Any experience with the Haswell T low power CPUs?

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
autobulb
Posts: 14
Joined: Wed Nov 05, 2008 10:54 pm
Location: New York City

Any experience with the Haswell T low power CPUs?

Post by autobulb » Mon Nov 25, 2013 8:39 pm

In particular the 4670T which is a Core i5 quad core at 2.3-3.3GHZ at only 45W. This is quite impressive considering my mobile Ivy Bridge quad core has the same TDP.

The T series doesn't seem to be available on Newegg, but here in Japan they are listed on various online shops. The 4670T seems like a sweet spot since it's quad core for around $230. The next bump goes to a Core i7 and it's another $100 for a 200-400mhz clock increase and the extra 2MB cache. Or, there is an even lower powered i7 version at 2GHZ at 35W. You can find the full list here: http://en.wikipedia.org/wiki/Haswell_(m ... processors

All the parts I am talking about have the T suffix.

Anyway, I'm wondering how much more power efficient these chips actually are compared to their S or regular counterparts. I can't seem to find any reviews on these actual parts. All the reviews that contain the search terms are just charts listing all the Haswell chips at the start of the articles, but they generally only benchmark the K versions.

Abula
Posts: 3662
Joined: Tue Nov 02, 2004 12:22 pm
Location: Guatemala

Re: Any experience with the Haswell T low power CPUs?

Post by Abula » Mon Nov 25, 2013 9:31 pm

autobulb wrote:In particular the 4670T which is a Core i5 quad core at 2.3-3.3GHZ at only 45W. This is quite impressive considering my mobile Ivy Bridge quad core has the same TDP.
Im not surpriese of the similar TDPs, they have almost the same clocks, Intel Haswell Core i7-4700MQ, 2.6~3.4GHz, you even get higher clocks on the mobile + hyperthreading.

Personally i dont think any T is worth it unless you dont have the cooling to handle a standard 4670, its just a nerfed CPU from factory, there is no specially low voltages nor cherry pick cpus, its just like buying a lower end cpu for the close to the same money you spend on the normal, but your cpu is less capable. I made the mistake of buying a 2100T, but i got it on release, i though it was really a special CPU but to my missforutune is just a nerfed 2100 that comes with an intel low profile cooler, never will be again a T or an S, unless they have something different.

All haswell CPUs will downclock to the same 800mhz on idle, most will draw similar power and heat up similar, its on load where each will different, but in the case of the T probably wont heat as much but wont do stuff as fast, so at the end is your choice, your money, but personally i dont recommend T or S versions, unless the build doesnt allow a cooler to keep up with the CPU.

autobulb
Posts: 14
Joined: Wed Nov 05, 2008 10:54 pm
Location: New York City

Re: Any experience with the Haswell T low power CPUs?

Post by autobulb » Tue Nov 26, 2013 6:17 am

Abula wrote:
autobulb wrote:In particular the 4670T which is a Core i5 quad core at 2.3-3.3GHZ at only 45W. This is quite impressive considering my mobile Ivy Bridge quad core has the same TDP.
Im not surpriese of the similar TDPs, they have almost the same clocks, Intel Haswell Core i7-4700MQ, 2.6~3.4GHz, you even get higher clocks on the mobile + hyperthreading.

Personally i dont think any T is worth it unless you dont have the cooling to handle a standard 4670, its just a nerfed CPU from factory, there is no specially low voltages nor cherry pick cpus, its just like buying a lower end cpu for the close to the same money you spend on the normal, but your cpu is less capable. I made the mistake of buying a 2100T, but i got it on release, i though it was really a special CPU but to my missforutune is just a nerfed 2100 that comes with an intel low profile cooler, never will be again a T or an S, unless they have something different.

All haswell CPUs will downclock to the same 800mhz on idle, most will draw similar power and heat up similar, its on load where each will different, but in the case of the T probably wont heat as much but wont do stuff as fast, so at the end is your choice, your money, but personally i dont recommend T or S versions, unless the build doesnt allow a cooler to keep up with the CPU.
Hmm, my experience with low powered CPUs has been always the opposite. Processors are generally "binned" and the better ones can either run at a higher clock speed, or can run at a lower voltage and consume less power, with a sacrifice in clockspeed.

Just looking at the chart I posted contradicts your claims. The easiest to way to see this is by comparing the 4400 and 4430 with their T counterparts. I chose these two because the specs between the regular and T version are exactly the same: same Turbo Boost, same GPU clockspeeds, same cache, except for one thing, a difference of 300mhz in the base clock speed. Yet, the difference in TDP between the regular and T versions is a whopping 19W. Have you ever overclocked a system? A simple increase in 300mhz will not increase your power consumption by 19W, that's for sure. The T chip uses less energy because it's a better binned chip.

Anyway, I am somewhat on your side that the S or T chips might not be worth it. But I don't know if so, and that's why I want to see comparisons, hence this post.

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: Any experience with the Haswell T low power CPUs?

Post by CA_Steve » Tue Nov 26, 2013 8:57 am

I doubt that Intel cherry picks for power efficiency for the T chips. They mass groups of part numbers within certain TDPs. First bin sort is most likely for speed and functionality as they determine the top $ parts. Actual power use may or may not come close to the TDPs (especially for a not-top speed rated part in a category). Take a look at the i5-4670 options along with the i5-4430 and i7-4770K. The 4430 creeps along at 3GHz base and has 6MB of cache, the 4670K is 3.4GHz and 6MB, the i7 is 3.5GHz and 8MB of cache and they all share the same 84W TDP.

Also, recall that power use (while linear to clock frequency) is a square function of core voltage, and that the operating core voltage steps up as you step up the clock freq. Vice versa for lower clocked chips. In the case of the 4670 vs 4670T, here's some math:

4670, 84W, 3.4GHz @ say 1.1V.
4670T, 45W, 2.3GHz @ say .95V. (A complete swag on my part)

Power use for the T part = (2.3/3.4)*[(.95^2)/(1.1^2)]*84 = ~53% of 84W = 44.6W.

autobulb
Posts: 14
Joined: Wed Nov 05, 2008 10:54 pm
Location: New York City

Re: Any experience with the Haswell T low power CPUs?

Post by autobulb » Tue Nov 26, 2013 9:49 am

CA_Steve wrote:I doubt that Intel cherry picks for power efficiency for the T chips. They mass groups of part numbers within certain TDPs. First bin sort is most likely for speed and functionality as they determine the top $ parts. Actual power use may or may not come close to the TDPs (especially for a not-top speed rated part in a category). Take a look at the i5-4670 options along with the i5-4430 and i7-4770K. The 4430 creeps along at 3GHz base and has 6MB of cache, the 4670K is 3.4GHz and 6MB, the i7 is 3.5GHz and 8MB of cache and they all share the same 84W TDP.

Also, recall that power use (while linear to clock frequency) is a square function of core voltage, and that the operating core voltage steps up as you step up the clock freq. Vice versa for lower clocked chips. In the case of the 4670 vs 4670T, here's some math:

4670, 84W, 3.4GHz @ say 1.1V.
4670T, 45W, 2.3GHz @ say .95V. (A complete swag on my part)

Power use for the T part = (2.3/3.4)*[(.95^2)/(1.1^2)]*84 = ~53% of 84W = 44.6W.
One thing you seem to be neglecting in your comparison is price. The thing about the 4670, 4670S, and 4670T is that they are all the same price. The only difference is their clock speeds and TDP. If you are telling me that a 4670 downclocked from 3.4 to 3.1 (a measly 300mhz, and not counting TB which is the same speed for both chips) will save 19W of energy and heat, I will be quite shocked. The reason is because I have overclocked or underclocked CPUs before, and without playing with voltage, a simple change of 300mhz will not save more than a few watts or so in usage, if that.

I can't really try to compare the other chips you listed, because two of them are K chips which are at a premium because of their unlocked cores, and one of them is an i7 which further complicates things with the extra cache.

However, if you look at the 4430 you listed (3-3.2GHZ @ 84W) vs. the 4670S (3.1-3.8GHZ @ 65W) the comparison is much clearer. The S has a slightly faster base clockspeed, a bit faster turbo clock speed and yet manages to do it at 19 less watts for a marginal price increase. This is not taking into account that the S also has a tiny bit faster GPU clock speed as well. In general, it's an all around faster chip in a smaller power envelope (and thus smaller heat envelope,) I don't see how that is Intel trying to take advantage, and I don't see how that isn't a result of CPU binning. The S chip clearly appears to be a better chip since it performs better in every major regard while doing it with less power. Am I missing something there?

CA_Steve
Moderator
Posts: 7650
Joined: Thu Oct 06, 2005 4:36 am
Location: St. Louis, MO

Re: Any experience with the Haswell T low power CPUs?

Post by CA_Steve » Tue Nov 26, 2013 10:51 am

autobulb wrote:If you are telling me that a 4670 downclocked from 3.4 to 3.1 (a measly 300mhz, and not counting TB which is the same speed for both chips) will save 19W of energy and heat, I will be quite shocked.
What I'm saying is:
- TDP is just a catch-all for a class of CPUs. What your individual i5-whatever consumes will be specific to your chip, it's operating frequency, and it's core voltage.
- Yes, undervolting your CPU will save a lot of power. Using the i5-4670 as an example, if stock core voltage is 1.08V (and this can vary depending on the chip and your particular mobo's default settings), dropping it by 0.1V means you've lowered the power use to (0.98/1.08)^2 or 83% of base. If your particular i5 consumes 84W at 1.08V, then at 0.98V it'll consume 69W. That's 15W just by lowering the core voltage a ten of a volt.

What's more important is to figure out if your apps need the speed or not. If they don't, and you don't want to futz with underclocking/undervolting or don't want to buy a mobo that allows it, then sure, get the S or T parts. Remember the Turbo speed listed only applies to single threaded/single core. The turbo bins for 2/3/4 cores are lower.

ggumdol
Posts: 154
Joined: Mon Aug 05, 2013 1:01 pm
Location: Scandinavia

Re: Any experience with the Haswell T low power CPUs?

Post by ggumdol » Tue Nov 26, 2013 11:58 am

To the best of my knowledge, it is a widely known fact that CPU power consumption in the full load state does indeed linearly increases with the CPU clock frequency. In plain words, if you downlock a CPU, its power consumption in full load state accordingly decreases.

My i5-4430 consumes no more than 34W in the load state (though I have disabled Intel's interal graphics module), which is way lower than its official TDP (84W). As I mentioned in one of my earlier posts, Intel was intent on giving the impression their low-end CPUs (e.g., i5-4430) consume as much energy as their high-end CPUs, perhaps to make low-end models less attractive to non-savvy people. There are numerous beachmarks confirming low-power consumption of i5-4430 or the likes. For example:

http://uk.hardware.info/reviews/4993/27 ... ch-115-max

I will have to concur with CA_Steve. It is pointless to buy T models unless your motherboard lacks the capability to downclock CPU.
autobulb wrote:If you are telling me that a 4670 downclocked from 3.4 to 3.1 (a measly 300mhz, and not counting TB which is the same speed for both chips) will save 19W of energy and heat, I will be quite shocked.
Last edited by ggumdol on Tue Nov 26, 2013 1:27 pm, edited 1 time in total.

ggumdol
Posts: 154
Joined: Mon Aug 05, 2013 1:01 pm
Location: Scandinavia

Re: Any experience with the Haswell T low power CPUs?

Post by ggumdol » Tue Nov 26, 2013 12:24 pm

This is exactly where Intel's impure profit-making motive behind their less-accurate-than-desired TDP table starts to deceive the majority of non-tech-savvy people. Microprocessor industry has long evoved over the last couple of decades and there is nothing new under the sun except for the ever-clear and ever more fundamental tradeoffs between "power consumption", "clock frequency", and the "CMOS technology" (e.g., Haswell's 22nm process). If you can achieve higher performance with lower power consumption within the same Intel CPU family, Intel will buy your patent at an astronomical price.
autobulb wrote:However, if you look at the 4430 you listed (3-3.2GHZ @ 84W) vs. the 4670S (3.1-3.8GHZ @ 65W) the comparison is much clearer. The S has a slightly faster base clockspeed, a bit faster turbo clock speed and yet manages to do it at 19 less watts for a marginal price increase. This is not taking into account that the S also has a tiny bit faster GPU clock speed as well. In general, it's an all around faster chip in a smaller power envelope (and thus smaller heat envelope,) I don't see how that is Intel trying to take advantage, and I don't see how that isn't a result of CPU binning. The S chip clearly appears to be a better chip since it performs better in every major regard while doing it with less power. Am I missing something there?
BTW, I would like to take a small deviation from the story. In fact, I would like to argue that it is necessary for people to rethink in a radical way about the real benefits of high-end Intel CPUs. High-end models outperform the rest simply because they consume more power in full load state due to the high clock frequency (I'm oversimplifying the exposition because hyperthreading, cache size, and number of core significantly influence the whole picture as well). If you seldom harness the enhanced performance power of your hign-end Intel CPU which must be true for at least 90% of people, it implies that you have paid extra premium for no reason, for nothing. On top of that, you have to buy more expensive CPU coolers to better dissipate resulting extra heat.

Abula
Posts: 3662
Joined: Tue Nov 02, 2004 12:22 pm
Location: Guatemala

Re: Any experience with the Haswell T low power CPUs?

Post by Abula » Wed Nov 27, 2013 10:17 pm

autobulb wrote:
Abula wrote:
autobulb wrote:In particular the 4670T which is a Core i5 quad core at 2.3-3.3GHZ at only 45W. This is quite impressive considering my mobile Ivy Bridge quad core has the same TDP.
Im not surpriese of the similar TDPs, they have almost the same clocks, Intel Haswell Core i7-4700MQ, 2.6~3.4GHz, you even get higher clocks on the mobile + hyperthreading.

Personally i dont think any T is worth it unless you dont have the cooling to handle a standard 4670, its just a nerfed CPU from factory, there is no specially low voltages nor cherry pick cpus, its just like buying a lower end cpu for the close to the same money you spend on the normal, but your cpu is less capable. I made the mistake of buying a 2100T, but i got it on release, i though it was really a special CPU but to my missforutune is just a nerfed 2100 that comes with an intel low profile cooler, never will be again a T or an S, unless they have something different.

All haswell CPUs will downclock to the same 800mhz on idle, most will draw similar power and heat up similar, its on load where each will different, but in the case of the T probably wont heat as much but wont do stuff as fast, so at the end is your choice, your money, but personally i dont recommend T or S versions, unless the build doesnt allow a cooler to keep up with the CPU.
Hmm, my experience with low powered CPUs has been always the opposite. Processors are generally "binned" and the better ones can either run at a higher clock speed, or can run at a lower voltage and consume less power, with a sacrifice in clockspeed.

Just looking at the chart I posted contradicts your claims. The easiest to way to see this is by comparing the 4400 and 4430 with their T counterparts. I chose these two because the specs between the regular and T version are exactly the same: same Turbo Boost, same GPU clockspeeds, same cache, except for one thing, a difference of 300mhz in the base clock speed. Yet, the difference in TDP between the regular and T versions is a whopping 19W. Have you ever overclocked a system? A simple increase in 300mhz will not increase your power consumption by 19W, that's for sure. The T chip uses less energy because it's a better binned chip.

Anyway, I am somewhat on your side that the S or T chips might not be worth it. But I don't know if so, and that's why I want to see comparisons, hence this post.
Read AVS renethx post #1473
So basically G620T is not a cherry-picked die with lower leakage current, but just a lower-clocked version with lower running voltage. Even I can obtain a G620T from G620 by simply limiting the max multiplier to 22!
He also quotes an SPCR review, read that also.

autobulb
Posts: 14
Joined: Wed Nov 05, 2008 10:54 pm
Location: New York City

Re: Any experience with the Haswell T low power CPUs?

Post by autobulb » Fri Nov 29, 2013 1:53 am

So the general consensus is that the S and T models are for motherboards that don't have downclocking or undervolting capabilities?

In my region it seems like the Core i5 4670K has the best bang for the buck then, since the price range between the lower ones and that one is only $50 less and with the K I can try overclocking via multiplier. The next step up would be the Core i7 but I don't think the increase of $60 for some extra cache is worth it for my needs.

flemeister
Posts: 346
Joined: Mon Dec 28, 2009 1:42 am
Location: Australia

Re: Any experience with the Haswell T low power CPUs?

Post by flemeister » Fri Nov 29, 2013 3:11 am

autobulb wrote:So the general consensus is that the S and T models are for motherboards that don't have downclocking or undervolting capabilities?
Agreed. I can make my "65W" Celeron G1610 into a "35W" CPU, by lowering the clock speed from 2.4GHz to 1.6GHz, and undervolting as well (though my Gigabyte H77M-D3H doesn't have any CPU voltage options...).

I used this site to try and calculate what the TDP would be: http://extreme.outervision.com/psucalculatorlite.jsp

Probably not accurate, but surely it can't be too far off. :)

FWIW, I also managed to get my single 4GB DDR3 module down from 1.5v to 1.2V. 48 hours Prime95 stable as well, not too shabby. :D

autobulb
Posts: 14
Joined: Wed Nov 05, 2008 10:54 pm
Location: New York City

Re: Any experience with the Haswell T low power CPUs?

Post by autobulb » Fri Nov 29, 2013 3:53 am

flemeister wrote:
autobulb wrote:So the general consensus is that the S and T models are for motherboards that don't have downclocking or undervolting capabilities?
Agreed. I can make my "65W" Celeron G1610 into a "35W" CPU, by lowering the clock speed from 2.4GHz to 1.6GHz, and undervolting as well (though my Gigabyte H77M-D3H doesn't have any CPU voltage options...).

I used this site to try and calculate what the TDP would be: http://extreme.outervision.com/psucalculatorlite.jsp

Probably not accurate, but surely it can't be too far off. :)

FWIW, I also managed to get my single 4GB DDR3 module down from 1.5v to 1.2V. 48 hours Prime95 stable as well, not too shabby. :D
Hmm very interesting. Thanks for your input. I guess I have to hope I get a chip that undervolts well.

I would still like to see a comparison between the regular and S/T versions of the chips just to see if there is a difference during idle. And, if the regular chips downclocked to the S/T clockspeeds would yield the exact same savings in TDP.

flemeister
Posts: 346
Joined: Mon Dec 28, 2009 1:42 am
Location: Australia

Re: Any experience with the Haswell T low power CPUs?

Post by flemeister » Fri Nov 29, 2013 5:55 am

autobulb wrote:I would still like to see a comparison between the regular and S/T versions of the chips just to see if there is a difference during idle. And, if the regular chips downclocked to the S/T clockspeeds would yield the exact same savings in TDP.
Already been done in the past, and no reason to think that anything notable has changed since then. :)

* http://www.silentpcreview.com/article954-page4.html
* http://www.silentpcreview.com/article1202-page3.html\

Note how close the undervolted i5 2400 was to the i5 2400S. If it was underclocked as well, it would surely match or even surpass the i5 2400S in power efficiency!

Post Reply