Intel Plans 3.40 GHz Celeron E3900 for 2010

[btarunr]

Although a large chunk of products based on the new sockets LGA-1156 and LGA-1366 arrive in 2010, effectively pushing LGA-775 out of the extreme, performance, and upper mainstream segments, Intel will continue launching new low-end LGA-775 processors in an effort to digest inventories of motherboards and memory compatible to that socket. The company's grassroot processor brand Celeron will be updated with the E3000 series, consisting of dual-core processors based on the Wolfdale-1M core. The 45 nm based core offers the essential features for basic computing, while running cool and priced low.

Two quarters ahead of its expected launch, sources in the Chinese media unearthed Celeron E3900, perhaps the last SKU in the E3000 series. Slated for Q1 2010, the Celeron E3900 runs at 3.40 GHz (200 MHz x 17), with an FSB speed of 800 MHz. It has a shared L2 cache size of 1 MB, and supports the most common instruction sets including SSE3, SSSE3, and x86-64, although SSE4.1 and Intel-VT isn't part of it. The latest company roadmap of the essential and value segments show that the company looks as far as Q3 2010 with its Celeron and Core 2 Duo brand products, with existing number schemes. In essence, retailers in Q3 2010 shouldn't have stock problems selling you these processors.



View at TechPowerUp Main Site

 

[AlienIsGOD]

Interesting...I bet these will overclock nice!!

 

[AltecV1]

i was exited until i got to the celeron part

 

[LittleLizard]

17 multiplier

400*17 = Almost 7 ghz

This should be interesting

 

[AltecV1]

17 multiplier

400*17 = Almost 7 ghz

This should be interesting

getting a 800mhz to a 1600mhz = FAIL

 

[Fitseries3]

add this to the list of non existent cpus.

LIST:
e8700
e6500k
q7600
e3900

 

[btarunr]

E8700 (tested, listed, sold in China). E6500K (Chinese overclockers having a ball with this, doesn't look like it will ever make it to US). Q7600 and E3900 are upcoming.

 

[JrRacinFan]

17 multiplier

400*17 = Almost 7 ghz

This should be interesting

getting a 800mhz to a 1600mhz = FAIL

If it clocks anything like it's 65nm e1k counterparts, 400mhz fsb wouldn't be far fetched but 7Ghz would

 

[thezorro]

celeron, what a lousy processor
just my two cents.

 

[h3llb3nd4]

why are the chinese getting all the fun?

 

[FreedomEclipse]

Intel will continue launching new low-end LGA-775 processors in an effort to digest inventories of motherboards and memory compatible to that socket

Proof that LGA-775 aint quite dead yet! its dying yes but the day to poke it with a stick is still a long way away... Until stocks last....& with how there are so many choices to chose from out there & since LGA-775 IS a dying platform, people will be less inclined to buy it unless they want almost i7 performances on the cheap which then also means that the life of LGA-775 will drag on till they forcefully decide to stop supporting it all together.

 

[newtekie1]

45nm Celeron! I want I want I want!

getting a 800mhz to a 1600mhz = FAIL

i was exited until i got to the celeron part

It is just a name, the processor at the core is what matters, and the current Generation of Celerons have been pretty damn good. I've gamed on them.

And getting 800FSB processors up to 1600FSB is pretty easy, I got my E1400 up to 1600FSB pretty easily.

With this thing being 45nm, and having double the L2 of the E1000 series...I really want to play.

 

[KainXS]

400fsb(1600) on air is easy for pentium DC's, even my pos E2140 did almost 500fsb(2000) when I had it on water

 

[Mussels]

celeron, what a lousy processor
just my two cents.

its a WOLFDALE celeron. you know, the E8x00 chips use the same core?


its only got 1MB cache, but it will OC like made to make up for it with that multi.

 

[PCpraiser100]

OMG, its still alive. I wish that existed I would fry the CPU to living hell due to its sloooow performance.

 

[Cheeseball]

It'd be pretty hard to fry this particular version of the Celeron though, heh.

 

[Mussels]

It'd be pretty hard to fry this particular version of the Celeron though, heh.

let alone call it slow, its wolfdale based.

the celeron name doesnt mean what it used to - this is going to be one of those magic chips that overclocks madly.

And in the thread with the quad core version, where was the whiners? oh, they were all busy praising it just because it didnt have the celeron name.

 

[Phædrus]

With a multi that high it may be reaching the point where you derive no benefit from the extra clocks. You see, the multiplier is a ratio, of how many CPU cycles occur per bus cycle. In this case, that means that the CPU receives data from the bus--and then goes seventeen cycles without receiving more data! Commands that will take a full 17 clock cycles to execute are fairly rare--most are under 10 from what I understand. Thus you receive a set of commands to execute them, you finish them in say twelve clock cycles--then you're sitting around for another five waiting to send the answer back to the RAM and receive a new command. Thus 5/17 of your ultra-high clock speed is wasted in that instance.

So in this case, although having a low FSB with an extremely high multi means that you can overclock to ridiculous numbers (you can break the 4.0GHz barrier with just a 40MHz increase!), it may actually give less real world performance-per-clock-cycle than a CPU with a high FSB and low multi, like the E8x00 series.

You know, I wonder how this thing would perform if you dropped the multi to x12 and increased the FSB? Normally I don't like that form of overclocking because it puts more strain on the motherboard and increases instability. However, if this CPU is capable of such high clock speeds then there's no sense wasting them twiddling its thumbs waiting for instructions. A multi of x12 and FSB of 375 (1500... high, but manageable on high end boards) would give you a clock speed of 4500MHz. I expect that would be the most performance you could get out of this chip, and it's a lot of performance! Again, though, there's the issue of instability, and the bus wall. I don't know. Maybe a compromise? Multi = x14, FSB= 300MHz?

But though this is obviously appealing to overclockers who care more about clock speed than real performance, I think the real reason Intel has gone with this combination of low FSB and high multi is because it gives them the ability to put a processor with a very, very high clock speed onto very low-end motherboards that don't support the higher bus speed needed for CPUs like the E7x00 and E8x00 processors. Thus I expect to see a lot of these in 2010/11's OEM machines.

 

[MikeMurphy]

Sounds interesting but I doubt it will be a better value than the e5200 / e5300.

 

[MN12BIRD]

Celeron @ 3.4GHZ?

Its kinda like a Pagani Zonda with Wal-Mart all season tires on it!

 

[mrhuggles]

something I'd like to add, i have a 200mhz fsb cpu myself, its able to get to 400mhz fsb, but it wont do 500mhz, cpu side fsb limit. i have heard of alot of people having trouble even getting to where i am, i did 266mhz BSEL mod, wondering if that has anything to do with it being able to hit 400, and wondering if 333mhz BSEL mod would get it further, im not really sure if BSEL mod changes the FSB interface [timings? not sure what to call that, or if there is such a thing] wondering if this cpu would have the same issues as the e21x0 series does getting high fsb

EDIT: MN12BIRD: nice c64 dealy

 

[Mussels]

With a multi that high it may be reaching the point where you derive no benefit from the extra clocks. You see, the multiplier is a ratio, of how many CPU cycles occur per bus cycle. In this case, that means that the CPU receives data from the bus--and then goes seventeen cycles without receiving more data! Commands that will take a full 17 clock cycles to execute are fairly rare--most are under 10 from what I understand. Thus you receive a set of commands to execute them, you finish them in say twelve clock cycles--then you're sitting around for another five waiting to send the answer back to the RAM and receive a new command. Thus 5/17 of your ultra-high clock speed is wasted in that instance.

So in this case, although having a low FSB with an extremely high multi means that you can overclock to ridiculous numbers (you can break the 4.0GHz barrier with just a 40MHz increase!), it may actually give less real world performance-per-clock-cycle than a CPU with a high FSB and low multi, like the E8x00 series.

You know, I wonder how this thing would perform if you dropped the multi to x12 and increased the FSB? Normally I don't like that form of overclocking because it puts more strain on the motherboard and increases instability. However, if this CPU is capable of such high clock speeds then there's no sense wasting them twiddling its thumbs waiting for instructions. A multi of x12 and FSB of 375 (1500... high, but manageable on high end boards) would give you a clock speed of 4500MHz. I expect that would be the most performance you could get out of this chip, and it's a lot of performance! Again, though, there's the issue of instability, and the bus wall. I don't know. Maybe a compromise? Multi = x14, FSB= 300MHz?

But though this is obviously appealing to overclockers who care more about clock speed than real performance, I think the real reason Intel has gone with this combination of low FSB and high multi is because it gives them the ability to put a processor with a very, very high clock speed onto very low-end motherboards that don't support the higher bus speed needed for CPUs like the E7x00 and E8x00 processors. Thus I expect to see a lot of these in 2010/11's OEM machines.

those 10 cycles you wait for, are measured in nanoseconds. it doesnt affect performance - there are CPU's out there with massively high multipliers, and they dont perform any worse clock for clock to their lower multiplier cousins.

lowering the multi to find the sweet spot between CPU and ram is standard OCing practice. You'd assume people would do that, yes.

 

[Phædrus]

What I'm saying is that it's giving the illusion of performance. Compare two CPUs running at the same bus rate, one has a multi of x12 and the other has a multi of x17. Except for in artificial benchmarks the one with the x17 multi will show barely any improvement over the x12 one, since there are almost no commands that require a full 17 clock cycles to execute. If both have the same multi and one has a higher bus rate, you'll see much more improvement per clock cycle increase. The increase in real-world performance you see from increasing the multiplier is logarithmic--it's steep at first, but it shallows out, and by x17 the increase is very slow. Increase in performance by bus rate is linear. And by lowering the multi to x12 to get the best performance-per-clock and CPU-to-memory ratio you're essentially making this into a 45nm Wolfdale version of the E2220 with a bus wall of 400 instead of 350. It's nice, but not as nice as it sounds at first.

Intel is giving this CPU a deceivingly high clock speed, so they can get idiots thinking that because the number before "GHz" is bigger that this CPU is better than, say, an E8600. The x17 multi is a gimmick, and the only people it should excite are overclockers who don't realize that clock speed isn't always the prime factor in determining performance.

 

[JrRacinFan]

No offense and I'm not attempting to flame(bait) you, good information for those who don't know but tell me something I don't know Phaedrus. I would still like to try one out.

EDIT:
Gave you a thanks to let you know I'm not trying to sound like a douche or insult in anyway.

 

[Mussels]

No offense and I'm not attempting to flame(bait) you, good information for those who don't know but tell me something I don't know Phaedrus. I would still like to try one out.

EDIT:
Gave you a thanks to let you know I'm not trying to sound like a douche or insult in anyway.

dont worry about it. (cause he's wrong)

Phædrus: show me anything - anything, with reliable testing that has a speed difference, with the only change being the multiplier. same CPU frequency, same ram frequency, same HT link if on AMD.

I've seen people prove time and again that cache makes a difference (if not in every application) but in all the years i've been using PC's - back when multipliers were first introduced i have never, EVER seen ANY evidence that higher multis cause poor performance.