i3-4370 - Dual Core or "Nerfed" Quad Core

A

Avexrion

n00b
Joined
Jul 23, 2015
Messages
15
Since 2010 or so, it looks like Intel has had two main dies of processors for their mainstream lineup. A quad core based one with 8MB L3 cache, and a dual core with 4MB. Intel then takes these two lineups and decides whether they are mobile or desktop, Xeons, unlocked, and so forth.

This is not outright confirmed, but similar TDPs between dual and quad core variants seem to support this along with actually looking at the mobile processors themselves. Intel does not seem to like divulging this information though.

For example, look at the i7-4900MQ and the i7-4600M which are quad and dual cores respectively. You can then look at a delidded i7-4790K and Pentium G3258 which should be similar (except the G3258 has hyper threading and 1MB of cache lasered off.)

Therefore, I assume that the desktop i3-4370 is a fully enabled dual core chip, the i3-4170 is the same thing with 1MB cut off, and the Pentium G3258 with more "nerfs" but with an unlocked multiplier.

Anyone delid these i3s (like the i3-4370) and find evidence for this? I am assuming that Intel would not necessarily salvage a quad core to make a dual core i3 as it might be considered too "defective" at this point.

Thanks!
 
defaultluser said:
Duh?

This is not news buddy. Intel either lasers off bad cache blocks, or just disables good cache to create multiple SKUs from the same hardware.

They've been doing this since the Pentium III Coppermine, with the Copppermine-128 processors branded as Celerons.
Click to expand...

You just restated what I said in my post.

I am trying to see if they do this for the i3s, or if they are native dual core processors. Not that there are lesser processors cut down from full dies (that's already known.)

There is some supporting evidence for this with Skylake. The quad cores i7-6700k and i5-6600k are both the R0 stepping, while a dual core like the i3-6320 has a S0 stepping. Usually a complete stepping change like this indicates a completely different die.

All Haswell products have a stepping of C0 though, regardless of it being the i7-4790k or the i3-4370 (so there is no evidence of different dies unless you physically look at them.)
 
Last edited:
defaultluser said:
Anandtech has done the analysis of chips Intel makes. They make a native 2-core processor for every single core revision.

They have in quantity for Haswell:

2 core + 2 GPU slice
4 core + 2 GPU slice

And in smaller quantity, versions of tthe above with 3 GPU slices.

Intel makes 4 CPU cores layouts. Search Google for Haswell introduction articles at Anandtech where they discuss die sizes.
Click to expand...

There we go, thanks for the information. I think I came across those before, but I wasn't sure of how accurate they were. I'll look at those again.
 
Avexrion said:
You just restated what I said in my post.

I am trying to see if they do this for the i3s, or if they are native dual core processors. Not that there are lesser processors cut down from full dies (that's already known.)

There is some supporting evidence for this with Skylake. The quad cores i7-6700k and i5-6600k are both the R0 stepping, while a dual core like the i3-6320 has a S0 stepping. Usually a complete stepping change like this indicates a completely different die.

All Haswell products have a stepping of C0 though, regardless of it being the i7-4790k or the i3-4370 (so there is no evidence of different dies unless you physically look at them.)
Click to expand...

The i3 is a different core all together, its their native dual core CPU, which they use for their Pentium chips as well.

i3's get turned into Pentiums (Celeron)
i7's get turned into i5's

defaultluser said:
Duh?

This is not news buddy. Intel either lasers off bad cache blocks, or just disables good cache to create multiple SKUs from the same hardware.

They've been doing this since the Pentium III Coppermine, with the Copppermine-128 processors branded as Celerons.
Click to expand...

They've been doing this since the 486 at least, their first Celeron (Pentium era) came from a normal Pentium with the cache shaved up.
 
Avexrion said:
You just restated what I said in my post.

I am trying to see if they do this for the i3s, or if they are native dual core processors. Not that there are lesser processors cut down from full dies (that's already known.)

There is some supporting evidence for this with Skylake. The quad cores i7-6700k and i5-6600k are both the R0 stepping, while a dual core like the i3-6320 has a S0 stepping. Usually a complete stepping change like this indicates a completely different die.

All Haswell products have a stepping of C0 though, regardless of it being the i7-4790k or the i3-4370 (so there is no evidence of different dies unless you physically look at them.)
Click to expand...
I don't believe the i3's are cut down i7's with 2 cores disabled. They are likely native dual core dies. I'm assuming this is because they sell a helluva lot more i3's than i7's and you could quite literally produce twice as many i3's on a single die than i7's.

It would be wasteful and costly to use the same die as the i7's to produce i3's, unless there was a manufacturing issue where they had to salvage what they could from the high end dies. A prime example of this situation occurred with AMD and their Phenom and Athlon II lines when they were making dual core and triple core processors from failed native quad core dies.


I'm fairly certain Intel has separate dies for most of their core counts, then separate SKU's are made with those dies. There are a few notable exceptions.

Examples:
-Desktop i7's, i5's are made from a quad core die. i5's are i7 dies with HT and a small amount of cache disabled.
-Desktop i3's, certain Pentiums, and certain Celerons are made from the same native dual core die. There are a few Pentiums and Celerons that are i3's with disabled HT and some disabled cache. Others are completely different cores and dies like Celeron J series (Bay Trail cores, replacement to the Atom series).
-E series and Xeons likely share the same dies, and a few inbetween chips maybe cut down higher end chips (possibly the 5820k). There are a few oddballs here as well, like the E7-8890v2 which is a native 15 core chip.
 
Last edited:
Yeah, this is what I believed as well, I just wanted some more people to weigh in what they thought.

My assumption with the binning process was something like:

Does it run with low leakage? Then it is a mobile processor,
Else, does it have all parts working? Then it is a xeon,
Else, is most of it working and clock high? Then it is a K i7
And this process would just continue until the processor is so terrible, they have to pitch it.

They seem to cut off cores in their 2011 processors, but the architecture of those (the "ring" style) seems to be more forgiving to that type of modularity. However, the mid-range stuff usually has all cores, they just laser off other things.

Now that just makes me want a K series i3 even more. A native, fully enabled, dual core chip would be reminiscent of Core 2 Duo. I'd buy that almost instantly.

Thanks again.
 
Closest thing to an i3-K is the G3258.

The 2011's HEDT processors are "disabled" Xeon's, kind of like the i7's and i5's.
 
Trimlock said:
The i3 is a different core all together, its their native dual core CPU, which they use for their Pentium chips as well.

i3's get turned into Pentiums (Celeron)
i7's get turned into i5's



They've been doing this since the 486 at least, their first Celeron (Pentium era) came from a normal Pentium with the cache shaved up.
Click to expand...

It was equally true with Celeron and Pentium Dual-Core and with the Celeron-G and Pentium-G of today. In fact, I looked up the feature set of the G3258 (which I am running right now) and the i3-4130, and found three (and only three) features missing form the cheaper part:

1. No HyperThreading Technology. (No surprise there. If it actually had it, it could be an i3-killer - instead of causing the pricier processor a ton of grief, which it does today by being a very attractive upgrade.)

2. No support for AES New Instructions. (Like the missing HTT, not surprising.)

3. There is one other (seldom used) feature missing from the G3258 - however, cache is NOT it; the G3258 has the same 3MB on-die as the 4130.

The bigger surprise is what it retains from the i3 it's based on - in addition to the cache, it retains the Intel HD Graphics from the i3 (VERY useful in streaming or game-recording - such as XSplit GameCaster). It also retains support for Hyper-V (just as the Celeron DC and Pentium DC retained the VT-x support from the Core 2s they were based on). The upgrade (from VT-x to Hyper-V) solidified the status of these sub-i3s as utility-infielder CPUs - in portables, low-end desktops - or even low-end workstations. (Remember, Hyper-V was added to Windows 8 - the issue for Celeron DC and Pentium DC was lack of EPT support - CeleronG and PentiumG fix this lack.) Considering Windows Server 2016 as workstation OS? Server 2016 has the same Hyper-V support requirement as Windows 10 or 8.x - EPT support in the CPU is a must. Again, CeleronG and PentiumG fix that issue.
 
There is one big thing that the pentium and celerons don't have... and that's AVX 2.0
 
Keljian said:
There is one big thing that the pentium and celerons don't have... and that's AVX 2.0
Click to expand...

What use does AVX 2 have on desktops - or workstations - right now? If it's a niche use, that's not really a lack. (It's like EPT and the Celeron DC, Pentium DC, and Core 2 - it was a niche use - and only on servers - before Windows 8 came along, and while Servers (Server 2003 and later) could leverage it, none required it - that changed ONLY with the release of Server 2016.) Virtualization in general (irrespective of Hyper-V) IS getting bigger in terms of desktops - however, that was due mostly to Core 2 and derivatives. Hyper-V (and EPT) is, for the most part, an upgraded VT-x (as far as users, and most uses, see it) - which is also part of the point for the Core i-series, and the Celeron and PentiumGs. That is why I called this whole subseries "utility infielder" CPUs, as they do a lot of what their bigger brothers do, without including features that the masses won't use (and all of that at a lower price tag), and if your CPU is socketed, you can still "drop and swap" - if something comes along that requires AVX 2, I can replace the G with an i5-4670K (for example) without replacing anything else. (Naturally, the same applies to any other "missing feature".)
 
Avexrion said:
You just restated what I said in my post.

I am trying to see if they do this for the i3s, or if they are native dual core processors. Not that there are lesser processors cut down from full dies (that's already known.)

There is some supporting evidence for this with Skylake. The quad cores i7-6700k and i5-6600k are both the R0 stepping, while a dual core like the i3-6320 has a S0 stepping. Usually a complete stepping change like this indicates a completely different die.

All Haswell products have a stepping of C0 though, regardless of it being the i7-4790k or the i3-4370 (so there is no evidence of different dies unless you physically look at them.)
Click to expand...

Apparently, the same C0 stepping applies to the sub-i3s as well (my G3258 is ALSO a C0 stepping according to CPU-Z).
 
PGHammer said:
What use does AVX 2 have on desktops - or workstations - right now?
Click to expand...

Really are we going there?

Avx2 is used for
-video compression
-encryption
-batch processing of game engine AI
-video decompression
-3D rendering
-batch calculations in Microsoft excel 64

Basically anywhere you need bulk number crunching or searching.

Many many modern applications use it if it is present..
 
Been doing it further back than the 486.
8088 and 8086. Could get math integrated, or buy a coprocessor. This also included instruction and bus width's
 
As I stated, how much of that is used (not usable, but actually used) by the masses? The first three require it to be built in to the application or game - what applications and/or games have it available? Same applies even to the LAST three - while Excel has supported it since 2010, how often do even Excel users use it? Also, how much is the penalty for NOT using it (in terms of performance)? If the penalty is (in real-world terms) ten percent or less, even among those that are [H], that is generally quibble turf; worse, the cost of the upgrade (even same-socket upgrades) is greater than ten percent (and that is just from i3 to i5 - naturally, PentiumG to i5 has an even wider spread). Most users can't upgrade "just because" - the upgrade has to have a point that makes dollars-and-cents sense.
 
alxlwson said:
Been doing it further back than the 486.
8088 and 8086. Could get math integrated, or buy a coprocessor. This also included instruction and bus width's
Click to expand...

Been doing what?
The 8087 was not a neutered 8088/8086. (And you couldn't get an Intel 8086/8088 that integrated the 8087 math).
 
defaultluser said:
Don't cherry-pick the best piece of shit out of the trash bin you posted, and pretend you have a point.

You claimed in your post that:8088 and 8086. Could get math integrated, or buy a coprocessor.

No, the 8086 does not have an integrated FPU, you have to buy an 8087 no matter which CPU you start with. The only difference was the data bus. l I already went into detail why this part existed.

You''re also claiming that the instruction width is different is bullshit.

Why don't you quit while you're ahead?
Click to expand...

Lol, u mad bro?

Also, I stated the external data bus, not the instruction width. L2r
 
Yes, my original post was incorrect. That's why I dropped it.
Then, I brought up the external data bus, and was very clear about that, and respectfully so. At which point you thought I was talking about instruction width. How instruction width = external data bus width in your mind, is curious.

Either way, I've conceded to my fault, and you're still attacking and behaving like a disrespectful teenager that thinks he's the only person in the world that knows anything. GJ!
 
I'm not sure exactly how it started, tbh. Obviously the 8087 was a different beast.
I'm sure that a few 8088's started life as an 8086.
Just going by the wiki article I read, it sounds feasible, but it doesnre say directly either way
 
You must log in or register to reply here.
Back
Top