will leak like crazy. Still though, I am just a builder by trade so what do I know:), We will find out in a few days.
im having high hopes for this cpu does any think there is a cahnce it can beat yorkfield or wolfdale in gaming? what r the chances
This may explain why leakage is reduced:Special thanks to RiverRicer who provided the SOURCE :clap:Quote:
Ultra-low-k Dielectrics. In some later 45nm products, AMD plans on using ultralow-
k dielectrics to reduce wire delays by as much as 15 percent and enable greater
overall processor performance.
1.168 sounds really too low. But on my System (9950BE on 790gx Gigabyte) AOD shows wrong voltage an CPU-Z shows the right.
But I can´t believe... So let´s see!
CLARIFICATION: To be clear, I was only speculating about how the OC'r who provided the snapshot could achieve 4.0Ghz on air at 53C under 4x 100% load. If this result is reproduced, then there must be new design elements that reduce leakage and heat. I just wanted to highlight a possibility.
My main reason for posting this was RiverRicer's source. These design elements are clues as to what we might see in an FX version of Deneb, e.g. high-K metal gates.
High-k is planned for H2/09, i.e. for the Istanbul 6 core chip. AMD stated this on one of the last press conferences.
What Du you mean with "memopery" ? Mem speed ? That is totally unaffected by the faster HT3 speed. Only exception are coherent memory transfers in Multiprocessor setups, but we are speaking about Deneb = 1P; not Shanghai, dont we ?
To the 4 GHz screenshot:
It is already from July:
http://www.overclock.net/amd-cpus/35...enom-45nm.html
Are there no new ones ?
cheers
Opteron146
well once a trust worthy site like toms gets there hands on it....were know what potential this chip has.....i want a review with deneb picked apart with an UNbiased view and benched to extreme conditions..:D then the dark will come to lite...weather this is 4800 series come back...or just a cooler running phenom that clocks higher...cuz if i wanted that i would just water cool a 9950 black edition and do every thing i could to push to 3.5ghz...we need the performance claimed...with out any twists in how there calculating this gain...i dont like cpu performance spin doctors trying to market remanufactured crap in box... show me 15% clock for clock......
I dont know why you guys keep thinking anything is going to change AMD always disapoints us well be lucky to hit 3.6ghz stable:mad: .........and I heard the first batch has the erata bug again. If theres one thing I have learned about AMD is their claims are always exagerated by 10-15% so that means we will have no clock for clock increase and only higher overall processor speed......I hope Bulldozer does better:shakes:
gotcha!:D
Heh. I was already rolling up my sleeves and breaking out the brass knuckles.
I have $400 sitting on the shelf in hopes the chip will be out by the time I get home in a few weeks!
So what roadmap have AMD released for Deneb speed grades at this stage?
When is 3.0Ghz due?
well they released a roadmap awhile back saying the chips would come very late December to early January. now it seems it could be this month and they would be releasing a 2.8 ghz 125W and a 3.0 ghz 125W both for am2+. early next year the am3 ones will be rolling in including a 2.8ghz 95W and a 3.0ghz 125W. thats all i can remember i know they have other chips lined up but they info for them is not released.
I have no qualms about putting 1.6 through a chip :D
My guess it'll hit 3.6-3.8ghz on air, around the same IPC as Yorkfield and perform similar to Yorkfields when both are oc'd to the max (on air).
AMD K10.5 Hydra to challenge Nehalem
:)Quote:
To have 1MB L2 per core
After Deneb and Propus 45nm K10.5 Quad-cores AMD plans to change the process and improve the existing K10.5 cores. The new process is called 45nm K10.5 Rev. D and it will bring high K to 45nm SOI cores. K10.5 Rev. C is the 45nm SOI process that AMD plans to use for Deneb and Propus 45nm cores. AMD calls Rev D cores Hydra and these are the chips that will really go against Nehalem.
The test production of these rather interesting cores will start in Q1 2009, but we expect to see the production cores by the middle of 2009 or a bit later.
The new K10.5 Rev. D supports 1MB L2 per core, which is twice as much as in current K10 CPUs and 6MB L3 cache memory. The most important feature is that it will be able to get you eight cores and we believe this is eight-core native design and not MCM (Multi Chip Module), as many have suggested before. This confirms our earlier story that AMD’s eight cores are native and you can read it here.
The new chip should be able to hit clocks in excess of 3GHz but we will hold our breath until we see it. This does look positive for AMD, but it still means that Nehalem will probably be the market leader for at least six to nine months before it gets a real competitor.
http://www.fudzilla.com/index.php?op...=7197&Itemid=1
The funny thing is, AMD is shipping shanghai right now. Im suprised some really wealthy person or site hasn't just gone out and benched some stuff, im sure some things scale a certain way or are clock limited so they could still test quite a few things that would be the same or very similar to deneb.
8 cores with 1MB l2 and 3ghz @ 45nm? No way. 32nm maybe, not 45nm. Way too big, it would be a mammoth die, and AMD has never been very good at making large monolithic dies.
SocketG is 6 and 12 core ready.It's rectangular and quite big :).Ready for Bulldozer too ;).
That isn't the issue. The problem is that on any given wafer of silicon, there will be some random defects. The more defects there are the more cores will be bad. The fewer chips you get out of one wafer, the more silicon you waste from a single defect. If you were to develop a core that is as big as a wafer, you'd get a 0% yield. There's plenty of space in the CPU package for a large die. It just isn't a good idea since larger dies generally equate to more waste. That will drive up per-unit costs beyond where you'd expect it to linearly be judging from area.
I dont believe this fuad story, cause the 1MB L2 Quad Core got deleted on the AMD Roadmap after AMD switched to the 6 Cores.
Its old code name was "Montreal". Furthermore Fuad sometimes writes really lots of trash ..
Besides that Spec Numbers for dual / octo Shanghais are out:
2384:
http://spec.org/cpu2006/results/res2...024-05683.html
http://spec.org/cpu2006/results/res2...024-05684.html
8384:
http://spec.org/cpu2006/results/res2...024-05685.html
http://spec.org/cpu2006/results/res2...024-05686.html
cheers
Opteron146
i don't believe fud period. they are just posting things over and over. most of the time they take both sides deneb sucks, deneb is good. i bet if someone tested the accuracy of their articles it would be below 25%. i don't even care about news anymore theres nothing else to say except for performance results and with shanghai coming out and deneb coming out soon that could be coming in. but really why does it matter? everyone is gonna get deneb because we all know how great it is going to be. :up::yepp::p:
Will we though? I have been an AMD fanboy since the days of k6II. I am now wanting/needing to upgrade but am waiting for fresh benchies and pricing of Deneb. If it fails to impress (which has quite honestly been the AMD trend lately) i will do the logical thing swallow my pride and get the base nehalem offering since I am already needing a new board, ram and cpu.
For the sake of a competitive market, i hope we get pleasantly surprised here. But personally, i am not expecting anything spectacular.
get the 3.0. go big or go home :). also i saw this today: http://biz.yahoo.com/rb/081112/business_us_intel.html intel just letting everyone know that they are going to be making less money this quarter. they blame it on the economy but i think theres a little something else to it as well.
True, but price is still a concern for me. Im pretty sure that the 920 nehalem will sell for around $300 and it appears to o/c nicely from what ive read so far. Hardocp took it to 4ghz on stock cooling with ease. I just keep thinking to myself that deneb will be priced around that same $300 price point. So, for me, the decision will be made on the initial deneb benchmarks and how well it will clock. Sure DDR2 is cheaper and the AM2+ boards are cheaper but this is already older gen tech as DDR3 is here and AM3 is right around the corner.
I personally cant believe i am thinking about an Intel build but it seems like a no contest decision to me. What would you guys do?
its pretty obvious that amd will be hurting intel's sales this quarter even if shanghai/deneb is a flop. with new parts theres always more sales. 9950 comes out and everyone buys it when the 9850 is pretty much the same and when overclocking they are both equal.
i would still get amd even if intel was better but theres nothing to say about that now because theres no comparison between nehalem and shanghai yet. but really the choice is obvious for me because i get amd cpus for free anyway.
Just to remind you the title is "Deneb vs Agena...feel Free to speculate" .You are flaimbaiting people into intel vs AMD argument yet again.The thread is about Deneb versus Agena,period.
Need i remind you you were already removed from one AMD thread couple of days ago?Should the history repeat itself?
they have performance wise i don't know how big of a market share they have compared to intel tho. just the release of shanghai should defenantly get some sales going there and intel can't do anything to beef up their sales xcept for drop prices.
lol anyone got any more speculation? there really isn't much to say.
Found this thread in the News section........early benchies of Shanghai with comparison to Barcelona and Zeon......
http://www.xtremesystems.org/forums/...d.php?t=207460
It shows some good gains in server apps as well as significant reduction in power use.
Yes the last 12 months have been pretty bad serverside if thats what you mean about AMD. They did lose marketshare.
The last few months it does seems as AMD has picked back small amount percentage.
But at least to upgrade on the AMD platform its more or less a bios update.
Not a full platform upgrade.
I can see Shanghai making Inroads to reduce the Bensley platform damage.
Now can you keep on topic please!
And with the AM2+ boards now being compatible with AM3 chips, there no need for a complete upgrade, and with no need for all the bandwidth that DDR3 has, theres no real need for DDR3 either right now. So, my "speculation" is that a lot of intel guys are gonna like what they see with Deneb and drop Intel due to the price savings of having to buy a complete new platform.
Even if they dont have any AM2+ boards laying around, they could still buy AMD stuff a lot cheaper. IMO I think that in itself will effect Intel!
Aussie pricing is a bit crazy at the moment due to the weak dollar and probably because we will be a month or so behind the USA in getting enough supplies, so I expect pricing to improve significantly in the weeks ahead.
Have AMD flagged what they are going to be charging for Deneb?
I know you won't need a new platform for it which will increase it's appeal amongst existing Phenom users.
But the the price of a top end X58 (not including CPU)
You can buy today a Phenom + 790FX + Ram
It will scale with current pricing im pretty sure about that and be very competative. As we have seen with Shanghai and in the past.
Dont forget the X2 guys too.
I agree that the platform cost for i7 is quite poor value unless you are some kind of crazy cruncher, so Intel will still be relying on Penryn to take on Phenom/Deneb for some time yet.
This is a laughable source I know, so please forgive me, but one of the crazies on AMDZone said that a distributor had told him that Deneb will come with a significant price increase over Phenom.Quote:
It will scale with current pricing im pretty sure about that and be very competative. As we have seen with Shanghai and in the past.
I guess we will know soon enough.
How true. Will be interesting to see what happens. I would imagine that the new chip would be more than the current ones - typical of new hardware releases anyway - but by how much remains to be seen. A lot of speculation at this point but then that's the point of this thread aye? :D
ehhhhh im guessing just about 200-230. not much more but a little. if they can price it right they will dominate for a few months.
Talking about price/cost. Strangely enough the 9850 and 9950 have been increasing in price the last month or so around resellers in the UK.
I would put a limit of around £175 /£150 for the Phenom X4 20550 3GHz and 20350 2.8GHz ,at retail level (and with prelaunch/order prices boosted another 15%) , with prices falling to around £150 £130. Thats my Guesstimate.
Too conservative ? Too Expensive ? Hard to predict how Intel will counter at the moment with bloomfield sprouting.
it's said to be $300 (slightly less then 50% more)
the 9850 came out a $235 and came down fast can now be had for less then $190.
still would like deneb to be 1-2% fast then penryn.
What's so laughable about it Chad?
That source is me. Here is the the email that I was sent as proof.
Not sure if anyone got back to you on this mate as I was out of the office. Our rep has said mid-January for them. Prices are expected to be much higher than the 9950.
Regards,
I then replied to him that Deneb is expected before the end of the year and got this reply.
Came from our AMD rep itself mate.
Now Chad I gave you a chance in the other thread that Fugger himself had to come and ban you from, then you go behind my back and start posting moronic crap like this
What's your problem little boy...? :shrug:
AMDZone is about 98% insane rants against Intel, and 2% general stupidity, so it only makes sense to qualify using anything from there as a source.
What a perverse interpretation of events. :shrug:Quote:
then you go behind my back and start posting moronic crap like this
What's your problem little boy...? :shrug:
we understand about amdzone already.Quote:
1. a. Flame Recruitment.
Posting a link to another website or forum with the intent of gathering XS members to participate in flaming, bashing, or any other form of attacking someone or something on the other site is forbidden. Furthermore, participating in cross-forum flaming will be handled as if the flaming were on XS itself.
:p:Well, I forget which post was about market share.. but I know personally, that when it comes to company's like mine, AMD has a hard time getting one of there product sold. The owner and brother of the company will not let AMD be sold to our clients due to an instruction set that was not available back in very early AMD chips, and there for they were unable to run a propriatary (how ever you spell that) software on the systems they sold. I had a hard time a couple years ago getting them to put dual core opterons in for some workstations when AMD had the upper hand. It is hard to gain market share when you are not being used. Intel was still selling tons of there notcompetive chips, while AMD was better and still trying to get partners. Then Conroe hit, and bam...AMD knocked down. They still hold a strong performance in the multi socket server space..but that isn't enough. The desktop side..well let's face it, AMD is behind. Although I feel that overall, the Phenoms are faster clock for clock then the k8's. Intel just has a better desktop chip out. Now Deneb, should be better then Agena..and if people are buying Agena...they will also get deneb since it will obviosly have less power draw and more cache and other things.
On that note: I wonder if the NDA will be lifted after tomorrow, since it is supose to be the official launch on Shanghai. SO in theory, since shanghai is basically deneb, we can get a idea of how it will perform..therefore no longer will the secret of 45nm be able to be kept....
To I type to much..
Post removed. - pointless
roofsniper, Nehalem IS coming out for server chips, 2nd Half of 09. With intel, and considering Nehalem is already shipping for desktop, that will mean early 2nd half, and they could easily pull it in if Shanghai is quite good.
AMD has about 2quarters advantage in the server market before they will face some extremely stiff competition(not that Intel is already the better choice in most regards for 1P and 2P servers, AMD still wins 4P though)
Also, of course Deneb/Shanghai will be more expensive than the older ones. Why they hell would they sell it cheaper than their older slower CPUs? That would completely osbourne massive amounts of stock and be an extemely stupid move.
As for pricing....
669 for top end X58 mobo here(450 for cheapest), and top end FX board is 393 , 9950 here is 381. So while the Corei7 is much more expensive it isnt even close to being a X58 board = 790FX+CPU+Ram.
Remember prices vary dramatically across countries depending on cross currency conversions and which cpu/gfx/mobo maker dominates a local market. Eg Nvidia is always way cheaper in NZ. The ATI cards just don't get the volume to compete on price or be cheaper than the equivalent Nvidia card like in the US.
Personally while I think Chad also has been a little harsh sometimes, some of you have also been just as immature. People could be much more polite and nicer here. I also only buy AMD ever, so don't think im some intel bigot.