AMD 8000M Series To Debut at CES

[RussianSensation]

Originally Posted by USER8000

It might not be slower. You are making the assumption it has the same performance or lower. It probably uses tweaked GCN shaders,appears to have greater memory bandwidth and also it probably uses boost clocks too. People judged the HD6870 and thought it would be "slower" than an HD5850 based on leaked specifications since it had far less shaders.

Ok how much faster can it be?

HD8870M
640 SP
700mhz
4500mhz GDDR5
128-bit
FP32 0.992 Tflops
FP64 0.062 Tflops

vs.

HD7870M
640 SP
800mhz
4000mhz GDDR5
128-bit
FP32 1.024 Tflops
FP64 0.064 Tflops

Even with Turbo boost and tweaks to GCN, I can't see much difference in performance. Comparing HD6870 to HD5850 is actually not making these products look good. HD6870 was one of the most underwhelming products to have come out from AMD. It launched 13 months after HD5850 and its performance was just 4-5% faster. Also comparing shaders of HD6870 to HD5850 is misleading because HD6870 was clocked at 900mhz and HD5850 was clocked at 725mhz. There were hardly any breakthrough efficiencies in HD6870, despite what AMD's marketing was spinning. It was a matter of upping clocks by 24% to alleviate the pixel fill-rate bottleneck AMD cards have every generation it seems and removing some shaders/TMUs to maintain power consumption/die size targets. The actual shader performance of HD6870 barely dropped (1120 SPs @ 900mhz = 2.016 Tflops vs. 1440 SPs @ 725mhz = 2.088 Tflops for 5850). Texture performance was also barely changed since the loss of TMUs was 22% but because of 900mhz GPU clock, this loss was largely mitigated. That was no magic, just fixing the pixel fill-rate deficiency, while minimizing the penalty to shader/texture performance. If you look at the specs of HD8870M, I am not seeing any clever tricks/rebalancing of the specs that address some particular weakness in a similarly clever way. Ya, the memory bandwidth went up, but barely.

I suppose since AMD doesn't position HD8870M as a replacement for HD7870M, then they'll say comparing HD8870M and HD7870M makes no sense as I just did, but then their marketing makes no sense.

[tviceman]

Originally Posted by RussianSensation

Even with Turbo boost and tweaks to GCN, I can't see much difference in performance. Comparing HD6870 to HD5850 is actually not making these products look good. HD6870 was one of the most underwhelming products to have come out from AMD. It launched 13 months after HD5850 and its performance was just 4-5% faster.

I personally thought barts was an amazing chip for it's size, speed, and power consumption vs. everything else on the market at the time. The only thing disappointing about it was it's name.

[RussianSensation]

Originally Posted by tviceman

I personally thought barts was an amazing chip for it's size, speed, and power consumption vs. everything else on the market at the time. The only thing disappointing about it was it's name.

I didn't. It came out 13 months later than 5850 and barely delivered 5% more performance, didn't fix tessellation issues of HD5850 and it was overpriced on launch compared to HD5850/GTX460s. In price/performance it got spanked hard by GTX460 cards that could be purchased for less and overclocked to 850mhz, at which point they were as least as fast. I also recall HD5850 market price for $180-190 or so. Launching 13 months late with barely any performance increase at $239 was not impressive in the least when a product with 95% of that performance was $259 more than a year ago. HD6950 2GB at least unlocked, went 2GB without a large price premium and had about double the tessellation performance of HD5870.

I guess we have different expectations of marketing names and performance increases. To me labeling a slower product as HD8870M is not something I think GPU makers should be doing since the average consumer doesn't follow GPUs like we do. From a consumer's point of view they would think it's faster than HD7870M. I guess AMD is replacing HD7600/7700M with HD8700/8800M and going to price them at the same level. From an overall perspective, the way I look at it is HD8870M serves as a mid-range laptop GPU part and yet it appears to be just a 25-30% lower clocked $100 HD7770 desktop part. In other words HD8870M will be barely better than HD7750 desktop part for games.

NV has 7 more GPUs above 650M and AMD is positioning HD8870M against 650M.
http://www.geforce.com/hardware/notebook-gpus

Based on this, I am guessing AMD is going to position HD8900M against GTX670/680M parts. How does AMD intend to compete with GTX700M parts? I guess they forgot NV is launching their 2013 Kepler parts too?

The marketing is just confusing.

[tviceman]

Originally Posted by RussianSensation

I didn't. It came out 13 months later than 5850 and barely delivered 5% more performance, didn't fix tessellation issues of HD5850 and it was overpriced on launch compared to HD5850/GTX460s.

I was just talking about barts in relation to it's die size and power consumption and speed in regard to other chips, not price. But I get what you are saying

[iMacmatician]

Originally Posted by RussianSensation

The next level up is just HD8900M series. It can't be 25-30% faster than HD7970M or they'll have a major gap between HD8870M and HD8900 parts.

Not saying that'll happen, but there's already a substantial gap (>2x the FLOPS and bandwidth) between the 7870M and 7970M.

There are two "Coming Soon" boxes above the 8800Ms. One might be a 768 SP Orust (whatever the Cape Verde successor will be) and the other might be a 1xxx SP Oland. (Well, if Oland has 1792 SPs as rumored, then that's still a huge gap. )

[The Alias]

http://www.theinquirer.net/inquirer/...hd-8000m-chips
it's official

[RussianSensation]

Originally Posted by iMacmatician

Not saying that'll happen, but there's already a substantial gap (>2x the FLOPS and bandwidth) between the 7870M and 7970M.

The confusing part is if HD7870M > HD8870M, then HD8950M replaces HD7870M? HD8970M replaces HD7950M? What replaces HD7970M, HD8990M (but it'll be a single GPU like the marketing misguided HD6990M)? You need an AMD GPU marketer on hand to understand their naming logic.

[iMacmatician]

Originally Posted by RussianSensation

The confusing part is if HD7870M > HD8870M, then HD8950M replaces HD7870M? HD8970M replaces HD7950M? What replaces HD7970M, HD8990M (but it'll be a single GPU like the marketing misguided HD6990M)?

I wouldn't be surprised. I'm sure they'll put some Oland as the cherry on top (8990M?) but what happens between that and the 8870M is anyone's guess.

Oh, and from Hardware.fr (via this post at Beyond3D forums), the new chip is 77 mm^2, and the 8500M/8600M series apparently have 64-bit buses (the 8700M series keeps the 128-bit bus).

[SickBeast]

I wonder if it will be a paper launch. If it's a hard launch that early then this is great news for AMD and for people who need a GPU upgrade.

[SickBeast]

Originally Posted by The Alias

http://www.theinquirer.net/inquirer/...hd-8000m-chips
it's official

So does that mean that these are just the mobile parts being launched or will there be high end desktop GPUs as well?

[Red Hawk]

Originally Posted by iMacmatician

I wouldn't be surprised. I'm sure they'll put some Oland as the cherry on top (8990M?) but what happens between that and the 8870M is anyone's guess.

Oh, and from Hardware.fr (via this post at Beyond3D forums), the new chip is 77 mm^2, and the 8500M/8600M series apparently have 64-bit buses (the 8700M series keeps the 128-bit bus).

64 bit buses.

what is this. I don't even...

No. Just...no. Seriously. AMD, did you just entirely forget about this whole fusion, heterogenous computing thing you've got going on? The whole point of it was to eliminate the market for 64 bit memory bus width GPUs like the 5450 and the 6450. And Intel is right there along side you making it happen with their iGPUs. And then you go along and make another? And you brand them as 8500M/8600M, when the last 64 bit memory bus width GPU under that number was the 4570?

You do realize what this means, AMD? Your Radeon HD 8600M might be slower than the Mobility Radeon HD 5650. Kaveri and Haswell might just eat them for breakfast.

I'm with Russian Sensation on this. What resort did AMD's marketing department just get back from? Because I want to book a trip there!

[3DVagabond]

Originally Posted by Red Hawk

64 bit buses.

what is this. I don't even...

No. Just...no. Seriously. AMD, did you just entirely forget about this whole fusion, heterogenous computing thing you've got going on? The whole point of it was to eliminate the market for 64 bit memory bus width GPUs like the 5450 and the 6450. And Intel is right there along side you making it happen with their iGPUs. And then you go along and make another? And you brand them as 8500M/8600M, when the last 64 bit memory bus width GPU under that number was the 4570?

You do realize what this means, AMD? Your Radeon HD 8600M might be slower than the Mobility Radeon HD 5650. Kaveri and Haswell might just eat them for breakfast.

I'm with Russian Sensation on this. What resort did AMD's marketing department just get back from? Because I want to book a trip there!

This could have just as likely been perpetuated by the OEM's. They are actually AMD's customers, not you and me. They want the model number to be higher to increase the perceived value of their product with the card in it.

[DiogoDX]

Originally Posted by SickBeast

So does that mean that these are just the mobile parts being launched or will there be high end desktop GPUs as well?

I do not know if it was serious or not but the "M" in 8000M indicates Mobile.

[iMacmatician]

Interesting note about the bus width: Since the same (128-bit bus) chip is used in the 8500M/8600M/8700M lineups, my guess is that the 64-bit buses of the lower two series may be due to some sort of product positioning/segmentation, since I doubt there are that many yield issues for a die so small. If so, that makes the narrow bus width even more ridiculous from a hardware specs standpoint.

[theprodigalrebel]

I'm in a bit of a dilemma. I put up a listing to sell my 5850 two months ago. Two people finally responded this week (I live in a country where trying to sell a GPU on the used market is near impossible - you normally end up holding on to it, like a relic worthy of cold storage; I still have a Radeon 9800 Pro lying around somewhere that I could never sell for a decent price). And they are offering a good price - about $150. The 7950/670 go for about $420 here. If I had sold the 5850 then, I'd been happy to spend $250-odd for the upgrade.

Now my dilemma is do I sell the card and buy a GPU - although I honestly don't have any games that are unplayable. I mean, the 5850 still plays all recent games at 1920x1080 at Low/Med settings (typically no MSAA). Upgrading will just allow me to bump up the visuals in BF3/Saint's Row III/GTA4 which really is all I play these days. I already have Civ V maxxed out.

Or do I wait till Q2 2013 and hope somebody is still interested in maybe giving me a few bucks for my old graphics card?

I am leaning towards the latter but it still feels like a tough choice.

[boxleitnerb]

Originally Posted by RussianSensation

So you are suggesting it's OK to release a crappy new generation because the low-end parts of previous generation were even more garbage based on rebadged HD6000 series? That's not exactly setting the bar very high when NV's GTX600 parts are eating away your market share Pac-man style in the notebook dGPU sector. Also, you didn't address HD7870M --> HD8870M, both based on GCN, the latter is positioned as the slower part. Desktop HD7770 1Ghz is faster than HD8870M. The next level up is just HD8900M series. It can't be 25-30% faster than HD7970M or they'll have a major gap between HD8870M and HD8900 parts. If it's barely faster, what was the point? Desktop HD7770 is selling for $100 on Newegg with 2 free games. 12 months later and all they could come up with for HD8870M was a 30% slower clocked variant of that product?

New generation by name only. I have given up to complain about "creative" naming in the OEM and notebook sector - with NV and AMD. To call these new chips HD8xxx is as bad as to call a heavily cut down GK104 "680"M(X).

[OCGuy]

This can't be a serious thread. notebook != desktop.

[riva2model64]

wait, if they're coming out with a Radeon 8800, that means that. . they'll have to release a radeon 9800! all over again, that card was king back in the day. or better yet, a radeon 9700. they could even use the pro/xt moniker again

[MarkLuvsCS]

It does seem pretty odd to release any low end dGPUs nowadays with both camps having iGPU options. I mean sure the HD4000 isn't going to give crazy FPS @ 1080p but it does at least make it playable at low/med settings. AMD's integrated solutions are typically even better than that as long as it's using dual channel memory. So where does that leave the dGPU that provides little benefit while being another power drain for the notebook.

[3DVagabond]

What's the performance going to be like if you use one of these with an APU and crossfire them? Would the perf/W make it an optimum choice?

[Silverforce11]

640 GCN SP is a lot more than 640 VLIW4/5 SP.

For a mid-range mobile GPU, that is very respectable, since its essentially 1/2 a desktop 7870 +/- tweaks and clockspeed (turbo up to 1.15ghz?) differences.

Its asking too much of mobile GPU that have to fit into very low TDP thresholds and limited by the external power plug. We know mid-range GCN is very energy efficient vs kepler, so this isn't a factor for their lacking in design wins.

[boxleitnerb]

Afaik you cannot CF Trinity/Richland+GCN, because they are different architectures (VLIW4 vs GCN).

CF is a bad choice for low end anyway, especially when you use differently performing GPUs -> worse microstutter.

[3DVagabond]

Originally Posted by boxleitnerb

Afaik you cannot CF Trinity/Richland+GCN, because they are different architectures (VLIW4 vs GCN).

CF is a bad choice for low end anyway, especially when you use differently performing GPUs -> worse microstutter.

This says you can mix different generations for crossfire. Or, at least they planned on it. Whether or not it was ever actually implemented, I don't know. It does raise the possibility of mixing GCN with VLIW4 though.

[boxleitnerb]

PCGH tried it and it is not possible currently as it seems.

[ShintaiDK]

Originally Posted by RussianSensation

Ok how much faster can it be?

HD8870M
640 SP
700mhz (775Mhz Turbo)
4500mhz GDDR5
128-bit
FP32 0.992 Tflops
FP64 0.062 Tflops

vs.

HD7870M
640 SP
800mhz
4000mhz GDDR5
128-bit
FP32 1.024 Tflops
FP64 0.064 Tflops

The Tflops numbers you listed for the 8870M only applies at full 775Mhz turbo. Without its 0.896Tflop. AMD took a 7870M chip, lowered the clocks and enabled turbo while speeding up memory. Placed a new sticker and called it a day. AMD writing 992Gflops on their slide without saying its due to turbomode is...misleading.