Haswell gt3e to crush Nvidia and Amd low end market gpus ?

Specifically that right there. Sometimes I wonder what these OEM's are thinking.

 

On one hand, the Intel name sells more expensive computers.

On the other hand, if they relegate AMD out of relevance entirely, Intel has no competition and can charge them whatever they want, or at least stop giving them millions of free processors.

I'm sure, like most business decisions, it comes down to NOW economics: we make more money NOW and we'll worry about the future when it arrives.

 

I wouldn't be surprised if the OEMs were stuck in the "chase the next quarters financial statement" game. Gotta look good now than later.


I recall there was one large Dell kiosk at a major shopping mall in the outskirts of Chicago. Banners with Intel CPUs everywhere, along with a couple advertisements within the store. The kiosk also had around one to two dozens of laptops.

Not a single mention of AMD.

 

All's well that Haswell, right?

(..seriously, though - multiple pages and no one did that one )

 

Thinking about it a bit more I am not opposed to a faster iGPU, and I would welcome it as a compliment to my dGPU, not necessarily an alternative. If it means that I can run applications like Photoshop and older games with Intel Graphics and conserve energy then great. Maybe this iGPU will also be capable of better output resolutions, I wouldn't be opposed to that either. Nvidia's Optimus has been great and it ensures that my performance orientated notebook gets decent battery life and gives me more flexibility. Because of my likes and needs I will always buy a notebook that has a upper range GPU and I would welcome a more powerful iGPU any day.

Now if Intel marketed their iGPU as such then fine, but I have a feeling they (and some others) are going to spin this thing as a replacement to dGPUs for some buyers. I don't see it as a direct threat to GPUs and their future pricing and availability necessarily, although I am sure Intel does.

 

If the new Haswell GPU has its own VRAM, I'd say it already fits the traditional definition of dGPU. They just sell a CPU and a GPU in one package.

If those onboard GPUs get much faster, there's no doubt that people who don't play latest AAAs on their mobile machines will have less interest in stand-alone MXM dGPUs. The downside is for those who need no GPU performance at all, the low-power benefit is lost.

 

Although I wouldn't say MXM GPU's those are typically reserved for the high end machines. But to have no dedicated GPU yes. They should really have two types of chips. One with a low power IGP for those that just want basic browsing, and those with a high power IGP so the price points will be different and the low power IGP's can offer better battery life. It's almost like putting a V-8 engine in every car you sell. Doesn't make much sense.

 

That's what Intel are doing with the external DRAM, I doubt its going to be in the main series of CPUs.

 

Was thinking, the thread title is appropriate. Intel already does and will continue to dominate the low end GPU market. You could say it's nearly a monopoly.

But despite my preference for dGPU, I really do think Haswell will also make entry level dGPU gaming GPUs obselete and definitely seal my argument for why AMD APU is obselete.

 

People are already buying those V-8s though. How many people buying i7 Quads actually need one?

 

By saying that Intel dominates the low-end GPU market, you mean that most consumers who don't have dGPUs in their systems have Intel processors for the same "it says Intel on the sticker" reasons that we've already discussed, right? As far as iGPU goes, they have lesser hardware and inferior drivers to AMD, and it's unlikely that any Haswell iGPU other than the gt3e will manage to match or outdo (by a small percentage) AMD's year-old Trinity, not to mention the incremental upgrade of Richland or the significant upgrade of Kaveri.

The fact is that the vast majority of Intel's market wins in the low-end market, including iGPUs and CPUs for people who only need basic computing power, are based on ignorant people who think Intel is inherently better than AMD and the OEM's that fail on a regular basis to put APUs in competitive machines. They don't sell a better product at the low end. They sell a product better.

 

It does seem like Nvidia have great belief in their upcoming Haswell with GT3 graphics.

According to them we should see desktop GT 640 performance from it.
Meaning it will beat GT 640M since it is clocked lower.
GT 640 GPU-Z
GT 640M GPU-Z

 

Slip of the fingers there with the Nvidia thing, we'll let it slide.

They may claim to have great faith in the product, but that infographic also implies that their integrated graphics have doubled in performance in every Core i-series generation, when Haswell is supposed to be by far their biggest iGPU improvement and I don't believe it's supposed to double Ivy Bridge's iGPU performance. Lie to me about your past achievements, how can I trust your future promises?

 

According to intel ofc.

 

Yes
There is an asterix there right next to the GT 640, but it doesn`t say what it means.

I believe that the IGP performance from Ivy Bridge will double.
Ivy Bridge IGP: 16EUs @ 350MHz-1350MHz. No dedicated memory. IGP use system memory DDR3 as VRAM, memory bandwidth on a 1600MHz DDR3 is 25.6 GB/s.
Haswell IGP: 40EUs @ 200MHz-1300MHz, dedicated memory with memory bandwidth of 64GB/s.

They are increasing the die size over Ivy Bridge to fit more EUs inside.

I understand that you don`t trust their marketing though, but the specs are there and it look pretty promising. If they promised double performance from Sandy to Ivy, then yeah thats bull

 

If Haswell's IGP can double Ivy Bridge's, that's great. The graphic showed that Arrandale was 10x as powerful as their original, motherboard-level IGP from 2006 - no knowledge, no comment. It then showed that Sandy Bridge was 20x as powerful, so double Arrandale, which is an absolute crock, and then that Ivy Bridge was 40x, meaning double Sandy Bridge, which is also fertilizer. GT3e may indeed double HD4000 (not really that great an achievement considering how unimpressive their IGP has been so far), but that will make it more like three times the power of Arrandale, not eight times or more. Let's also not forget that GT3e is one of four IGP designations for Haswell, and most Haswell chips will be significantly slower, not least for lack of onboard memory.

If they gave a more realistic accounting for their past achievements, I would be more willing to trust their claims for their future achievements. As it is, the marketing is really showing through in that graphic and calls into question the legitimacy of anything they say until they can back it up.

 

I agree. I'd like to know what metric they use for those claims. If it's merely bandwidth then likely. But if it's 3D performance related, not even close. And as you stated most chips will get the gimped version of the GT3. It seems they're doing this solely for trying to trump AMD. To compete with the GT 640 (that's desktop not even mobile) they'd have to exceed 2200 3DMark 11 performance. Not that I'm a firm believer in 3DMark as representative of actual performance, but I guess it's the standard we all measure by. They will be lucky to exceed 1200 3DMark 11. Even then I'll keep my excitement at a minimum until I see it's actual 3D performance numbers.

 

Actually, it depends what metric they chose as baseline for the performance chart. Look at 3DMark Vantage for example. HD 3000 (Sandy Bridge) scored 1500 in TOTAL score, HD 4000 (Ivy Bridge) score 3029. Thats double. Same with Arrandale and Sandy Bridge. Is it sneaky, to include CPU power as well, yes, but is it wrong, no

And if Haswell now really is double the graphic performance (not just total score) of a Ivy Bridge, thats not really that great? Seriously? Not only will Intel beat AMDs Richland, it will mean that Intel owns AMD in both CPU and GPU performance. And that is really embarrassing for AMD consider the gigantic hole between those two a few years ago...

You`re right. GT 640 score about 2200 in Graphic score in 3DMark11. I`d imagine 1300-1500 maybe from GT3. Thats not very far from GT 640M btw, since it score 1700. I remember a Dirt 3 video where they put 640M against GT3. But anyways, it was just a video where they showed no FPS or settings. A GT 640M will beat GT3e, but we are slowly getting there

 

cloud it was a 650m, in a POS notebook known as the U500 or UX51 from the asus prone to problems zenbook line.

But according to you guys (I dont play rally games) even the throttling that model has, it wouldnt affect so much the performance

I really still have good hopes of a score of 1600 ish in 3dmark 11 for the gt3e off course

 

Their chart shows GT 640 not 640m, and 1700 is still ~ 25% faster than a 1400 score. The Trinity 7660G already scores 1300, and I'm the one that showed that video of the 7660G next to a 650m at 1080p and you could not tell which was which either. Such a farce by Intel. I am not anti-Intel or pro-AMD, I'm just pro-reality. They offer too much ambiguity and not enough hard facts or numbers.

If Haswell GT3e is really 2x the performance of HD 4000 then that would mean, for example, 3DMark 11 of ~ 1200 to 1300 since HD 4000 is about 600-650. AMD is already equal or beyond that performance with Trinity.

 

No a 3610QM will score about 740 in 3DMark11. Double that you have 1500, which considering the specs should be perfectly plausible. There was also a recent update on HD4000 which promised 10% boost, I don`t know if that increase the score even further.
AMDs Richland APU, A10-5750M score 1400. Meaning that Haswell will beat Richland, a company that have never touched a graphic card is beating a company that have that as its expertise. Thats a pretty good development by Intel if you ask me considering Fusion APUs have always been so far ahead of Intel previously.

 

So.. do we know what the watt-drain will be in order to get a result like that?

And are anyone possibly interested this time around, after Ivy and Sandy bridge essentially toasts your average laptop if the gpu is run at normally at "high performance" presets?

We're possibly also interested in the actual gpu score, compared to the gpu score from earlier, to see the actual increase.

And maybe in whether or not the igp will support dx11 fully this time around.

 

Intel had its own integrated graphics for some time now... so saying that they 'never touched a graphics card' would be quite inaccurate.
I think their biggest challenge were their drivers (even though on the gpu side they were always well behind AMD).

One other thing... the difference between Haswell IGP and Richland (if those benchmarks are any indication) are mere 7%(100 points in 3dMark11), and that's with Haswell introducing modifications similar to the ones that Kaveri is supposed to feature (and Kaveri was stated to be a relatively larger 'jump' from Richland - just how much will that translate to performance remains a question).
This is effectively comparing Intel's architectural change to AMD's refresh.
Wouldn't it be more fair to compare Haswell to Kaveri (when it comes out)?

Also... we have yet to see in-game performance compared between Richland and Haswell (at least to my knowledge there hadn't been any such comparisons).

 

Do not get hopes so high, if they measured somewhere double performance than that was very special case, but sure not 3DMark 11. I assume only the 57W version flagship will be competitive with A10-4600M and the mainstream Haswell APUs just below that.
But for sure Intel like to beat AMD on paper (at least with the 57W monster), therefore you can expect little higher 3DMark score than A10-4600M's official (1150), so around 1200-1250.

 

as far as we know there are no 57w gt3e cpus, we only saw 47w ones

also haswell wont bring a change the igpu arch, which is the same since SB. Broadwell will implement changes to the igpu arch

depending on when kaveri actually launches it should be better to compare with broadwell, but I no less expect that kaveri should be much faster than haswell gt3e will be

and again the hd4000 where it matters, i.e. standard voltage cpus is around 700-800 in 3dmark11, not lower than that, even good cooled ulvs reach 700+

when we get gaming results we should have a better picture

 

Well in theory I suppose IGP and graphic cards are almost the same. I was just saying that since AMD is such a big player in graphics, they should not be beaten by Intel who doesn`t make dedicated graphic cards. I guess Intel`s famous fab comes to play here. They are one step ahead with nodes, Haswell 22nm while Richland is 32nm.

Kaveri sounds promising, with shared memory between the GPU and the CPU, but its out in late 2013. When that time comes, yes Haswell will compared against it, but not in June. In November or whatever when Kaveri is released, I expect AMD to surpass Intel by a lot again. Thing is though, that Intel is planning a brand new IGP architecture for Broadwell which will come out in 2014. So we might see Intel and AMD have a real graphic fight in the coming years, something that was not really possible with the gap that was between the two for a few years back.

100 points in graphic score is one thing. But since Intel have the clear advantage in CPU performance, many games who scale pretty good with CPU, will be in Intels advantage with Haswell vs Richland while the rest they might come even.

Like karamazovmm says, 47W is what will have the GT3e. 4850HQ and 4950HQ. Based on the specs I posted earlier you dont think 2x the performance of Ivy is possible?

 

not that it actually makes any sense to put the gt3e in a 57w extreme cpu, though I do expect that we have some 35w cpus with gt3e as well.

and according to my math on this thread they are quite possible actually

 

I can remember 57W parts, what were those? The 47W version sounds more fair against 35W AMD and bigger hope to see GT3e graphics in thin 14" laptops. I actually hope it could take the integrated graphics crown, than AMD should came out with a 45W part to get it back...
No, I do not think they can double performance again. While they could in the past because their integrated graphics was so week, but Ivy Bridge is not week in terms of integrated graphics.

 

ivy is weak on igpus, in x86 there are 2 companies, if one company is lagging behind drastically in one topic its considered weak. or in other words weak is defined by the strongest or the average.

700-800 is quite different from 1300

not to mention that 40eu vs 16eu is quite a far cry in terms of units that are actually doing some work. and with the 20eu version delivering 20% more performance, I dont see doubling performance as neither impossible nor improbable.

 

Well that is the other thing. 22nm Intel new gen just equal to AMD's current gen 32nm. Let's see what AMD can offer once they get to 22nm. /shrug/

I'll be sure to pit my A10-4600m with 7660G vs the Haswell GT3 (or whatever it is in i7-4850HQ something like that... confusing)

 

They doubled the eu numbers, added side port memory, can all these take the full advantage within similar TDP range? As Hopelessfaitfull mentioned, Ivy Bridge CPU cores already can take the whole TDP room. What is left for GPU?

That is gonna be very exciting test, hope Intel won't pay you off, to tell which benchmarks, settings to run .

 

not necessarily. and given that there is a drop of 400mhz, you can guess where the heat is coming out of

 

While that will certainly be accurate compared to Richland... we don't know how much Intel CPU performance will give it leverage compared to Kaveri (which shouldn't have modules that cripple its single-threaded performance).

 

Kaveri is still modular with improved performance, up to 3 modules (6 Cores). Because of new generation consoles, we will probably see better multi-threaded games where AMD APUs can benefit more.

 

Though AMD mentioned (if I'm not mistaken) that they added more FPU's per module (among other things which should result in about 30% to 40% higher performance in single threaded tasks), which effectively means that instead of having to share resources and have 2 modules behave like a single intel core... each core should behave much more in like like Intel's do.

 

They can pay me, and I'll still run the tests honestly. What could they do then? Say they paid me off? lol.

I am curious about the TDP too. Ivy Bridge definitely will consume the full TDP with CPU only. Wikipedia shows nominal TDP for GT2 and GT3e chips as 47W, but GT2 as high as 57W, and unknown for GT3e. You're probably looking at a 65W TDP with GT3e. Granted that's still less than a 45W TDP CPU + 35-45W GPU but not by much. GT3e CPU's also have a much lower base clock as well.

 

4800MQ: 47W @ 2.7GHz + GT2
4850HQ: 47W @ 2.3GHz + GT3e

They lowered the clocks to accommodate the extra EUs

Also, the CPU itself doesn`t use the whole TDP. That is wrong. Because if it did, then the CPU would run at much lower clocks when using the IGP.
Same logic is applied to the 4850HQ: Why not program the CPU to run at higher clocks when not using the IGP? They don`t do this because they use the 2.3GHz as max clock regardless to safeguard that the CPU never exceeds the TDP when firing up the IGP. Meaning there are more room for the CPU to operate when the IGP is not in use.

Atleast thats what I get from it.

 

cTDP is the programable TDPs available in the cpu.

The unspoken norm is that the OEM and entry line cpus of a said family for example: i5 3615m or i7 3630qm are not using the whole TDP presented in the CPU when doing only cpu apps, and will do use when they are using the cpu and gpu extensively, if you get the i7 3520m or the i7 3840qm while they share the TDP with afore mentioned cpus, they do dissipate more heat, on the case of those 2 much more actually

This was very easy to see in the rmbp 15 temps thread, which is still an imprecise way to measure tdp, but we get some rough results

As I said before, ivy doesnt use more than the TDP always, depends on which one. This started in SB actually

 

Intel HD 5100/5200 IGP = Intel Iris / Iris Pro.

AnandTech | Intel Iris & Iris Pro Graphics: Haswell GT3/GT3e Gets a Brand




The GT3e (Iris Pro) is exclusive to quad core processors with the 'HQ' suffix on the end of its model designation, Ultrabook processors will not get the Iris but still have decent boost in performance.

 

cant believe I missed it, do you have a cache to show? the page was taken down

 

Good to see progress, but that ULV processor doubles performance with an extra 2/3 TDP. To be fair, I'm excited to see the improvements with Haswell IGP at since lack of AMD options will probably leave me getting an Intel notebook this summer, but I'm still expecting Kaveri to wipe the floor with Haswell as regards IGP

 

I just switched to another computer and it's still there?

Anyhow, this is probably not the be-all, end-all IGP everyone was hoping for, especially when that particular one is only available in the high dollar processor SKUs.

 

well thanks the article is back again in their servers

and we already knew where it was going to be, I was hoping more info on 35w parts

and I love the increase in that yellow balloon, I remember that when I was 3 years old I drew something very similar

 

Those huge GPU performance increases, now I start to believe the twice GPU performance and 3DMark points. As I count the 28W version will already crush the 35W Trinity? That is very impressive!

 

So GT3e 45W is supposed to score 2,5x more than HD 4000 in 3Dmark11. It would score 1750 pts if you take 700 pts for HD 4000. Near 640M GT. Even a little more with the TDP pushed at 55W. It's very good but Kaveri should blast it without troubles knowing Richland is already at 1500ish at only 35W TDP. Though Intel igp will obviuously perform very well in cpu dependant games so Kaveri will have to bring decent cpu power to really crush it.

28W parts are pretty impressive also. 2X in 3Dmark11 is very good given the TDP. Of course it'll have to be paired with fast dual channel ram. But i'm not even sure it will beat 35W Richland in games despite its cpu power advantage.

 

With iGPUs (from both Intel and AMD) getting this fast, will any notebook manufacturer put a 100W desktop one in a notebook and sell it to people who need more computing power than a standard notebook one but don't need a high-end mobile dGPU?

 

The only notebook I know of that currently uses a desktop processor is made by Clevo and it's a powerful-GPU beast. I don't think there's a great deal of motivation for desktop CPUs in laptops; they don't perform much faster than laptop CPUs because they get throttled by heat issues and most people who want power to go want GPU power as part of it, whether it's for gaming or CUDA/OpenCL.

A high wattage desktop CPU in a laptop without a dGPU would be very much a niche product; there just aren't a lot of good reasons to use a machine like that over a machine with a powerful mobile i7.

 

and given that the R desktop with the gt3e is 65w, I dont see the reason

 

So an extra 8W of "boost" TDP (47W to 55W) offers a very marginal increase in performance, why offer it at all? That's a significant TDP hit for very little gain. In any case I'll hold further reservation until chips are released. Apparently the HQ chips won't come until sometime in Q3 2013, where the others will be available June 3.

And as noted before, 3DMark is not really indicative of actual game performance. You can have a GPU perform as much as 50% faster than another GPU in 3DMark, but in gaming the FPS are within 10% of each other. A perfect example is the Radeon 8870m which scores a 3300+ 3DMark 11 score, and nVidia GeForce 660m scores only 2500, but the 660m beats or meets the performance of the 8870m in most every game benchmark.

 

As for 8870M, it could be just a drivers issue cause it's a fresh gpu. 3dmark is pretty accurate most of the time.