I'm waiting for the release of low end Ivy Bridge CPUs. For one application, I want something with a low TDB like a Sandy Bridge G630T and for the other, something like an i2100.

Bill

A couple of the dual core i3 mobile chips are shipping now...but no news on the desktop variants. Rumor has been summer for the i3 and fall for the Pentiums.

Thanks, C A_Steve.

Thanks. You can ignore my other post in the other thread. I didn't see this.

This rumor is too optimistic, unless perhaps you define "fall" as december or something.

Well, until Dec 21, it is fall.

Thanks for the crystal ball reading.

Hmmm, December.

For my dedicated MusicPC, a low end processor like the G630T would be fine. Plenty of CPU power and the application doesn't require 3D graphics. A low TDP would let me make desirable choices for case and CPU heatsink fan.

I think that picking a B75 motherboard with a Sandy Bridge CPU would give me the functional characteristics I want. I though that waiting to see what the corresponding IvyBridge CPU chips offered would be prudent. (A 10-20% increase in CPU speed isn 't significant for this application.) Waiting until December (+ 3 months for BIOS bugs to shake out) seems a long time to wait.

Any thoughts?

Bill

B75: You can save ~$20 over an H77. Has 1 native SATA3 port instead of 2, lose RAID support and SSD caching.

Pentium for music server? Sure. I don't know that I'd bother with a T suffixed part.

I don't see why you'd be concerned with BIOS bugs at this point.

The G530 is cheaper and there's very little difference in TDP with the G630T (for the upteenth time, Intel's ratings are for the whole line, not for individual CPUs). In case you have a 50W power supply or something and want really low TDP, you can get the G530T. But the G530 would be fine, really.
It'll be a long wait before they sell Ivy Bridge CPUs that cheap.

Or you could get an Atom board. Even cheaper and even lower TDP...

I've found that waiting 3 months after a motherboard becomes available to buy one avoids some hassles. User reports show up new problems and BIOS updates are often frequent during that period. In this case, the motherboard would be well past 3 months old but the Ivy Bridge CPUs would be new to the motherboards. I'm content to avoid being a pioneer when I can.

I have done my reading on the subject of TDP for Sandy Bridge and Ivy Bridge. I've observed my Sandy Bridge system and know how it behaves with respect to power consumption and temperature levels. My considerations may be different from yours. The difference in price between the G630 and the G630T is not large ($ 7) so getting the G630T rather than a G630 is not a big deal. The G530 is significantly cheaper ($ 25) so I'll think about that.

I use a ~ 2.5 year old CULV laptop so I have a sense of how Win 7 might feel on an Atom CPU system. I've considered an Atom based system with each new generation. I don't think I'll go that way.

I was hoping to get some ideas about reasons getting a low power Ivy Bridge CPU might be worth the wait for either of my two applications.

> It'll be a long wait before they sell Ivy Bridge CPUs that cheap.

An interesting view. How do you think that Intel's prices for Ivy Bridge Pentiums and Celerons will be different from those for Sandy Bridge equivalents?

Bill

Chances are even the fastest Atoms wouldn't match your CULV for responsiveness. For number crunching, they might beat it (depending on the exact operations) because they can run 4 threads.


I have no idea. Probably the same (assuming Intel does upgrade the Pentium and Celerons).

What I'm saying is that they're going to be released later than the more expensive models (if at all). The dual-core Ivy bridge which are shipping right now are expensive (mobile or i5). At some point the cheaper i3 models will be released but this was already delayed. The cheaper models will be released even later, probably in two waves (first the Pentiums and then the Celerons). This has been Intel's MO.
We might experience further delays this time around because there's no reason Intel should hurry (other than manufacturing and logistic considerations). The Sandy Bridge parts match the needs of the market and there's no real competition. When the Sandy Bridge cheapos were released, 1155 lacked low-end CPUs and Intel's value CPUs were obsolete.

Is there a way to, effectively, simulate/force a lower TDP using some logicboard BIOS tweek? (I read something about this somewhere but I'm not clear on it.) If so, then I think the T SKUs are a waste of money. (Not only is the G630T $7 more than the G630, but it also runs at only 2.3GHz as opposed to 2.7GHz for the G630.)

Here's a comparison of the G620 vs 620T @ Anandtech. 4W difference in load power, 1.5W diff at idle (via Bench), performance hit in line with the clock frequency drop.

The 1.5W figure is at variance with what others have found and plain unbelievable considering how much power these CPUs actually draw at idle. It goes to show how credible Anandtech's numbers are.


You can get about 90% (number out of my most amazing butt) of the TDP benefit by lowering the CPU speed. You can do it in most BIOSes ("most" mainly because of pre-built computer with custom boards) or in software.

My concern with that test relates to the use of the Radeon graphics card for the power test. The integrated graphics portion of the CPU chip was probably not active at all. I also have a concern that the test did not push CPU utilization to the max.

I've seen tests of i32100 / i52400 level processors that showed power use of about 30 watts for the IGP portion of the chip on a graphics benchmark. My memory was that a CPU only benchmark showed about 30 Watts being used by the CPU. I'd want to know how much the IGP functionality was throttled back on the T version.

So for the full picture of power use for the G620 versus G620T, the test in the Anandtech should be run with integrated graphics active for a CPU + graphics benchmark. The benchmark should push both CPU and IGP functionality to their max.

Tests like this one in the anandtech article produce results that may or may not be relevant to what I want to know. It's up to me to figure that out.

Bill

Quibbling over 1.5W when the test setup.measurement has a 1W accuracy at best isn't worth it.

Here's another review at digitalversus.

"At idle, there's no difference between the Pentium G620T and the standard G620. When using the integrated graphics circuit, we took a reading of 46 Watts for both. In full load, there's a gain but it isn't all that significant: 78 Watts on the G620 and 68 Watts on the T version."

X-bit Labs' site is down. They have a review as well.

30W seems too high. I bet that includes conversion losses and possibly CPU/RAM overhead.

Regardless, the Pentiums have a less powerful GPU so you don't have to worry about anything like 30W.
But your concern is well-founded: the GPU has a different clock in the G620T (as compared with the G620). But some boards allow you to underclock the GPU as well.

@ HFat and CA_Steve: Thanks for all the good posts.

@ CA_Steve: Thanks for the link to the digitalversus review. I had high expectations when I clicked on the graph for power consumption graphs. "Image currently unavailable...". Well, nothing's perfect. The 10 watts difference seems more reasonable than 4 watts but the exact nature of this benchmark wasn't described either.

@ HFat: I didn't expect the 30 watts IGP use to transfer to the Pentium CPUs. I made the argument to show that IGP power use is a significant part of the total for the chip.

Bill

XBit seems to be back up now. Here's the idle power graph:

G620T actually idles at 0.1W higher than the non-T part.

On the B75, if you use the single SATA3 for a SSD system/boot drive, then do you really even lose anything by not having SSD caching?

Only if you put commonly used apps on an HDD.

Doesn't the system power, which includes the draw of the motherboard, seem a bit low to you on this graph?

The XBit CPU Burn and GPU burn tests were of interest to me.

G620 - G20T use
CPU burn: 6.7 W
GPU burn: 1.9 W

combined: 8.6 W

Bill

The CPU difference can easily be drastically reduced my making the G620 as slow as the G620T. The G620 mostly consumes more because of the higher clock.
In many cases, it's probably possible to reduce the GPU difference as well (see above).

Another way to look at it: If your applications don't use 100% of the G620T's CPU/iGP, and the core voltages for the two are the same, then the G620's power consumption for those same tasks will be the same. There's nothing special built into the "T" part.

Yeah. With the T and S series, Intel marketing is successfully charging money for making something slower. They effectively take a Pentium DualCore 605 or so that would retail for 25 bucks, call it 620T and add those 25 bucks to the price of the 620. Amazing, really.

I mean, Ferrari wouldn't disable one cylinder bank of a V8, downtune it's power to 180 HP and sell you that miserable 4-cylinder as premium Eco-Version costing 50k more than the 8-cylinder model.

Idle is the same, and load power consumption is proportional to clock speed.

Why would anyone buy the -T, unless you can't lower the multiplier yourself?

In a music PC, how often does it run on full load? And if it does, I bet those extra 400 MHz are useful, and the added 1.9 W won't make any difference.

I'm very satisfied with my G540, it's really fast.

Nice to know. You can't beat the price/performance on those. I get the sense the Sandy Bridge Celeron is not like the Celerons of past years.

The Celerons of old were unfairly maligned to begin with. But CPUs were not as overkill as they have become so the low-end models actually felt like low-end models. You got good value for your money though if a low-end PC is all that you wanted.


You can always make it go slower in software even if you can't lower the multiplier.
Most boards don't support undervolting but I don't think that makes a real difference anyway. Some people might not be able to underclock the GPU but that's no big deal either.
You need to change your perspective to see the real value: it's not about who can lower the multiplier but who can raise it back. Meddling users can't raise the multiplier on a factory-underclocked CPU. No, that's not a concern for most people and that doesn't justify the price of the T models... but if you need the feature, you're going to pay. That's why I bought one (that was before the Celerons became available obviously).

No I'm aware of the value in the higher multiplier. Just because I don't mention it, it doesn't mean I'm unaware of it.

In general, people are obsessed about load power consumption, without realizing that most CPU's rarely runs at 100 % for more than a minute at a time.