6-core 32nm Westmere

[Idontcare]

6-core 32nm Westmere

Hong Kong's first trial Intel "Gulftown" Processor

Original (Chinese)

Google Translated to English

They seem to have run into the wall when it comes to scaling number of cores with software, much as AMD's tri-cores did. In many of the non-synthetic benches the 6C gulftown performance matches that of 4C bloomfield.

More of these tests means that, if the user does not need a lot of applications simultaneously, or enforced by the software and is not optimized for multi-core, there is no support of more than four or more threads, then, Gulftown six core are totally useless.

Gaming is coming back as expected, you need the GPU sauce to make games fly.

In the 3D gaming tests, due to the reduced graphics performance bottlenecks in order to achieve maximum difference in processor performance, we have adopted the most current GeForce GTX 295 graphics card and the resolution set at 1024 x 768 and the Detail set to Low, The number of test-core gaming FPS gain.

The result is that most 3D games do not have access to the six core is optimized to enhance the number of processor cores, and there is no appreciable gain for the FPS.

Power consumption is nice, right where you want to see it. Basically you get 2 more cores and another 4 MB L3$ all fit into the same power-consumption envelope.

Some new instructions added to the ISA for handling encryption. Not clear how this impacts the desktop consumer markets yet. I'm assuming this will make adding encryption to software far less penalizing to the customer's experience when using that software, thus opening the opportunity for a whole new market of encryption products to come to the market. Hopefully someone can illuminate further.

 

[ilkhan]

Personally, theres only a couple things I do that could possibly use 6 cores (namely: encoding FRAPS videos), but the cooler process should be great for hitting higher OCs.
Im tempted to rebuild on i7-920 and go straight to sandy, D0 chips are hitting 4Ghz much more consistantly than C0 (C1?) chips did.
Of course, Id need money to do that.

Basically, the chip is wider than almost anything a consumer does, even with HT turned off.

 

[Idontcare]

I agree, the term "overkill" comes to mind.

Which is why the design itself is no doubt targeted at the Xeon segment (hence the delay in release to Q 2010, same timeframe as the Xeon release schedule) and just bled over onto the x58 platform as a enthusiast chip for that 1% of the 1% of desktop users that want one, and no more than one, Gulftown in their gaming rig.

If you don't need one in your gaming rig then you don't need the desktop consumer segment gulftown, you'd just get a workstation mobo (either 1S or 2S) and nehalem 6-core xeon and use it as such.

I know what I'd use the chip for, but really that is workstation work in which I want to be cheap by buying low-cost chips and overclocking them which you use to not be able to do on workstation mobo's. Nowadays though OC'ing a workstation mobo is not so problematic, so the barrier has kind of evaporated.

One gulftown rig would replace three of my Q6600 rigs. My power consumption would drop by about $30/month which is intriguing too. But I'd need a mobo that let's me load it with 24GB of ram to house all the data for running 12 apps in parallel to saturate the cores and I'm not going to put $3k into 4GB dimms just for the pleasure of saving $30/month.

 

[KingstonU]

$3,000 for 6x4GB ram?!? Hmmm that's $500 per stick of ram, or $250 per 2GB to give it a perspective I'd be more familiar with, but still are these the higher-end triple channel DDR3-2200?

 

[aka1nas]

Hehe, I still want one.

 

[exar333]

Originally posted by: KingstonU
$3,000 for 6x4GB ram?!? Hmmm that's $500 per stick of ram, or $250 per 2GB to give it a perspective I'd be more familiar with, but still are these the higher-end triple channel DDR3-2200?

4GB DDR3 modules are VERY expensive. You can't really say "it's 250 per 2GB" because it's really $500 per 4GB. If you only wanted 12GB (6x2gb), it would cost only a fraction of this price.

 

[exar333]

Do we know what speed the cores will launch at? I assume they will be bumped-down a little from what we are used to for quad cores?

Six cores, along with the extra cache, for the same thermals is very encouraging.

 

[scooterlibby]

I really wish they would apply the 32nm process node to the Core i7 quad cores.

 

[Fox5]

Originally posted by: ilkhan
Personally, theres only a couple things I do that could possibly use 6 cores (namely: encoding FRAPS videos), but the cooler process should be great for hitting higher OCs.
Im tempted to rebuild on i7-920 and go straight to sandy, D0 chips are hitting 4Ghz much more consistantly than C0 (C1?) chips did.
Of course, Id need money to do that.

Basically, the chip is wider than almost anything a consumer does, even with HT turned off.

I think you'd see a bigger gap with HT turned off.

Hyper-threading is a big win for a lot of apps, but honestly, twelve threads is probably well past the point of diminishing returns. It's probably why AMD isn't adding hyper-threading, it adds substantial complexity, size, and heat to the core, at a time when we have good multi-core architectures in place and core numbers that are rapidly advancing beyond what is usable.

I think we'll see the industry standardize on 6-8 core processors, and then see a switch to heterogeneous computing.

 

[VirtualLarry]

Originally posted by: Idontcare
6-core 32nm Westmere
They seem to have run into the wall when it comes to scaling number of cores with software, much as AMD's tri-cores did. In many of the non-synthetic benches the 6C gulftown performance matches that of 4C bloomfield.

More of these tests means that, if the user does not need a lot of applications simultaneously, or enforced by the software and is not optimized for multi-core, there is no support of more than four or more threads, then, Gulftown six core are totally useless.

Wasn't there another thread on here, about CPU affinity settings internal to WoW, that allowed for a speedup? But the value only ranged from 1 to 255, so it was only allocating 8 bits to CPU affinity. With a 6-core CPU with HT, that leaves out some of the cores. No wonder the software wasn't using all of the cores.

The result is that most 3D games do not have access to the six core is optimized to enhance the number of processor cores, and there is no appreciable gain for the FPS.

 

[Jovec]

Most games don't do multi-threaded rendering, so after you get 1 core for rendering, 1 core for AI, 1 core for physics, and 1 core for other game functions you'll see diminishing returns. Even this is a stretch, as the majority of games are simply 1 core for rendering and a 2nd for everything else.