• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

OK, who's waiting for Nahalem?

The only thing that worries me, is which socket`s will intel keep and which one`s will they discard at a later time you think?
 
Last edited:
hyperthreading, mostly worthless due to not being a single core long pipeline incredibly inefficient design.

On board mem controller, questionable theres only so low you can get on latency. Core2duo does insanely well with latency, but it has a LOT of on die logic and massive cache levels to give it such low latency, they are essentially "replacing" much of that latency dropping predictive logic with the mem controller. latency drops aren't going to be massive like p4 to core2 or ath xp to ath 64 like drops so that in and of itself really isn't going to be improving IPC a lot. L1 cache is still very small per core, L2 cache is dropping to "amd" style levels, IE 256kb per core as opposed to the 6 mb per dual core they have at the moment(quad core having 12mb L2 on the high end) and will have 6MB L3 cache shared.

As to the ddr3 only, i do wonder if in a space saving design they won't have a budget range without mem controller with a much smaller chipfootprint, high yields per waifer and stick the mem controller on northy, which could bring about some boards with ddr2 support. If not the fact you can buy 8gigs of ddr2 for what, £100 now, and ddr3 is still over £100 for a 2gb set thats not really that good and the fact you will need 6gb's for triple channel all seems a little pointless.

Bandwidth isn't exactly massively needed either, the current chips really have very little and only take a small hit on the 4th core when all are loaded. Just on board mem controller does take the bandwidth limit off the limited FSB bus which boosts it to amd and beyond levels, they simply don't need more. Phenom is very similar in speed with single channel as it is in dual channel. Triple channel is really complete overkill and in need of expensive ddr3 mem.

This is all basically to give you a chip that 99% of common applications will never use the full speed of anyway. Games won't use a Q6600 fully for a couple years.

Am i waiting for an expensive upgrade to nehalem that is completely not needed with current mem prices or current and a year or two into the futures severe lack of power requirements...........no.
 
drunkenmaster said:
hyperthreading, mostly worthless due to not being a single core long pipeline incredibly inefficient design.

On board mem controller, questionable theres only so low you can get on latency. Core2duo does insanely well with latency, but it has a LOT of on die logic and massive cache levels to give it such low latency, they are essentially "replacing" much of that latency dropping predictive logic with the mem controller. latency drops aren't going to be massive like p4 to core2 or ath xp to ath 64 like drops so that in and of itself really isn't going to be improving IPC a lot. L1 cache is still very small per core, L2 cache is dropping to "amd" style levels, IE 256kb per core as opposed to the 6 mb per dual core they have at the moment(quad core having 12mb L2 on the high end) and will have 6MB L3 cache shared.

As to the ddr3 only, i do wonder if in a space saving design they won't have a budget range without mem controller with a much smaller chipfootprint, high yields per waifer and stick the mem controller on northy, which could bring about some boards with ddr2 support. If not the fact you can buy 8gigs of ddr2 for what, £100 now, and ddr3 is still over £100 for a 2gb set thats not really that good and the fact you will need 6gb's for triple channel all seems a little pointless.

Bandwidth isn't exactly massively needed either, the current chips really have very little and only take a small hit on the 4th core when all are loaded. Just on board mem controller does take the bandwidth limit off the limited FSB bus which boosts it to amd and beyond levels, they simply don't need more. Phenom is very similar in speed with single channel as it is in dual channel. Triple channel is really complete overkill and in need of expensive ddr3 mem.

This is all basically to give you a chip that 99% of common applications will never use the full speed of anyway. Games won't use a Q6600 fully for a couple years.

Am i waiting for an expensive upgrade to nehalem that is completely not needed with current mem prices or current and a year or two into the futures severe lack of power requirements...........no.
Click to expand...


Very nicely put, a cynical point of view backed with good facts and an obvious depth of knowledge...A pleasent change from the arrogance of todays regulars :)

So, this nehalem sounds pretty pants then? Is that what your effectively saying? Not the revolution some are implying it might be?
 
Drunkenmaster the integrated memory controller will also serve to lessen the difference between motherboards, the P35 motherboards for instance do have a bit of extra memory bandwidth over the P965 chipsets.

DDR3 will fall in price, the current DDR2 market is just fantastic for end users but 2 gigs of DDR3 currently costs less than my first 2 GB DDR2 sticks cost just 1 year ago, that must mean something.

And you do have to give AMD credit for getting massive bandwidth on the memory and low latencies, which in combination often give more snappy systems in Windows, as per my own experiences.

Also Nehalem will serve as a refined 45nm processor, i'm sure they will be able to reach further into 3ghz stock than we've seen so far, and this is a good thing.

I'm really looking forward to Nehalem, integrating the memory controller is a step towards centralizing everything on the chip, and eliminating a difference in motherboards that has been a source of concern for me. I know that Nvidia have done a good job at making efficient memory controllers on the northbridge but really, all chipsets under the same chip design should have the same high performance in regards to memory.

3 channel memory is friggin overkill though, agree with that.
 
with games at the moment are we limited by graphics cards more than processing power? I think its going to be a pretty huge investment for very little gain (currently anyway) I mean the motherboards will easily be £200+ the cpu's i dread to think (low end probably £200+) the buying enough ddr3 to take advantage of the triple channel (£200+ going by current prices) it just seems like an insane amount of money to spend on something that wont even see a decent performance return for a good while. I say just wait till there is a need for it, most games arent even optimised for quad core yet so whats the rush?
 
I went from P4 570 to X2 4400 to E6700 to a Q6600. Each one gave me a massive boost from the previous. Don't see anything doing that to my Q6600 for some time tbh, maybe nehalem... maybe not!
 
Motherboards will not be those prices, DDR3 will be much cheaper, it's already reaching fair prices, and processors would not be any more expensive than the current ones, except for early adapters. This is a completely new socket, platform and an investment for Intel, why would they only make new technology available for the high-end?

The rush is also singlethreaded performance, it's increasingly important to address this as well, seeing as developers are reluctant to invest the time and money to develop efficiently for quad core processors. Therefore if you can get any more speed(clock frequencies) or efficiency(instructions per cycle, how much is done in one single clock cycle) Intel needs to go along with this. DDR3 is a natural evolution, no one would suggest anything differently, even though the bandwidth is not currently needed, why not have it along with lower power consumption?

Don't forgot that the RAM manufacturers said that DDR3 will be only 10% more expensive than DDR2 very soon, that's interesting to say the least.

And it's not like gaming is the only use for CPU's, we're all more or less enthusiasts here, we should be interested about other prospects as well. I never thought i'd actually need it but i have done movie work lately where i could have used a faster CPU, and have been very interested in POV Ray and would like to run that program faster just for the heck of it.

Let's see if these processors are better clock for clock, i don't see why Intel would launch them without this being the case, and if the boost is as large as C2D was, we're in for another great ride. If not well, AMD finally has a chance to get back into the market we all once loved them in.
 
Talrinys said:
Motherboards will not be those prices, DDR3 will be much cheaper, it's already reaching fair prices, and processors would not be any more expensive than the current ones, except for early adapters. This is a completely new socket, platform and an investment for Intel, why would they only make new technology available for the high-end?

The rush is also singlethreaded performance, it's increasingly important to address this as well, seeing as developers are reluctant to invest the time and money to develop efficiently for quad core processors. Therefore if you can get any more speed(clock frequencies) or efficiency(instructions per cycle, how much is done in one single clock cycle) Intel needs to go along with this. DDR3 is a natural evolution, no one would suggest anything differently, even though the bandwidth is not currently needed, why not have it along with lower power consumption?

Don't forgot that the RAM manufacturers said that DDR3 will be only 10% more expensive than DDR2 very soon, that's interesting to say the least.

And it's not like gaming is the only use for CPU's, we're all more or less enthusiasts here, we should be interested about other prospects as well. I never thought i'd actually need it but i have done movie work lately where i could have used a faster CPU, and have been very interested in POV Ray and would like to run that program faster just for the heck of it.

Let's see if these processors are better clock for clock, i don't see why Intel would launch them without this being the case, and if the boost is as large as C2D was, we're in for another great ride. If not well, AMD finally has a chance to get back into the market we all once loved them in.
Click to expand...

i hope there is truth in these words, i would love a huge performance boost out of these and if the prices are reasonable hell yeah i will get them but i really doubt that they will be.
 
StriderX said:
Im waiting for sandy bridge as it looks more promising, its the the last chip on intel roadmap.
Then the infamous 2012 chip :).
Click to expand...

that's liek talking about the infamous 2003 chip...

you realise you're talking about northwood? urgh, dirty netburst ;)

(I loved pentiums 4s!)

Point being thats a LONG way away, filed well and truly under "may never happen"
 
I'm not saying its going to be pants, or slow. But we are used to going from, i dunno a 1.6ghz p4 slowly over 3-4 years to 3.4Ghz or so, at a couple hundred mhz every few months. Then out of no where we simply doubled performance, then again, a fairly damn short amount of time later we doubled that again. The gains were stagnant for a long while in the cpu arena and then suddenly we really shot through the roof.

Sure some people use POV ray, but at the end of the day 95% of the forum, probably more, really do only intensively use their PC's for gaming. Office stuff, most work at home, surfing, watching vids its all a breeze for the lowest end £40 cpu you can get right now. Gaming its almost impossible to tell the difference between a dual core at 2.4Ghz, and a quad core at 4.5Ghz, in encoding aswell. To be honest most people don't encode or render graphical work, but those that do tend to leave those things working overnight anyway so it doesn't make a huge difference.

I'm sure Nehalem will provide a boost to IPC, its just most people mention the IMC as if the new super low latency will somehow boost IPC on its own. Its a modest change, mostly because the core logic for predicting branches on the C2D is INCREDIBLY good, and they are all but mutually exclusive technologies. What it does offer is getting rid of the FSB< but again, that has limited uses, in theory its nice to have a potential 50gb/s bandwidth(and that potential is there btw ;) ) but at the end of the day again, going from 5gb/s to 10gb/s barely gives a gain at all, so that 10 up to 50gb/s is going to go almost unnoticed.

The IMC isn't that small, probably takes up similar space to the cache its replacing, but the fact that the cache isn't there tells you it would have been to big a die, L2 cache is still massively faster than L3 and would have helped, there just wasn't space, so i don't see great prices coming on Nehalem parts.

AS to why would they not give it to us on mid/low end range. Well intel have done that lots of times before, P4 extreme editions had more cache, they were really xeons, rebadged and stuck in a different socket. Its long been a rumour that Intel might go back that way, extreme edition is actually a xeon with the ondie mem controller, and mid/low end versions will lack the IMC. We'll see, the only way to really save money would be to still keep a low amount of cache. We'll have to see tbh.

In terms of gaming Crysis is about as cpu intensive as it gets and my Q6600 overclocked doesn't even hit 50% load, i can't see games using much more than that within the next couple years.

AS to memory prices, its been a long time you could get ddr2 very cheap unless buying a fancy overclocking pack that frankly has always been the same mem as on crucial value sticks for a heck of a lot cheaper.
6 gigs half decent mem will cost what, at least £300, maybe more like £450? for a cpu upgrade that you'll be hard pushed to notice at all.

Now also, not saying there will be issues, but sticking something running at different speeds, diff voltages, and more complex logic on the same bit of silicon is fairly difficult. Theres no guarentee that the mem controller will overclock well.
 
No there is not, but it's still exciting technology and while there is the potential that Intel are in way over their heads on the first run, at least the 32nm shrink has the opportunity to get more cache on there and clock better in the process.

6 gigs will simply not cost that, it'll cost at max 200£, that's your word against mine but we'll see in even a couple of months time how much it has dropped. The memory fabs are all moving to smaller processes right now, and they are all talking about prices on DDR3 falling, some of that drop already evident on OcUK.

And well i'm just a fan of getting it onto the die, i'm not really sure i can argue for that, but besides the extra bandwidth which may or may not be needed, it just sounds more simple and less of a hassle to pick motherboards that way. And really, no one has ever proven that you need a Q6600 for anything, i just think it's great that we actually have cheap processors that are able to run all the newest games. If Nehalem gets even better processors here, the chips will just have an even larger head start before graphics cards catch up.

Also i wonder if some of the extra bandwidth gained from DirectConnect or whatever they called it isn't related to the possibility of an integrated GPU, i'd guess those will require quite a bit of bandwidth, espically if that GPU is overclockable.
 
to be honest, intergrated gpu's won't be anything special for a long while to come. When we hit say, at least the 16 core, or maybe 24/32 core stage only then will it be suitable to drop in multiple "gpu" styled cores for speed. Where you might get a 31 cpu cores + one gpu for desktop for ultimate encoding, rendering or server performance. but there could be 16/16 versions with 16 cpu/16 gpu cores for a fantastic gaming chip. But with only 4 or 8 cores you're droping a significant amount of power to drop in gpu cores.

I don't think we'll get a big boost in cache from the next die shrink either, cache is an exponential usage situation. say the prediction unit predicts this particular thread might need one of 2 different bits of info so it stores them, then each or those might use 2 after that, and each of those 2 after that you end up with , 2-4-8-16-32-64 etc bits of into ready to be streamed into the core. With relatively high latency access to the memory but not realistically hugely less bandwidth, the key is to be constantly preaccessing that info rather than constantly waiting. So the more steps you store in cache you need exponentially more cache. But when you cut that latency beyond the 3rd or 4th step where you'd need to stream in silly amounts of extra data, its simply quicker to access the memory with an onboard mem controller. So large cache is simply not needed in the slightest on a chip of that design. Really an L3 big enough to store the next 3 bits of info is more than enough and really, that L3 design is going to be based more on number of cores. more cores more data it can get through, give it a proportional increase in L3 cache.

One thing i dislike about ath 64 and onboard mem IS the lack of motherboard upgrades. 965 was decent, but overclocking on the p35 is noticably a lot more reliable in regards to getting much higher. When you have an onboard mem controller tweaks to that controller come out VERY rarely and mostly with a big upgrade. For instance, if the Phenom's mem controller is whats holding it back right now, as say a 790fx board can do 300-350Htt with a X2 easily, but 220ht is hard for a phenom. A simple new spin of a VERY simply northbridge could be a simple quick fix. But we had to wait a heck of a lot longer for the B3 silicon from AMD and even then theres nothing known about its HTT overclocking ability. Through my fairly long use of the 939 there was only really the NF4 available, which I didn't really like. Nvidia had no reason to upgrade it, there was really very very little to upgrade as 99% of the performance was down to the chips not the northbridge version.

An intergrated memory controller works very well on the northy, gives more than enough bandwidth. A simple move from a traditional Fsb quad pumped thing from Intel to something more updated and faster along with triple channel memory would have been far easier, cheaper, faster and easier, and allow multiple versions of motherboards and ddr2 use would have been easy to put in.

I'm not a fan of the mem controller as i think its hurt AMD quite badly in some cases, though helped in others. At the end of the day, if intel or amd make a crappy design with poor overclocking a new mobo chipset won't help with that, and a long wait for a better clocking chip is in order. Also, the IMC uses a fair amount of heat and you are moving it onto the cpu. Cooling a AMD chipset is incredibly easy as theres just no heat there. Intel have difficulties there, we'll see if it works out or not.

EDIT:- just checked the likes of Dell and on their highest end gaming system, no sign of DDR3 anywhere at all, even on their 2.3k gaming system with 2gigs of mem its £90 to upgrade to 4gigs of ddr2. Which to be honest is fantastic in Dell's terms, nto long ago it would be £300 to have 4gigs of ddr2 in one of their systems, it just shows how cheap ddr2 has gotten.

The thing that brought about cheaper ddr2, as in the drop from £200-250 ddr2 kits down to £100-150, was when 775 came out with from what i recall only ddr2 use. Even then, if i recall correctly, it was expensive until Dell and the likes started to sell more 775 systems and eventually ddr2 prices dropped. We won't hit the point where DDR3 is actually needed, until a chip that only supports it comes out, that IS Nehalem and DDR3 won't be cheap till after that happens. 1 in every probably 10000 systems sold uses ddr3 at the moment, memory makers just aren't making it as no one really wants it. When it shifts as from ddr/ddr2 to higher production which will only happen when demand goes through the roof, prices will drop and fast.
 
Last edited:
no thats not right, 775 was ddr2 only from day one which was something like 2003. Not sure what made the price collapse tbh, I expect it was some sort of improvement to the fabbing process.
 
You must log in or register to reply here.
Back
Top Bottom