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

while I want to see kaveri in action, its based on piledrive, which is a very different design from anything that the xbox and ps4 will pack.

And HSA needs apps to be coded to use it, its not something that you magically happen to put and everything is unicorns, it will take time, how much? I dont know. Is it a good move? its the only possible move

we could see how that HSA works tday with the 4950HQ review, along with how the edram will work on intel system, how it will work in AMD system I have no clue

 

According to latest news no GDDR5 for Kaveri.

http://www.brightsideofnews.com/new...aps2c-shows-28nm-kaveri-for-socket-fm22b.aspx

 

Based on past AMD/Intel iGPUs and costs of those APUs by both. Other than that I got nothing. Pure speculation.

 

now that is a bummer, I hope they can come up with something very decent

 

They don`t need GDDR5 for a simple IGP. Thats overkill delux

 

If the processor designation (i7-4950HQ) is any indication, that's going to be a very expensive part (typically the X9XX parts are over $1000/chip, though the 4900MQ is listed at $568 on Intel's site, so it's more likely to be that plus the GT3e upgrade cost which will probably be significant), and that article states that it's the lowest-end part with GT3e. Intel's site doesn't seem to show any GT3 or GT3e parts right now, but considering that similar GT2 processors are listed at well over $500/per, it's reasonable to suppose that a system with GT3e is going to cost a lot of money.

It's also a 47W part - between the gaudy price and the high TDP, Intel's going to have a hard time squeezing these parts into the kind of dGPU-free machines that would attract a lot of customers. A chip with that TDP (did you notice that the Anand reference machine was a desktop?) isn't going to work in an ultrabook-style chassis, and if you get much bigger than ultrabook size, you can find machines with i7 quads and better-than-GT3e GPUs for less than a GT3e machine is likely to cost, making its usefulness on the market shaky until Intel can prove it has its place. At that TDP and price, GT3e isn't competing in the iGPU realm, it's competing with dGPUs, and that's a fight it's not going to win.

Let's also not forget that, like it or not, Anand is a name that has long been associated with biased tech journalism, so as much as it delights you to bump up the font size and tell us how the not-yet-released 47-watt flagship processor beats a 35-watt APU that's been on the market for 14 months and costs less than a quarter of the price, you're going off one set of numbers devised on a test system built specifically to allow the CPU to run optimally in a way that it won't necessarily be able to do in a real laptop, especially not one designed to take advantage of the chip by billing itself as a thin-and-light powerhouse.

 

not really, it could work as L4 cache like crystalwell, though it would need some serious tweaks to work with the cpu and the gddr5 itself

 

4850HQ = $468
4750HQ = $300-350?

Both with HD5200

AnandTech | Hit the Road, Jack: Intel


@Karam: Crystalwell use plain DDR3 on 128bit. Worked well for them. Why you assume Kaveri need so much more bandwidth?

 

I am impressed with the GPU performance, but again it does all come down to price and Intel's listed price means squat at the retail end. Plus we don't have any idea on temperatures or CPU performance with IGP engaged. If they can come out with a quality well equipped 13-14" machines for < $700 with the 4850HQ then I'll be impressed. Otherwise the GT 740m with a 4700MQ will likely cost under $800 and perform better.

And as I stated somewhere else (wow these threads become so redundant and entangled) the Anandtech performance of the 7660G and 650m are low by 15-20% from performance I've measured.

 

I couldn't find those H-processor prices on Intel's site or a quick websearch. Judging by those numbers, $375+ sounds like a much more realistic number for the 4750HQ, though it's nice in that it's much less than the 4950, which Anand apparently misrepresented as the lowest specced GT3e processor. They're still 47w, which makes them restrictive for the kinds of machines you really want to offer fast IGPs, and it's still kind've childish to call this an in-your-face moment to AMD fans because of GT3e being faster (in a desktop setup on its $678 flagship processor) than a year-old processor that costs significantly less than any GT3e part. Granted, AMD has screwed the pooch for a long time thanks to not getting its APUs in quality notebooks, and Richland is, IMO, a debacle to cover the fact that Kaveri should have come out at the same time and as a direct competitor to Haswell, but I'm still rooting for the underdog because their product is more along the lines of what I want, and cheering for the team that always wins is boring.

What I think I'm getting at here is that you must be a Yankees fan...

 

again, the arch on the gpu is as old as sandy, they added more EU and tweaked some things. I want to see a battle of new VS new, I want to see kaveri VS broadwell

intel doesnt compete on price at that market, it competes on performance, or tries to, there is no performance from amd coming near those guys.

intel competes on price where the gt3e aint available, and its simple really, those cpus are not to compete with anything that amd offers, they are not meant to. AMD doesnt have good design wins, quality thin and light notebooks, the maximum you see amd is the gx series, which in turn is not enough, the other maximum in terms of what they deliver is the very hard to find asus "ultrabook", those envy 4t and envy6t are pos.

For average users, AMD is quite enough, but its delivered in a cheap package with the pricing of i3 and pentium offers, probably to recoup costs spent on the cheap intel sku, while it could be delivered in a premium package at 700, or at least something resembling quality

@cloud edram is edram

 

Actually, Haswell's iGPU is Gen "7.5", because Ivy Bridge is Gen 7. Why is Ivy Bridge Gen 7? Because each EU can output 2x the flops of Sandy Bridge.

Gen 4: GMA 3 series
Gen 5: GMA 4 series
Gen 5.75: Clarkdale/Arrandale
Gen 6: Sandy Bridge
Gen 7: Ivy Bridge
Gen 7.5: Haswell

Pricing would be $468, at least the official one, because the lowest quad core CPUs are listed as $378, and GT3e adds $90. Also, the 4750HQ is showing "SBA" on the list, meaning its a business oriented SKU(honestly, GT3e doesn't make sense for a business oriented SKU but whatever). The 4850HQ is the "gamer" oriented one, if you ask me.

 

the design of the EU remains largely the same, with tweaks on the pipelines aint it? the SUB-fil are basically what sb delivered.

vliw4 aint that new either, its from 2011

and another thing, if you flop a lot, will that make you a flopper or a loser?

PS: I shouldnt have gone to that stand up, terrible, just terrible

 

If you say it like that though, you can say that Broadwell is based on the same generation, and even GMA 3 series does too. Broadwell will beef up some weak parts, and probably boost Flops/EU, but other parts will be largely based on previous generations.

Also, Intel uses a different architecture from VLIW. Basically, the difference is, while VLIW4 can do 1 FMA per SPs, Intel's EU has two pipelines, of which each can do 4 FMA operations.

Made me smile though, that counts.

 

So why only Anandtech tested GT3e, the site which last year under measured A10 Trinity performance and loves badmouth AMD?

 

Techreport has a GT3e as well, they are working on it. Anand tested a very fast 650M= 900/2500 GDDR5

Usually 650M in notebooks is slower. I think GT3e is comparable to 650M with DDR3. Big advantage is that GT3e can reach that performance level within a 50W power envelope while a 650M alone eats up 50W or so.

 

Then if it's 900/2500 those numbers are PLAIN WRONG. I get much higher results from a 900/1800 DDR3 650m, like as much as 30% higher.

 

So haswell onboard GPU will make even the weakest laptops perform like a nividia 650m?

Thats pretty cool, than everyone can do some gaming and people dont have to worry about specs that much anymore

 

Don't quite trust those results yet. There are a few things to consider:
- those results are at least 20% less than they should be for the 650m based on what I've tested
- cost of the CPU/system. The system with a 4850HQ may cost more than a system with a GT2 Intel CPU and GT 750m and will be 50% faster.
- power consumption. CPU's without the GT3e already consume about 90W at full power load
- stutter. Any IGP I've used and tested have had horrible stutter in some games. It makes 30-40FPS look and feel like 10FPS, so numbers don't tell the whole story.

 

ok but i hope its true, than dedicated GPU's would only be for the specially interested gamers and such who want top performance in the newest games.

 

The answer to your first question is no, not yet. The weakest laptops will not have the GT3e chips everyone is talking about -- they'll have the significantly wimpier GT1 or GT2 parts -- and even the GT3e does not quite match the 650M. However, your second statement is where things appear to be headed. Intel is clearly increasing the die area devoted to the GPU. Their integrated GPUs used to be the laughingstock of the industry, but now the mobile GT3e beats all other integrated solutions (including even the desktop Trinity). The "base" configuration has gone up to the point where most (although not all) available games are playable (not at the highest settings, but not looking like year 2000 graphics either).

I am actually quite curious about the die shrink of this (Broadwell). It's hard to get a die shrink to increase CPU performance much, but GPUs work with parallel workloads so adding more of the same causes a nearly linear increase in performance. The current top-of-the-line incarnation is quite awesome (I didn't think it could beat desktop Trinity!), but with a die shrink they can bring even the lower-end parts up to this level.

 

Im not saying like that, the principles of how that EU operates are still extremely similar. But they did put a very different thing here

AnandTech | Intel's Haswell Architecture Analyzed: Building a New PC and a New Intel

I know intel uses a very much different arch from vliw4, which is what is in trinity right now (and that was the point), and the trinity with a speed bump (richland)

thats something at least, I didnt even find it funny

 

Spinning has initiated I see

 

Just because it's not in large bold font doesn't mean it's spinning Cloudfire. There's more than just performance numbers involved here. I don't doubt that the GT3e has good performance. I don't doubt that it is a significant first for Intel to compete with low end dedicated GPU's. But it gets back to cost and usefulness. If a system with a GT3e CPU has to be 14-15" and cost $1000 then it really has limited appeal because you know there will be much faster machines at that same price point with a dedicated GPU that will put the performance of the GT3e to shame. If Intel can manage to shove that chip in a 13" form factor, keep it cool without throttle, and cost less than $700 then they've won my interest.

 

700 is unlikely, but you are right, that should be their target, prime candidates new sony z, rmbp 13, asus crap in 13 size, and so forth.

 

Right. But my whole point is if it only works in 15" notebooks due to cooling and power, you can find some pretty good laptops for $800-$1000 with a dedicated GPU that will double the performance of the GT3e.

 

agreed. that power consumption figures got me really worried in notebookcheck.

 

HT, the pricing scenario reminds me of something you said the other day. It went something along these lines:
For AMD APUs: Their products are great but if OEMs are putting them in notebooks nobody wants then whats the point?

I say:
For Intel APUs: Their products (GT3e specifically) are great but if OEMs are putting them in notebooks MOST cannot afford then whats the point?

I agree, price and usefulness.

 

Do you think it's fair to compare an i7 to an i3 on a pure raw performance, and base the i3?

No, because like the CPUs, the GT3 and the APUs' IGP serve different price targets.

 

actually depends. and depends on a lot of things

We have to understand that what most people buy, dont offer a good benefit for the OEMs right now, or back them when it started. It simply doesnt, there is no way you can cut this to me that will change me or the numbers.

Cutthroat margins, killed and moved to change the pc industry. Its a major indication that the cheapest most unreliable OEM, Acer, has basically and purposefully continues to move out of that sector.

and thats the thing, its like what do we do to compete against say basic utilities like pans and fridges if the chinese can produce and ship across the world and deliver it in our shores at a much lower cost? Is there really a point in competing at that place? PCs are more expensive products, but not much more than actual good set of pans. How is that possible? its simple stamping!

Simply put, intel has the luxury to let amd keep themselves at the 600, and they are quite happy with that, OEMs on the other hand that still insist on delivering the cheapest most unreliable trash possible, are still tied to that thinking, HP is still one of those, dell as well, sony I dont know why in the world tried that and is trying to back out

Simply put I dont care what most will buy, for the OEMs that did that, its death, for intel its yay, for amd its yay, and for us that try to keep a notebook for 3-4 years is consumerism at its best, i.e. we are in politics of carrot and stick, shove the carrot and beat us with the stick.

I think entry level should be at 800, and high end double that.

GT3e is not going to happen at 600 bucks, its not. GT3e wasnt designed to compete against AMD igpus, it was designed for premium thin and light notebooks, it was designed for workstations.

intel with gt3e aint delivering a gpu that can only be used for gaming, its delivering the cache and gpu power for pro apps

 

Comparing performance independently of price is not unreasonable. Prior to Haswell, Intel simply did not have an integrated GPU that beat Trinity -- regardless of price. If you did not want a laptop with a discreet GPU, your choices were to get a slow CPU (AMD) or a slow integrated GPU (Intel). No amount of money would get you a chip that was best at both; the thing simply did not exist. Haswell changes this.

Agreed. This chip is for the MacBookPro and the like.

 

If it's not meant for gaming then why does Intel keep showing 3DMark performance numbers and images of gaming characters in some of their marketing slides?

 

because its catchy, and of course you can game on it. doesnt mean that this is going to be the direction that they are going to follow in their future cpu development for workstation users

 

Why would pro apps benefit more from "cache and gpu power" compared to games? The both need compute power.

Even if the i7Q/GT3e is an option for workstation workloads, Why worry about iGPU performance on a workstation, you'll be using the dGPU anyway. Even if there's no dGPU available, you'll need DCC app venders to optimize for Intel, which, if the history is any indication, takes forever.

Also, "thin and light notebooks" and "workstations" appear to be very different things.

 

I thought the same thing. Thin and light and workstations are complete opposite ends of the spectrum. I can see thin and light, but I doubt it will be in many other than the MBP.

 

If MBP counts, some other slim gaming laptops are also in the category.

Non of them count if large drive space, reliable power & cooling, ws GPU (I'm not even talking about wd/server CPU here) and abuse-resistant casing are criteria for workstations.

Whether those matter or not, of course, depends on the use case. Workstation workloads are diverse, usually much more diverse than AAA gaming.

 

That's kind've a silly way to put it: "if you want a laptop without a dGPU..."

Most people want a laptop that can do X or fit in a certain size package. Due to their TDP, GT3e parts are unlikely to fit in a notably smaller package than a faster, less expensive notebook with a dGPU.

 

I would say your comment is nonsense. Or how can you reproduce anands game tests? You don't know what scene he tested. You could only reproduce 3dmark results. So if you have 3dmark results from your 650M share with us.

 

They explained what they tested in their article if it wasn't a built-in benchmark:

Metro Last Light - Built in benchmark
Bioshock Infinite - Built in benchmark
Sleeping Dogs - Built in benchmark
Tomb Raider - Built in benchmark
Battlefield 3 - Single player and they even note " Our goal here is to crack 60fps in our benchmark, as our rule of thumb based on experience is that multiplayer framerates in intense firefights will bottom out at roughly half our benchmark average"
Crysis 3 - No mention of where benchmarked, but indicative of general performance of the game, it doesn't vary much except in certain scenarios
Crysis Warhead - Frost benchmark
GRID 2 - Built in benchmark

And LOL, just check my sig you'll find every kind of benchmark for the GT 650m, 680m, Intel HD 4000, AMD Trinity 7660G and AMD Llano 6620G.

Here's my best 3DMark11:

http://3dmark.com/3dm11/3945706


But here you go rest of GT 650m:
http://forum.notebookreview.com/sag...er-np6110-clevo-w110er-first-look-review.html
http://forum.notebookreview.com/sag...dual-core-vs-quad-core-gaming-benchmarks.html

 

A dGPU inevitably means a heavier notebook and almost inevitably means a larger size. The TDP of the GT3e is usually 47W (there's supposedly also a 37W version). Can you think of a CPU/GPU combination that has a lower TDP with similar performance? For example, the 650M that people keep talking about has a TDP of 45W alone -- add a decent CPU to that and you'll get close to double the TDP of the GT3e.

 

and they are very different things. doesnt mean that you cant produce for both, check ivy ex and xenom, same cpus different purposes

still to the point, the idea is that that edram cache would help on certain apps, thus, workstation. There is a reason that IBM put those things at their servers, its not gratuitous.

and for the thin and light premium, you have that igpu performance

that much is true, I dont even see that happening on the rmbp 13, which would be the prime candidate, it doesnt mean that those OEMs cant produce new models that can take advantage of the HQ models

no one can claim that the mbp or rmbp are workstation, far from that actually, they dont fit in any criteria that can be used to define workstation

its exactly like that, most people arent concerned about the specs, its actually boring and irritating for them, they really want something that works for their lives, simple as that

and again its not going to be in inexpensive notebooks, no one in their right minds would do that, 600 bucks notebooks? thats 2/3 of the cost just for the cpu, 900? more plausible, still if models are to go by we dont have much of those dgpu less with quads floating around, though it can change

Im not anticipating a major success for haswell or anything like that, Im merely pointing what I think was their choices, and in part how the market will behave

most sales will still concentrate on the triangle, 600, 15, crap screen, and that triangle aint served by those cpus at all.

 

Would you please give a specific DCC-related application/implementation/algorithm as an example?

Server workloads are quite different from what workstations usually do.

 

That's a fair point.

The difference won't be as big as 1:2 (CPU power draw would be much lower when you're not using the iGPU), but one single chip with lower power consumption would still be easier for a smaller/slim machine.

 

The problem is that you can't spread the heat of the chip over the entire chassis, it's centralized around the CPU. Since you have to accommodate the localized heat of a 47W chip in the chassis, very few designs currently exist that could deal with the heat of a 47W CPU running at full blast but couldn't also fit a dGPU in another part of the chassis. With a dGPU like the GT650M (or its 700-series or Radeon equivalents) you're talking about a few ounces of weight added to a chassis that can already accommodate the space it takes up, with much better performance in games or (esp. for workstation equivalents) CUDA/OpenCL productivity.

Give me GT3e in a chip no higher than 27W (accepting the performance losses of the lower TDP) for $300 or below and I'm excited. A 47w GT3e that costs as much as a normal i7 + mid-range GPU combined doesn't excite me in the slightest.

 

currently no idea, but all software should benefit from it, specifically one or 2 I cant provide at all.

its indeed faster than to access ram, to storage your workload, when we code things to benefit from it we will a boost in performance.

and you are right, its very different workloads, doesnt mean we cant take advantage of that

then we better wait for june 4th

 

So I clicked on both of those links, and I could only find Bioshock Infinite and Sleeping Dogs in the benchmark. Also, 3DMark11 score for your 3610QM HD 4000 gets 873, while Anand's NV56M with 3720QM(which has a faster CPU and GPU) got only 800. The HD 4000 in the Iris Pro benchmark gets even slightly less than 800, despite being a desktop 3770K version, and despite that latest 15.31 drivers improve 3DMark11 scores by 5% or so.

According to Bioshock Infinite, Anand's results show 59.6 fps for the GT 650M and 54.6 fps for the GT 640, which is a 925MHz part with 1.7GHz DDR3, which means bandwidth impacts are pretty small. Anand's results show 59.6 fps for GT 650M, while your GT 650M gets 61-62 fps with DDR3.

So let's assume GDDR5 will increase your fps by 10% to 67-68 fps. Fair enough. But then again, your HD 4000 shows it beating Anand's results with quite similar degree. That's why we need to see more reviews with GT3e rather than claiming "oh Anand is wrong simply because...".

 

I see what you're getting at by suggesting that all of HTWingNut's results are faster than Anand's by a similar margin (suggesting perhaps that he would get 10% faster results with GT3e than they did), but you failed to finish the last sentence: the claim is "Anand is wrong simply because they have a long history of biased journalism, and the known-quantity (i.e. parts available on the market) benchmark results they publish in the article in question do not reflect actual results submitted by users of those parts, calling into question the legitimacy of all of the benchmarks in said article."

 

Hey guys I have a question if you know.....
The Mobile Intel® HM87 chipset is it compatible with sandybrige-ivybrige processors ??
Thanks

 

I don't think either AnandTech or HTWingNut is necessarily wrong. Modern chips (both GPUs and CPUs) from all three major manufacturers (Intel, AMD and Nvidia) are very good at adapting to the thermal environment they are in. HTWingNut used a Clevo machine and such machines usually have some of the best cooling around. AnandTech's results are for a MacBookPro which tends to be as thin and light as it can be without frying the enclosed hardware. It's entirely possible that they do actually get different results.

 

no .