AMD competition for Core i3 (Gamers thread)

[IntelUser2000]

If it has 12MB of l3, its physically a hex-core. Remember, there are "dual" core bloomfield-DP chips as well.

The E5502 has only 4MB L3 though.

 

[DrMrLordX]

The thing that looks interesting to me about i3 is that the memory controller, PCI-e controller, and GPU will be located under the same heatspreader but on a different die in an MCM arrangement. This could mean that the PCI-e controller limitation that holds back i5 overclocks may not exist on i3 chips (might be possible to avoid OCing the PCI-e controller when OCing the cores, or it might be possible to raise PCI-e controller voltage by overvolting the imc instead of the cores). Time will tell whether or not i3 will have nice, low vcore OCs like s1366 i7s have.

 

[cbn]

The thing that looks interesting to me about i3 is that the memory controller, PCI-e controller, and GPU will be located under the same heatspreader but on a different die in an MCM arrangement. This could mean that the PCI-e controller limitation that holds back i5 overclocks may not exist on i3 chips (might be possible to avoid OCing the PCI-e controller when OCing the cores, or it might be possible to raise PCI-e controller voltage by overvolting the imc instead of the cores). Time will tell whether or not i3 will have nice, low vcore OCs like s1366 i7s have.

Well there is a report of a 4 Ghz overclock @ .832 volts using a ES sample.

Its at Coolaler.com.

Other OC reports (by different people) at 4.7 Ghz use much higher vcore though (1.4 volts). Maybe someone figures out how to do this differently? What you are saying about the imc sounds interesting.

 

[DrMrLordX]

4 ghz at what is basically .85v is bad-ass (super htpc anyone?). 4.7 ghz at 1.4v on a 32nm processor . . . not quite so bad-ass. In fact, that much vcore would bother me on a chip from that node.

But that first OC (the 4 ghz one) sure beats the stock vcore OCs of i5, and i5 has a higher stock vcore . . .

 

[cbn]

4 ghz at what is basically .85v is bad-ass (super htpc anyone?). 4.7 ghz at 1.4v on a 32nm processor . . . not quite so bad-ass. In fact, that much vcore would bother me on a chip from that node.

But that first OC (the 4 ghz one) sure beats the stock vcore OCs of i5, and i5 has a higher stock vcore . . .

Yep, I wonder if OC technique will affect results?

Someone like you knows much more about this than I do. In fact, I basically know zero about overclocking. (Raise vcore, raise FSB...that is about all I do. Then I set clocks back down to stock for 24/7 use)

 

[DrMrLordX]

Yep, I wonder if OC technique will affect results?

Probably . . . however, some "extreme" overclockers have complained that it's not hard to hit clockspeed walls with Nehalem and its offspring that seem somewhat-reminiscent of clockspeed walls that various AMD K8 processors had in the past. In other words, i3, like i5 and i7, may scale up in clockspeed extremely well with very little voltage until some arbitrary point beyond which ridiculous amounts of voltage may be necessary to continue increasing clockspeed, even though thermals might not indicate that such a wall should be present.

It may be that i3 will have a "wall" in the 4.0-4.5 ghz range, and the guy who had to push 1.4v to get his chip to 4.7 ghz may have done well to get that far. By all indications, a chip that can run at 4 ghz with .85 vcore SHOULD be able to do more than 4.7 ghz, you'd think, but that's not always how it works out . . .

Someone like you knows much more about this than I do. In fact, I basically know zero about overclocking. (Raise vcore, raise FSB...that is about all I do. Then I set clocks back down to stock for 24/7 use)

I must confess that most of my experience has been with AMD machines. I never really got the hang of the memory strap phenomenon from s775, and until I get my hands on a s1156 or s1366 machine, I won't consider myself to be an authority on the matter.

As an aside, I've been thinking about the question of whether overclocked i3s will match or beat the performance of overclocked AMD offerings at the same price point (especially since I may get a new machine in the next year or so and would be looking at something like a Propus-core X4 or maybe an i3 as a possible candidate).

Based on pricing, it looks like the i3 would be positioned against the Athlon II X4 if it is to be positioned against any quad at all, though AMD may be able to position a Deneb against it if the cheapest i3 really does launch at $119 (I was thinking Intel was going to go sub-$100 with i3?).

If we're going to look at i3 as being competitive due to hyperthreading, then we should compare it to the X4 when contemplating workloads that would fully load 4 or more cores . . . in other words, something that would help the i3 get the most out of HT AND help the X4 get the most out of its four physical cores. Obviously any workload (game or otherwise) that taxed 3 or fewer cores would probably run best on the i3 just due to the superior ipc and clockspeed of the i3.

I've decided to use some "fuzzy math" to do a simplistic comparison of what we should expect from i3 versus the Propus we have today, under the assumption that the X4s currently occupying the sub-$100 quad market will be the same ones on the market when i3 launches (this assumption may prove to be untrue).

Since i3 is based on the same core architecture as i5 and i7, then we can probably assume that it would have about a 20% ipc advantage per core over Deneb. Propus is about 10% slower than Deneb due to a lack of l3. So . . . unless I'm way off-base here, per core, i3 should be about 33% faster than Propus.

Hyperthreading should increase i3's performance by about, oh I don't know, 20% when fully utilized. These are just fudge numbers and can be swapped out if my assumptions are off-base (feel free to correct me anyone).

So, for the purpose of this "fuzzy math" thought exercise, we can (sort of) calculate how fast an i3 would need to go to beat the prevailing OCed Propus chips out there right now, which currently top out at around 3.4 ghz in most machines that run them. All we have to do is this:

(i3 clockspeed * 1.33 * 1.20) / 2 = Propus equivalent clockspeed

So that 4 ghz i3 is about equal to a 3.2 ghz Propus when dealing with a workload that can fully tax four cores. As good as that might seem from a dual-core chip, that isn't going to set the world on fire or be the best available gaming CPU in its segment, since any game that CAN utilize four cores is probably going to be faster on the Propus chip @ 3.4 ghz than the i3 at 4 ghz.

If we give i3 the benefit of the doubt (i.e. that it can hit 4.5 ghz reliably without stupid voltage levels), then we're looking at an i3 that will be competitive with a 3.6 ghz Propus. Since most Propus chips out there with reasonable levels of cooling do not hit 3.6 ghz, that would make i3 a clear winner since it would probably trump Propus even in heavily-threaded apps.

If AMD positions Deneb against i3 instead of Propus, the picture would be less rosy for i3, but I'm not sure if AMD can really afford to do that. That being said, unless s1156 board prices come down a bit, it may be possible to get a C3 Deneb + board for about the same price as an i3 + board in the future, so i3 may well be competing with Deneb instead of Propus. Deneb currently enjoys a higher ipc than Propus and hits clockspeeds as high as 4ghz, making clear victory for i3 much less likely. I wager a 4 ghz i3 = 2.9 ghz Deneb, so if you want to challenge Deneb at every possible workload with an i3, you'd better get it running at 5.5 ghz or so. Good luck with that.

If it's my money on the table (and at some point, it will be), a cheap C3 Deneb would probably win my cash over the i3 unless I really needed my CPU to be lower-power for some reason. I might lose some performance in sparsely-threaded apps but I'll win in apps/games that load 4+ cores.

I will say one thing, though . . . i3 will kill the market for E8x00 chips. Thank goodness for that.

 

[GaiaHunter]

So that 4 ghz i3 is about equal to a 3.2 ghz Propus when dealing with a workload that can fully tax four cores. As good as that might seem from a dual-core chip, that isn't going to set the world on fire or be the best available gaming CPU in its segment, since any game that CAN utilize four cores is probably going to be faster on the Propus chip @ 3.4 ghz than the i3 at 4 ghz.

If we give i3 the benefit of the doubt (i.e. that it can hit 4.5 ghz reliably without stupid voltage levels), then we're looking at an i3 that will be competitive with a 3.6 ghz Propus. Since most Propus chips out there with reasonable levels of cooling do not hit 3.6 ghz, that would make i3 a clear winner since it would probably trump Propus even in heavily-threaded apps.

If AMD positions Deneb against i3 instead of Propus, the picture would be less rosy for i3, but I'm not sure if AMD can really afford to do that. That being said, unless s1156 board prices come down a bit, it may be possible to get a C3 Deneb + board for about the same price as an i3 + board in the future, so i3 may well be competing with Deneb instead of Propus. Deneb currently enjoys a higher ipc than Propus and hits clockspeeds as high as 4ghz, making clear victory for i3 much less likely. I wager a 4 ghz i3 = 2.9 ghz Deneb, so if you want to challenge Deneb at every possible workload with an i3, you'd better get it running at 5.5 ghz or so. Good luck with that.

If it's my money on the table (and at some point, it will be), a cheap C3 Deneb would probably win my cash over the i3 unless I really needed my CPU to be lower-power for some reason. I might lose some performance in sparsely-threaded apps but I'll win in apps/games that load 4+ cores.

OMG - you crushed Computer Bottleneck dreams of i3 setting up the world on fire. :biggrin:

 

[jvroig]

OMG - you crushed Computer Bottleneck dreams of i3 setting up the world on fire.

Computer Bottleneck is a fun poster. On one hand, we're still currently trapped in a world with majority of apps not multi-threaded, so I do understand where he's coming from with his dual-core preference. On the other hand, we're also in a very multitasking world - browsers open, some even with flash animations, all the while an MP3 player is blasting out tunes, and perhaps an office suite is also open, and perhaps a few times we zip and unzip some files, and we haven't even mentioned some background tasks that are running transparently. Regular users still won't come near enough to actually stress two powerful cores, but at least I can see some benefit to going multi-core especially if, along with everything else that's been mentioned, the PC is used for productive work like multimedia, engineering or software development.

I have a strong preference for Propus, and I agree with DrMrLordX's assessment of the i3 vs Propus / Deneb future situation. Of course, for Intel, it hardly matters. i3 isn't targeted to "passionate" (for lack of a better word) users like us who do care about the hardware. i3, if it sells like hotcakes (and I most certainly think it will), will do so because of being offered in lots of OEM packages. Smaller, cheaper, "Intel performance"... and if it does come with an on-die GPU, all the better.

Intel's newer i3 doesn't have to beat the much older Propus on technical merits. It just has to become the apple to the OEM's eyes.

 

[ilkhan]

The E5502 has only 4MB L3 though.

Its still a bloomfield die.
Im still hoping for 5Ghz on great air, as thatd crush most people's i5-750s at the same cost in *most* [not all, but the ones important to me at least] tasks.

 

[cbn]

If we're going to look at i3 as being competitive due to hyperthreading, then we should compare it to the X4 when contemplating workloads that would fully load 4 or more cores .

The problem is Games that fully load all four cores with 100% scaling don't exist. Therefore we would have to adjust your fuzzy math calculations to account for this.

 

[cbn]

I wager a 4 ghz i3 = 2.9 ghz Deneb.

But in gaming benchmarks this won't be true.

 

[Markfw]

The problem is Games that fully load all four cores don't exist.

Wrong again. Stop with the "quad is useless" crap please.

 

[cbn]

Wrong again. Stop with the "quad is useless" crap please.

See my recently done edit in post #210.

 

[Munky]

Not sure if anyone mentioned already, but there are some games which decrease in performance with HT (GTA4 comes to mind, if I recall correctly), AND at the same time a number of modern games will use more than 2 physical cores if they are present (I've ran some tests myself). So there are several potential pitfalls for those hoping the i3 is a good gaming cpu.

I still don't see why a gamer or enthusiast would care much about the i3 when we have plenty of quads available for well under $200, and triple-cores for about $100.

 

[cbn]

Not sure if anyone mentioned already, but there are some games which decrease in performance with HT (GTA4 comes to mind, if I recall correctly), AND at the same time a number of modern games will use more than 2 physical cores if they are present (I've ran some tests myself). So there are several potential pitfalls for those hoping the i3 is a good gaming cpu.

I still don't see why a gamer or enthusiast would care much about the i3 when we have plenty of quads available for well under $200, and triple-cores for about $100.

We wouldn't want to turn on HT with Core i7 in current games. The four physical cores are more than sufficient. In fact, I don't even know of a single title that use more than four threads (someone correct me if I am wrong).

P.S. Core i3 works just fine in GTA IV. Notice it beating the triple and quad cores.

http://www.pconline.com.cn/images/h...926714_gta4.jpg&namecode=diy&subnamecode=home

 

[toyota]

Not sure if anyone mentioned already, but there are some games which decrease in performance with HT (GTA4 comes to mind, if I recall correctly), AND at the same time a number of modern games will use more than 2 physical cores if they are present (I've ran some tests myself). So there are several potential pitfalls for those hoping the i3 is a good gaming cpu.

I still don't see why a gamer or enthusiast would care much about the i3 when we have plenty of quads available for well under $200, and triple-cores for about $100.

it has been mentioned and the few games that were slower were only by less than 5%. also in Windows 7 there is supposedly zero negative effect from HT because it was designed to use it from the get go.

btw GTA 4 performs best with an i7. http://www.pcgameshardware.com/aid,...ead-of-Core-2-Quad-in-CPU-benchmarks/Reviews/

 

[Markfw]

We wouldn't want to turn on HT with Core i7 in current games. The four physical cores are more than sufficient. In fact, I don't even know of a single title that use more than four threads (someone correct me if I am wrong).

P.S. Core i3 works just fine in GTA IV. Notice it beating the triple and quad cores.

http://www.pconline.com.cn/images/h...926714_gta4.jpg&namecode=diy&subnamecode=home

Thats a picture, from a site I have never heard of. Until Anand, or some reputable source reviews the CPU (with all the numbers and environment, and more than one game), I will not even give credit to a "picture", becuase thats all it is.

EDIT: Perfect example above. An I7@2.67 blows a E8600 @ 3.33 away, not even close. And ALL that quads dominate, big time.

 

[cbn]

So, for the purpose of this "fuzzy math" thought exercise, we can (sort of) calculate how fast an i3 would need to go to beat the prevailing OCed Propus chips out there right now, which currently top out at around 3.4 ghz in most machines that run them. All we have to do is this:

(i3 clockspeed * 1.33 * 1.20) / 2 = Propus equivalent clockspeed

I like this formula but we also need to correct for differences in scaling. I propose adding a correction factor to the left side of that equation for games that scale below 100% on the 3rd and 4th cores.

If a game scales 100% on the first two cores but 75% on the 3rd and fourth cores the correction factor would be 100%+75%/2= 87.5%

If a game scales 100% on the first two cores but 50% on the 3rd and fourth cores the correction factor would be 100%+50%/2= 75%

If a game scales 100% on the first two cores but 25% on the 3rd and fourth cores the correction factor would be 100%+25%/2= 62.5%


(i3 clockspeed * 1.33 * 1.20) / 2 (correction factor)= Propus equivalent clockspeed

So for games scaling 50% (Left 4 dead 2 and RE5) with quad core we would apply the .75 co-factor for comparing Athlon II x4 to 4 GHz Core i3. (See example below with co-factor bolded)

(4 GHz * 1.33 * 1.20)/ 2 (.75)= 4.256 Ghz Athlon II x4 (in games scaling 50% better FPS with quad core vs dual core)

 

[Munky]

We wouldn't want to turn on HT with Core i7 in current games. The four physical cores are more than sufficient. In fact, I don't even know of a single title that use more than four threads (someone correct me if I am wrong).

P.S. Core i3 works just fine in GTA IV. Notice it beating the triple and quad cores.

http://www.pconline.com.cn/images/h...926714_gta4.jpg&namecode=diy&subnamecode=home

I'd like to see the whole article, not just one picture of one game at one setting. Even in that picture, the i3 just barely beats a lower-clocked X3 and a L3-crippled Athlon X4. The Q8300 is L2-starved as well. So far I'm not seeing anything impressive from the i3.

 

[cbn]

http://forums.anandtech.com/showthread.php?t=2027820&highlight=clarkdale+arrandale

Here is where we first found the article. Lopri posted the link @ post 15.

But it looks like they changed the benchmarks due to the NDA...however you can still see our discussion of the then published results on the following pages of that thread.

 

[IntelUser2000]

I'd like to see the whole article, not just one picture of one game at one setting. Even in that picture, the i3 just barely beats a lower-clocked X3 and a L3-crippled Athlon X4. The Q8300 is L2-starved as well. So far I'm not seeing anything impressive from the i3.

http://www.parkoz.com/zboard/view.p...=off&select_arrange=headnum&desc=asc&no=17816

If they squeeze 5% more in the final version, I think the 530 might catch up to X4 620.

 

[cbn]

http://www.parkoz.com/zboard/view.p...=off&select_arrange=headnum&desc=asc&no=17816

If they squeeze 5% more in the final version, I think the 530 might catch up to X4 620.

Call of Duty and and Resident Evil 5 (clearly a quad threaded program) comparisons are pretty interesting there.

 

[DrMrLordX]

The problem is Games that fully load all four cores with 100% scaling don't exist. Therefore we would have to adjust your fuzzy math calculations to account for this.

Ah, but therein lies the rub: One of Core i3's major advantages is that it has HT which can boost performance in those instances when you're trying to manage more threads than your physical cores can handle and when your physical cores are leaving resources on the table.

If you want HT to matter, at all, you have to introduce workloads that load four cores, period. You're taking around 20% (or more) of i3's performance away when you move to apps that are not aggressively multi-threaded. You take more of Propus' performance away, granted, but this fact still remains.

You may also seem some "quad capable" games loading phyical cores 2 and 3 to maybe 30-50% on a Propus or Deneb that would load logical cores 0 and 1 to 80-100% on an i3, allowing the Propus/Deneb to handle background tasks flawlessly. i3 might not fare so well.

And, sad though I am to say it, if you're dealing with a game that only really loads two cores after taking graphics drivers and miscellaneous background threads into account, you're probably looking at something that will be bottlenecked by the GPU rather than the CPU unless you're running at low resolutions. As luck would have it, I *do* run low resolutions, a lot, due to my small monitor and "tin eye" (so to speak). Many other more-sane individuals do not.

So, to summarize, if you're going to compare a HT-capable i3 to a quad and list HT as an asset, it is fitting to consider workloads where HT is going to matter. If HT is going to be underutilized or completely unutilized due to sparsely-threaded workloads, then you might get better gaming performance at low res with the i3, but you'll have all kinds of computational resources available with the Deneb/Propus that will put such chips in a different performance class when it comes to multitasking (encoding video + gaming on Deneb/Propus: yes! On i3? Not so much). It doesn't help that i3 is going to be launching at $20 more than the Athlon II 620's current street price. If you look back at that pconline.com.cn link you pasted, the fact that AMD's $99 quad is within 3 fps in GTA IV versus the $119 Core i3 in an antiseptic benchmark environment where background threads are limited, then Propus is lookin pretty good. Post-OC, my guess is that i3 may pull ahead, or you may simpy get GPU-bottlenecked, making the i3 look more or less attractive depending on how CPU-dependent the game is.

Don't get me wrong, I think i3 is a potentially-exciting product, even though I would liked one without an MCMed IGP better (not that I think the IGP will be a huge limitation, since I'm betting Intel can mostly cut power to it if it's disabled, or come close to it). It screams clockspeed potential provided the Nehalem clockspeed walls do not emerge at 4.5-5 ghz (and they may, given early attempts to overclock s1366 Nehalem chips under phase/ln2 often hitting at wall at 5 ghz).

I may still get an i3 since it is currently uncharted territory in the consumer realm, and I may succumb to arrogance and believe that I can get a quad-killer OC on it with enough lapping, Shin-Etsu/IC Diamond 7 paste, a Megahalems, and whatever crackpot ideas I can throw into the mix (it seems readily apparent that the OCD OC treatment may not yield great results on a Propus. Le sigh). I also only do one thing right now that really loads four cores (occasional video encoding). There's no telling what gaming there will be in my future (quad-aware games or not) since when I do game, it's with a buggy hand-me-down GeForce 8800GTX and this 17" gateway LCD that looks like ass and misbehaves if you cycle power to it or let it stop drawing to the screen for more than a few seconds.

Once I regain my sanity, I'll probably just get a Propus or discounted Deneb and call it a day. If I regain my sanity.

 

[cbn]

Ah, but therein lies the rub: One of Core i3's major advantages is that it has HT which can boost performance in those instances when you're trying to manage more threads than your physical cores can handle and when your physical cores are leaving resources on the table.

If you want HT to matter, at all, you have to introduce workloads that load four cores, period. You're taking around 20% (or more) of i3's performance away when you move to apps that are not aggressively multi-threaded. You take more of Propus' performance away, granted, but this fact still remains.

You may also seem some "quad capable" games loading phyical cores 2 and 3 to maybe 30-50% on a Propus or Deneb that would load logical cores 0 and 1 to 80-100% on an i3, allowing the Propus/Deneb to handle background tasks flawlessly. i3 might not fare so well.

Hmmm....I was thinking HT could be maxed out at much lower loads.

 

[GaiaHunter]

Call of Duty and and Resident Evil 5 (clearly a quad threaded program) comparisons are pretty interesting there.

This is still very dependent on game, GPU you utilize and resolution you play.

Have a look at this article http://www.legionhardware.com/document.php?id=869&p=0, where they pit a i7 920 going from 2.0 GHz to 4.0 GHz vs a Phenom II 965 BE also going from 2.0-4.0 GHz while paired with the 5970.

I know you were talking about multi -threaded games, but in here, it shows speed boosts doesn't seem to do much for the i7 architecture.