Deneb just around the corner?

[Sentential]

I suspect hitting 4 gig will be notably less common than it is for Penryn Quads.


That is quite a perverse interpretation you got going for yourself there.

Bulldozer will be a new architecture, not Deneb.

Ok both statements listed above are complete bull. Easier to hit 4 on penryn than it will be on Deneb? You gotta be kidding me; clearly you dont own a Q9300/Q9400 on any kind of SLI or Xfire board Show me ONE person on these forums who has 533mhz FSB+ and runs any kind of multi GPU setup without using an ES (you wont find anyone).

Again its been said multiple times from several sources that Deneb will be close to or match penryn which @ 4ghz (or even close) will make it quite competitive with i7 in most games. Deneb will lag in multi-threaded setups however their power consumption will be significantly lower which will still make them competitive in the server markets.

The key thing you fail to grasp is how cache dependant AMD cpus are and always have been. Do you remember the days of 128kb Semprons whooping 6mb P4EE's? No. So what do you think its going to happen when they give current Agena's who are about 25% down from i7 50% more cache and up the extensions by 33% and reduce the latency on the L2 and L3 caches by the same amount?

Ive constantly swapped between intel and amd depending on which is the better of the 2 processors and in this occasion Deneb will end up on top for a whole host of reasons:

Primarly the main reason why I moved back to AMD was because of Intel's move to integrate the IMC to the CPU. That was also the same reason I left AMD 3 years ago. Why? When they moved the MC to the cpu die they sacrificed reliability to save die space and as a result there was significant corrpution and instability with a whole host of A64s. It happened when clawhammer was first launched and happened again with Winchester.

Just like now the initial samples did not show any issues UNTIL it went retail. I gaurn-damn-tee you the intel forum is going to be filled with people having stability issues with RAM. Infact its already started with people not being able to change their memory dividers on certian boards. If its bad now on bloomfield, lynnfield will be even worse why do you think it was delayed because they had to re-wire the CPU socket?

Just like we saw with Penryn we will be limited by the multiplier (since it will be locked on retail chips in addition to the ram multi) I seriously doubt we will see *ANY* i920 or i930 over the 3.5 mark for anything other than benchmarks because there isnt a single board that is able to get over 200 QPI with any sort of reliability. And dont bother citing the one 920 on the forums now because the load temp he is getting is 80c*+ on suicide runs, it by no means is stable nor is it even retail chip to begin with.

AMD's platform for better or worse is far more mature than intel's so even if its 5 or 10% slower than i7 Ill take that anyday if it means it wont crash all the time and I dont have to spend my money on very expensive DDR3 which will be a requirement to even get i7 close to the 3.5 mark

[soundood]

Sentential this guy has posted basicly 3 pages of bull, he really hasnt got a clue what he is on about.

read back (if you can be bothered)

its sort of funny, the inane comments he has come away with

we really need to start to ignore these intel fanboys and wannabes, heres an idea, shall we have like an International Alarm?

like the US Advisory System or something?

hows this one?



that way we all know when something is up....

Chad Boga heres yours

[Sentential]

Maybe so but this is the same thing I told Gautam:

So you're telling me that Nelhalem using tri-channel ram WITH an imc that is ring-bus based controller and 8 threads only beats Penryn by maybe 5% in heavily multi-threaded games? And it does this while dissapating 175W+ (high 60s low 70s load temps, ie Smithfield)?... Wow intel is hosed

This is the same thing people were saying when AMD had 2 consecutive issues with scaling when a new fab process rolled out. I and others said this was bad news for AMD long term and now look at where we are.

Just like back then the only thing AMD had to keep the Pent 4s at bay was efficiency because despite being a brand new tech it all scaled like crap. While going up against a 5 year old design (which is exactally what is going on now) that was worn out and well past its prime.

The fact that nelhalem scales poorly and has significant heat dissapation without bringing anything notable to the table in raw thoroughput is a BAD sign for intel.

[mesyn191]

The fact that nelhalem scales poorly and has significant heat dissapation without bringing anything notable to the table in raw thoroughput is a BAD sign for intel.

Not really. Intel targeted Nehalem for processing lots of threads vs. emphasis on single threaded performance. It still does quite well on single threads, at least as good as current C2D's, its just not a huge jump like some people are expecting for some reason. Intel (and IIRC AMD) has said many times that many-core is the future since the effort needed to get more single threaded performance became burdensome a long time ago and is now becoming impracticle. At some point in time you're going to see chips from Intel/AMD which have down right crappy single thread performance on a per core basis, but will have hundreds if not thousands of cores. Nehalem is just Intel's 2nd step in that direction (IIRC quad/dual cores were their first).

[Sentential]

Not really. Intel targeted Nehalem for processing lots of threads vs. emphasis on single threaded performance. It still does quite well on single threads, at least as good as current C2D's, its just not a huge jump like some people are expecting for some reason. Intel (and IIRC AMD) has said many times that many-core is the future since the effort needed to get more single threaded performance became burdensome a long time ago and is now becoming impracticle. At some point in time you're going to see chips from Intel/AMD which have down right crappy single thread performance on a per core basis, but will have hundreds if not thousands of cores. Nehalem is just Intel's 2nd step in that direction (IIRC quad/dual cores were their first).

I'm completely with you on this one, the problem that I have is that this chip is supposed to be the basis for all of their new cpus for the next 5 years or so and if this is all they have to offer they are in trouble.

I understand the arguement about threading, I do, but my concern is that we are comparing a brand new arch vs a design that is atleast 4 years old (Clawhammer -> Agena)

It would be one thing if Nelhalem was a 16/32 core chip with 64k of cache per core and it wound up peforming this well but its not, its a quad with hyperthreading and tri-channel RAM that provides zero peformance gain in multi-threaded games.

So what should we expect of lynfield? Is it going to be worse than C2D? By all accounts it should be. As for bloomfield and 1366 I believe this socket will quickly dissappear just like Socket 940 did back in the A64 days leaving only LGA1158 as being the widely adopted platform which will have less peformance, and poorer overclocking than bloomfield does.

Then what does this say of K11 when it arrives sometime late next year? Multi-thread or not if they dont get the peformance up they are going to get whomped by whatever AMD has to offer come next year.

This is the same arguement made about physics cards. Oh its wonderful this, oh its next-gen that. Bottom line is that it didnt deliver. Sure it may be future-proof but what good does it do if no one programs for it?

Hell 64bit procesors have been around over 5 years now and we have still yet to see any widespread adaptation to the new format. So what does that say to this philosophy of "more cores?"

Believe you me Ive got no axe to grind here, hell I argued that Pentium Ds were a better buy back in the hay-day of the A64s because of the early issues with AMD flawed initial mem-controllers and said that Conroe would *destroy* any market share AMD had gotten once it was released and I was ridiculed for it.

In this occassion I just dont see Intel as viable anymore. The heat dissapation issues with Nelhalem and its lack of ability to scale is going to ultimately doom Intel unless they can fix this with the 32nm process.

The only way that Intel is not going to get overrun next year is if they can get some serious clock scaling because if AMD makes any substantial changes to their arch they will quickly catch up to Intel and we will be right back to where we were back in the Netburst days. Intel will once again have a bloated chip that is only competitive due to its enormous cache size because it was made on a smaller fab process.

We are pretty much at the same exact apex right before the release of Clawhammer back in 2002. Intel had a faster chip that had significant heat dissapation and cost significantly more than the AMD equivalent. AMD chips were cheap as hell and when on certian combinations could come very very close to their Northwood counterparts. AMD was left with an ageing arch and were thought left for dead since they were a good 10-20% IPC below current HT enabled P4s.

Then AMD after many many delays finally got a viable new arch and then everyone hopped back on the bandwagon.

Only difference is I'm hopping on now while I can get top dollar for my existing C2D equipment

[mesyn191]

I'm completely with you on this one, the problem that I have is that this chip is supposed to be the basis for all of their new cpus for the next 5 years or so and if this is all they have to offer they are in trouble.

5yr.? Source please.

I understand the arguement about threading, I do, but my concern is that we are comparing a brand new arch vs a design that is atleast 4 years old (Clawhammer -> Agena)

Its my understanding that K10 has about as much to do with K8 as C2D had to do with the PIII, strongly related/inspired by yes but not the same thing.

It would be one thing if Nelhalem was a 16/32 core chip with 64k of cache per core and it wound up peforming this well but its not, its a quad with hyperthreading and tri-channel RAM that provides zero peformance gain in multi-threaded games.

I'm sure I've seen benches where it does great in UT3 based games and that is heavily multi-threaded, AFAIK not much out there ATM to compare...

So what should we expect of lynfield? Is it going to be worse than C2D? By all accounts it should be.

Can you expand on this? I didn't know we had heard so much about Lynfield yet. Same in regards to K11 too.

This is the same arguement made about physics cards. Oh its wonderful this, oh its next-gen that. Bottom line is that it didnt deliver. Sure it may be future-proof but what good does it do if no one programs for it?

AFAIK both major x86 CPU makers are going in the same direction, just taking slightly different routes to get there, if programmers want to get performance out of either of them they won't have much choice.

Hell 64bit procesors have been around over 5 years now and we have still yet to see any widespread adaptation to the new format. So what does that say to this philosophy of "more cores?"

Historically it normally takes ~10 yr. for any major architectural changes to achieve widespread use in programming. There were similar issues when switching over from 16 to 32 bit software IIRC.

[Sentential]

5yr.? Source please.

Thats generally about how long each major revision lasts generally speaking (or atleast these last couple of years it has been)

Its my understanding that K10 has about as much to do with K8 as C2D had to do with the PIII, strongly related/inspired by yes but not the same thing.

K10 and K8 are pretty much identical iirc the only thing that seperates them is the cache sizes, latency and instruction sets but the basic building blocks are the same which is why their IPC is almost identical. K8 to K10 chips are about as similar northwood to prescott/pressler. Ya its a different chip but its more or less the same.

As far as conroe vs p3 ya you could make that arguement as well but the chief difference would be conroe's out-of-order read/write in its internal pipeline plus a whole bunch of tweeks.

I'm sure I've seen benches where it does great in UT3 based games and that is heavily multi-threaded, AFAIK not much out there ATM to compare...

Ive seen a couple crysis benchmarks and those are threaded, frankly I would like to see a world of warcraft comparison as that is probably the most highly threaded game I can possibly think of.

Can you expand on this? I didn't know we had heard so much about Lynfield yet. Same in regards to K11 too.

Lynfield is identical to Bloomfield with the exception that lyn is strictly dual-channel and in addition it incorporates a PCI-E controller on-die like Phenoms do

AFAIK both major x86 CPU makers are going in the same direction, just taking slightly different routes to get there, if programmers want to get performance out of either of them they won't have much choice.

Again I agree, but Nelhalem is a step in the wrong direction. By all accounts it ought to slaughter Penryn by 30% or more because the main bottle-neck for C2Ds was the available bandwidth between the cores / memory. So one would think that there would be a difference between dual and tri channel. So far its only a measly 200 points... thats NOT a good sign.

Historically it normally takes ~10 yr. for any major architectural changes to achieve widespread use in programming. There were similar issues when switching over from 16 to 32 bit software IIRC.

Ya again I agree, the problem is that Intel is betting the farm on the programs actually being there for this CPU to shine, so far it hasnt happened nor does it look like it will anytime soon.

[Chad Boga]

Chad Boga heres yours

I find it really strange how you have chosen to identify so closing with the corporation AMD.

Do you have difficulties forming normal relationships with people and have become attached to AMD as a result?

[soundood]

Do you have difficulties forming normal relationships with people and have become attached to AMD as a result?

err mmmm errr yes Chad Boga, please tell me how i can make some friends? maybe in the intel forum? while your there, maybe you can make some too, and right off? lol

[biohead]

err mmmm errr yes Chad Boga, please tell me how i can make some friends? maybe in the intel forum? while your there, maybe you can make some too, and right off? lol

don't feed the troll.

[Chad Boga]

don't feed the troll.

The trolls are the imbeciles determined to turn this forum into a mirror image of the mentally disturbed AMDZone.

[informal]

The trolls are the imbeciles determined to turn this forum into a mirror image of the mentally disturbed AMDZone.

Why do we care what you think about this forum?
Oh and if you have some unfinished business with amdzone go there and sort it out,stop mentioning that website in almost every post of yours since it's getting old.
As you can see from the posts in this thread,even intel users came in to tell you to stop your useless trolling.Take their advice and stop boring us with your useless rants here.

[soundood]

The trolls are the imbeciles determined to turn this forum into a mirror image of the mentally disturbed AMDZone.

where you are moderator?

Chad have you thought of killing yourself and your family?

i donno it seems apt.

how are those voices anyway? still screaming at you?