AMD competition for Core i3 (Gamers thread)

[cbn]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3648&p=7

According to this the Core i3 has two very fast cores (perfect for maxium performance with 99% of the games out there) while performing on par with Core 2 quad (q9400) with respect to four threaded programs.

What does AMD have planned to counter this? (in the near future)

To me this idea of having maximum performance in dual threaded games is more attractive than have medium performance spread across four cores. In fact, it makes me want to think twice about being an early adopter with respect to CPUs.

Who would have thought a future dual core would be better with four threads than a dedicated quad core like Q6600? So much for "future proofing" strategies.

EDIT: Here is the Anandtech review thread with testing completed--> http://forums.anandtech.com/showthread.php?t=2038294

Another review from Anandtech focusing on Core i3 530--->http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3724

 

[Hyperlite]

To me this idea of having maximum performance in dual threaded games is more attractive than have medium performance spread across four cores. In fact, it makes me want to think twice about being an early adopter with respect to CPUs (this chip will no doubt be faster four way multi-threaded than a dedicated quad core like Q6600)

Lynnfield makes your argument completely moot....

Turbo mode is the future. Cost aside, there is literally no reason not to buy a multicore processor.

The more cores/slower clock vs less cores/higher clock argument is so Q1 2009. sheesh.

Is the next generation of dual cores (with HT) going to beat out some quads? Sure it is. Is that news??? That's just progress.

 

[cbn]

Lynnfield makes your argument completely moot....

I would rather buy a cheaper CPU that is actually faster in two cores.

Lynnfield has more TDP spread out over four cores so I wonder if Core i3 will actually beat it in dual threaded games. Besides 32nm > 45nm.

 

[cbn]

The more cores/slower clock vs less cores/higher clock argument is so Q1 2009. sheesh.

I wasn't in this forum for those arguments. But in general (to me) it makes far more practical sense to have two faster cores than four slower cores. A situation with a server would be different I am sure.

To me buying Core I7 860 is like buying Q6600 back in 2007...a complete waste of money unless you are constantly using specialized programs.

 

[Hyperlite]

I would rather buy a cheaper CPU that is actually faster in two cores.

Lynnfield has more TDP spread out over four cores so I wonder if Core i3 will actually beat it in dual threaded games. Besides 32nm > 45nm.

i did say:

cost aside

It's obvious what you would rather have, my point stands.

I wasn't in this forum for those arguments. But in general (to me) it makes far more practical sense to have two faster cores than four slower cores. A situation with a server would be different I am sure.

You're missing the point, bottleneck. Do you know how turbo mode works and what it intends to accomplish? The idea is that, cost being equal, the consumer no longer has to choose between more cores or a faster clock.

of course 32nm > 45nm. That doesn't even apply. Turbo mode, and whatever they call it and however they implement it in future iterations, is here to stay on multicore processors. I totally understand this is moot for someone shopping below that pricepoint. I can see the i3 edging out the i5 in some scenarios due to HT, yes, i'll readily admit that. What's this future proofing nonsense you speak of, anyway?


Sorry, to answer your original question, http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3674

I'm still trying to figure out where all those arrows go, but i'm sure it's brilliant

In short, AMD's answer to hyperthreading is done in hardware, through the use of multiple integer clusters on a single core.

A single Bulldozer core will appear to the OS as two cores, just like a Hyper Threaded Core i7. The difference is that AMD is duplicating more hardware in enabling per-core multithreading. The integer resources are all doubled, including the schedulers and d-caches. It’s only the FP resources that are shared between the threads. The benefit is you get much better multithreaded integer performance, the downside is a larger core.

My take on it is that we are going to see some wicked, wicked competition between AMD and intel in 2010 and 2011.

 

[ilkhan]

You're missing the point, bottleneck. Do you know how turbo mode works and what it intends to accomplish? The idea is that, cost being equal, the consumer no longer has to choose between more cores or a faster clock.

of course 32nm > 45nm. That doesn't even apply. Turbo mode, and whatever they call it and however they implement it in future iterations, is here to stay on multicore processors. I totally understand this is moot for someone shopping below that pricepoint. I can see the i3 edging out the i5 in some scenarios due to HT, yes, i'll readily admit that. What's this future proofing nonsense you speak of, anyway?

While I normally really agree with you, (seriously, I agree with you for 99% of people) for the games who ARE CPU limited at high resolution (in my case, supreme commander, which sees minute gains from 3/4 cores but huge gains with clock speed) an overclocked dual makes the most sense per dollar.

 

[Hyperlite]

While I normally really agree with you, (seriously, I agree with you for 99% of people) for the games who ARE CPU limited at high resolution (in my case, supreme commander, which sees minute gains from 3/4 cores but huge gains with clock speed) an overclocked dual makes the most sense per dollar.

oh yes absolutely, if your game isn't multithreaded (beyond 2) and cost is a major factor, then yesh!

Bottleneck, i may have completely missed the "in games" part of your OP. i apologize.

 

[cbn]

You're missing the point, bottleneck. Do you know how turbo mode works and what it intends to accomplish?

It is supposed to help a quad core work more like a dual core (which is a smart idea) by increasing processing speed on selected cores.

But something tells me this is still less efficient than having a true native dual core.

The best situation would be a multi-cored CPU that is able to share a single thread on two or more of its cores at the same time.

 

[cbn]

While I normally really agree with you, (seriously, I agree with you for 99% of people) for the games who ARE CPU limited at high resolution (in my case, supreme commander, which sees minute gains from 3/4 cores but huge gains with clock speed) an overclocked dual makes the most sense per dollar.

Isn't AI normally what makes games more CPU dependent?

So if the game is a multiplayer online shooter we are talking some pretty low CPU requirements right? These people would probably want to spend money on 120 Hz Samsung monitors than quad, hex or octo core processors.

 

[ilkhan]

Isn't AI normally what makes games more CPU dependent?

So if the game is a multiplayer online shooter we are talking some pretty low CPU requirements right? These people would probably want to spend money on 120 Hz Samsung monitors than quad, hex or octo core processors.

My SupCom games are almost always skirmish with >2 AIs and 500 units per side. Slows to a crawl at around and hour. Building a P55 and a clarkdale to play to replace my slow C2Q is a viable option.

OP- lynnfield's turbo mode makes it identical in 90% of dual core tasks. Fewer cores active, more clock speed per core.

 

[cmdrdredd]

It is supposed to help a quad core work more like a dual core (which is a smart idea) by increasing processing speed on selected cores.

But something tells me this is still less efficient than having a true native dual core.

The best situation would be a multi-cored CPU that is able to share a single thread on two or more of its cores at the same time.

Wrong...

Turbo is overclocking your CPU beyond it's stock clock speed, for free, wth no adverse side affects depending on the workload. It's FREE performance for all applications that aren't stressing multiple cores. It is not supposed to make it a dual core.

There are many games that benefit from multicore CPUs now and even drivers from Nvidia and ATI allow faster performance with multicore systems. The future is and will be multi threaded applications and not raw clock speed. It's time to move on really. The idea of saving a few dollars to get a faster dual core might seem good now, but in 6 months you'll wish you bought that i7 or i5 and overclocked it.

This is how the technology is moving.

 

[cbn]

Turbo is overclocking your CPU beyond it's stock clock speed, for free, wth no adverse side affects depending on the workload.

How is this any different from what I said below....

It is supposed to help a quad core work more like a dual core (which is a smart idea) by increasing processing speed on selected cores.

http://www.intel.com/technology/turboboost/

Essentially the idea is to allow greater performance for smaller threaded programs (which sounds for all practical purposes like the opposite of hyperthreading).

Still none of these strategies would be as good as multiple cores being able to simultaneously work on single threads.

 

[cbn]

but in 6 months you'll wish you bought that i7 or i5 and overclocked it.

In the same way I wish I bought a Q6600 back in 2007.

At the same time though I admit I am not a heavy computer user (encoding doesn't take up a large part of my day, etc)

Seriously if a 32nm dual core gives me better performance than a 45nm quad core and costs less money....I wouldn't be looking back.

What I need is AMD to come out with something competitive with this Core i3 idea. Then I could use it on one of their mini-dtx boards. (mini-dtx--->mini-itx with two PCI-E slots for Crossfire).

 

[Hyperlite]

In the same way I wish I bought a Q6600 back in 2007.

At the same time though I admit I am not a heavy computer user (encoding doesn't take up a large part of my day, etc)

Seriously if a 32nm dual core gives me better performance than a 45nm quad core and costs less money....I wouldn't be looking back.

What I need is AMD to come out with something competitive with this Core i3 idea. Then I could use it on one of their mini-dtx boards. (mini-dtx--->mini-itx with two PCI-E slots for Crossfire).

Ok you have to realize that's a bit too much of a blanket statement. a 32nm dual might outperform a lower end 45nm quad in the right situations, just like a 65nm single probably outperformed a 90nm pD in some situations...(wasn't the pD 90nm?)

i'm just saying, don't blame it on the process

 

[cbn]

Ok you have to realize that's a bit too much of a blanket statement. a 32nm dual might outperform a lower end 45nm quad in the right situations, just like a 65nm single probably outperformed a 90nm pD in some situations...(wasn't the pD 90nm?)

i'm just saying, don't blame it on the process

My guess is that Intel will cripple the Core i3 mainboards in an effort to maintain the image of Core i5 quad. But all things equal I see these 32nm duals scaling better with online FPS.

Now it is up to AMD to make a cheap affordable alternative. An overclockable mini-dtx CPU/mainboard combo for $150 would be ideal. (Leaving more money for SSD, 120 Hz monitors, Eyefinity, Crossfire, etc)

P.S. Something tells me there will be a lot of potential out there for people writing sophisticated AI (RTS/RPG games). No doubt by the time we hit 16 to 32 core CPUs....most of the gaming potential of Gulftown will still be left untapped (I might be exaggerating here a little bit but you get the point). Either that or 11nm single cores will be faster than Gulftown in hex threaded programs. The lesson I am learning is only buy what you need at the moment.

 

[ilkhan]

My guess is that Intel will cripple the Core i3 mainboards in an effort to maintain the image of Core i5 quad. But all things equal I see these 32nm duals scaling better with online FPS.

Now it is up to AMD to make a cheap affordable alternative. An overclockable mini-dtx CPU/mainboard combo for $150 would be ideal. (Leaving more money for SSD, 120 Hz monitors, Eyefinity, Crossfire, etc)

P.S. Something tells me there will be a lot of potential out there for people writing sophisticated AI (RTS/RPG games). No doubt by the time we hit 16 to 32 core CPUs....most of the gaming potential of Gulftown will still be left untapped (I might be exaggerating here a little bit but you get the point). Either that or 11nm single cores will be faster than Gulftown in hex threaded programs. The lesson I am learning is only buy what you need at the moment.

P55 motherboards will accept clarkdale chips just fine. Intel's already confirmed that.
And 32nm will OC better than 45nm, maybe significantly. We dont know that for sure, but if lynnfield does 4Ghz, I expect clarkdale can hit 5Ghz with 4 threads. Not too shabby.

You're looking at $200 for an i5-750 with 4c/4t, or $200 for an i5-660 with 2c/4t. The lynnfield doesn't have a prayer of hitting 5Ghz, though. Turbo on a 750 maxes at 3.2Ghz on a single thread.

 

[Hyperlite]

Turbo on a 750 maxes at 3.2Ghz on a single thread.

unless you OC!

The i5 750 is going to be in a strange place when 32nm drops. It will be the only processor in the ix lineup that lacks hyperthreading, it seems...

BUT it will be interesting to see what clock speed 32nm at 2c/4t will have to hit to best it. i'm guessing MORE than 5ghz, so it may just not happen.

 

[MrK6]

unless you OC!

The i5 750 is going to be in a strange place when 32nm drops. It will be the only processor in the ix lineup that lacks hyperthreading, it seems...

BUT it will be interesting to see what clock speed 32nm at 2c/4t will have to hit to best it. i'm guessing MORE than 5ghz, so it may just not happen.

Physical cores > logical cores. My i5 750 is sitting at 4GHz with very little effort. 32nm will have to bring a LOT to the table in regards to overclocking if it wants to topple the i5 750 in the gaming market. However, the combined GPU/CPU package is going to rule the general market. It's a great all-in-one solution for the majority of users, well done Intel.

 

[Hyperlite]

Physical cores > logical cores. My i5 750 is sitting at 4GHz with very little effort. 32nm will have to bring a LOT to the table in regards to overclocking if it wants to topple the i5 750 in the gaming market. However, the combined GPU/CPU package is going to rule the general market. It's a great all-in-one solution for the majority of users, well done Intel.

yup :thumbsup;

 

[Ben90]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3648&p=7

According to this the Core i3 has two very fast cores (perfect for maxium performance with 99% of the games out there) while performing on par with Core 2 quad (q9400) with respect to four threaded programs.

What does AMD have planned to counter this? (in the near future)

To me this idea of having maximum performance in dual threaded games is more attractive than have medium performance spread across four cores. In fact, it makes me want to think twice about being an early adopter with respect to CPUs.

"This data wasn't run by us and is straight from Intel."

"...however Intel mentioned that a big portion of that are the AES-NI instructions on Clarkdale accelerating one of the tests."

Note how the SPEC*CPU2006 score is an estimate==its not at there yet... I remember 65->45nm yielding 20% improvement across the board from an intel estimate

That whole page is a very bad place to base an argument off of. It only shows two things:
The integrated GPU is faster than a g45...Cool, none of us are going to use it anyways...
The processor has a IMC... Cool, everyone knows FSB sucked, but we don't know how much of an improvement its going to bring

Who would have thought a future dual core would be better with four threads than a dedicated quad core like Q6600? So much for "future proofing" strategies.

Who would have thought a future single socket system would be better with four threads than a dedicated DS like dual Pentium IIIs? So much for "future proofing" strategies

Who would have thought a future single core (sempron) would be better with four threads than a dedicated dual core like P-D820? So much for "future proofing" strategies

Who would have thought a future video card (5870) would be better with thousands of threads than a dedicated dual card system like 6800 ultra SLI? So much for "future proofing" strategies

Oh...and bulldozer

 

[cmdrdredd]

How is this any different from what I said below....



http://www.intel.com/technology/turboboost/

Essentially the idea is to allow greater performance for smaller threaded programs (which sounds for all practical purposes like the opposite of hyperthreading).

Still none of these strategies would be as good as multiple cores being able to simultaneously work on single threads.

Because it is NOT ACTING LIKE A DUAL CORE! You can get turbo enabled with 3 cored active. Imagine that

 

[SunSamurai]

I dont think its smart at all to have a 2< core system act like a 3> core system. Game are only going to be MORE multitasking and BETTER at multi CPU performance, not LESS.

Name one game coming out next year that isnt going to be 3< core aware and Ill name you one game behind the times.

 

[cbn]

unless you OC!

The i5 750 is going to be in a strange place when 32nm drops. It will be the only processor in the ix lineup that lacks hyperthreading, it seems...

I think the cheapest i3 is supposed to lack hyperthreading also.

 

[cbn]

However, the combined GPU/CPU package is going to rule the general market. It's a great all-in-one solution for the majority of users, well done Intel.

If the GPU becomes more and more important and switchable graphics are enabled, I think a lot of people will be rethinking how many cores they need.

Something tells me the upcoming 22nm Octocore will become the Q6600 of 2011.(Only able to be fully utilized by a very small number of people using specialized applications). I hope I am wrong about this though.

 

[ilkhan]

Because it is NOT ACTING LIKE A DUAL CORE! You can get turbo enabled with 3 cored active. Imagine that

For the same 2 core active clock speed, the quads will cost more.
For the same price, you get less clock speed for the same 2c active.

The price list makes that pretty simple. The quads CAN get to the same clock speed (in some cases), but not for the same value. What turbo does is remove the single threaded disadvantage that occurred with C2Q chips vs C2D chips (IE getting 2 more cores but losing clock speed). Ill repost my chart when I get home, but the breakdown is pretty simple.