[techpowerup]piledriver emerges.. vishera details show promise:quad channel ddr3, HTA

[SickBeast]

so why are we supposed to be excited about an unreleased cpu that will be slower than a 2500K that came out over a year ago?

bulldozer is so bad that it should be scrapped IMO. they shouldn't be trying to re-vamp it. it's like flogging a dead horse.

 

[BallaTheFeared]

It's like flogging eight dead horses, then adding two more dead horses to the pile..

 

[Rvenger]

More cores = more power consumption... period.

 

[BD231]

Four channel DDR3 for CPU's that can't even saturate AMD's current dual channel bus?? Don't get me wrong I'm interested in seeing what I can throw into this AM3+ board but that's just pure marketing there's no reason for that. Hopefully the interconnect jargin has a lot more to do with it, as AMD's PR team needs to be lined up and blind folded don't trust a word you hear from them.

 

[Tsavo]

Meh, quad channel memory is so last year.

I want 5.1 channel memory so my data transfers will be in Dolby.

 

[frostedflakes]

If it was quad channel it wouldn't even be AM3+, that would require a new socket.

 

[Smartazz]

Bulldozer isn't good, but it's certainly not totally worthless. It's decent in multi threaded applications and it's not outrageously priced. Still, Zambezi is far less appealing than the 2500k and 2600k.

 

[TakeNoPrisoners]

10 Cores

I stopped reading right there. Looks like AMD is going for quantity over quality. Looks like AMD is trying its best to alienate gamers. :thumbsdown:

 

[BD231]

If it was quad channel it wouldn't even be AM3+, that would require a new socket.

:sneaky:

 

[IlllI]

w/o bothering to look at the article, i would assume that vishera is some kind of server chip?

 

[Phynaz]

bulldozer is so bad that it should be scrapped IMO. they shouldn't be trying to re-vamp it. it's like flogging a dead horse.

But what do they sell in it's place? They have nothing.

BD is it for the next 3+ years. Get used to it.

 

[IntelUser2000]

as far as techpowerup article is aware L1 cache i doubling from bulldozer to piledriver.

It isn't L1 cache that's doubling, it's L1 TLB.

Piledriver is similar to Ivy Bridge in a sense that per clock the improvements will be marginal. The big gains they claim over previous chips is when its compared to Llano, since Trinity will have a significant clock speed advantage.

 

[frozentundra123456]

The only place AMD interests me now is the mobile area. If Trinity can improve the APU some more and also improve the CPU it will be very attractive. Are there any major improvements coming on the E-450? That chip is very attractive in a netbook, but has not been improved for a while. Otherwise, give me an Intel quad and a discrete card.

I got some gift cards for Christmas and seriously looked at buying a Llano laptop. I didnt even look at Intel laptops because I wanted to be able to use it for light gaming and did not want to go up to the price range of an i7 and discrete card. But I ended up getting a cheap tablet for e-mail and light surfing the web when at work away from my main computer. I will just game on my desktop. So far I am happy with the tablet. I got an Acer 7" open box for 200.00 so I feel it is worth it. However, I am not sure a 500.00 tablet is worth it. In that range I would prefer a brazos netbook.

 

[IntelUser2000]

That chip is very attractive in a netbook, but has not been improved for a while.

That chips is between rock and a hard place between their own Llano chips and Celeron/Pentium chips.

 

[infoiltrator]

Throwing this out for response.
On desktops the basic gpu need is a moving target, as I understand it. Outside of Crossfire/SLI I do not understand the avaiabilty and pricing of low end cards (8400, 210, etc).
On APUs the problem I have is they do not seem future proof, especially if there are fundamental changes. If GPU memory becomes more important in CPU functions where does that leave APUs dependent on shared memory?
Is this a valid or silly concern?
What effect will increased bandwidth capabilities make on discrete GPU lowend performance?
If I were Intel HD3000 would become the base in chip video and an HD3900(3890?)
would appear much improved.

But for fundamentals, future proof prospects, what shows more promise?

 

[Concillian]

Four channel DDR3 for CPU's that can't even saturate AMD's current dual channel bus?? Don't get me wrong I'm interested in seeing what I can throw into this AM3+ board but that's just pure marketing there's no reason for that.

It's for the GPU side on their APUs.
The dual channel bandwidth is definitely limiting GPU performance unless you OC the memory up to 2133 or something. Quad channel 1600 will be a benefit there.

The whole thing points to them continuing to target the lower end market, which is what they did with llano and that worked very well for them. They need to be sure they stay well ahead of the IB GPU to keep some advantage in that lower end market, and quad channel memory is a very important part of keeping that lead. I think it's absolutely critical, since they were already bouncing their head on the 2 channel bandwidth ceiling with llano.

As for being interested in throwing something into an AM3+ board, they've stopped making CPUs that aren't 32nm. Right now that means BD and FML socket llanos. These are clearly a new socket at quad channel, so looks like AM3+ is done.

 

[ShadowVVL]

Quad channel would make alot more sense for the apus.

Would be a bonus if they can get apus and non apu cpus running on the same board.

 

[nismotigerwvu]

Would be a bonus if they can get apus and non apu cpus running on the same board.

I thought that was the general plan all along. The main reason it has been a bumpy transition so far is that AM3 was so established in the desktop space, but lacked the routing for the GPU side. FM1 never really stood a chance for gaining significant traction, but FM2 should be able to get things moving down this path. I doubt we can have as clean of a transition as AM2/AM3, at the very least comparable (but incompatible) Piledriver chips should launch side by side on AM3(+) and FM2. At the very least this is tossing a pretty decent bone upgrade wise to existing owners and offering an enticing package (PCI Express 3.0..ect) to move to the new platform (FM2).

 

[Obsoleet]

The only place AMD interests me now is the mobile area. If Trinity can improve the APU some more and also improve the CPU it will be very attractive. Are there any major improvements coming on the E-450? That chip is very attractive in a netbook, but has not been improved for a while. Otherwise, give me an Intel quad and a discrete card.

I got some gift cards for Christmas and seriously looked at buying a Llano laptop. I didnt even look at Intel laptops because I wanted to be able to use it for light gaming and did not want to go up to the price range of an i7 and discrete card. But I ended up getting a cheap tablet for e-mail and light surfing the web when at work away from my main computer. I will just game on my desktop. So far I am happy with the tablet. I got an Acer 7" open box for 200.00 so I feel it is worth it. However, I am not sure a 500.00 tablet is worth it. In that range I would prefer a brazos netbook.

I'd like an AMD ultrabook. I look at my ultrabook of choice (Samsung 9 series) and find myself massively disappointed it's still not good for light gaming. I have a similar laptop I'm typing on now (Intel 640M) and it's not quite good enough, almost.

So that leaves me waiting for Intel to get their game together (Haswell/Broadwell), or, just go AMD for mobile.
Which I'm fine with, but Intel has the market on the well-built, sleek ultrabooks..

I'm willing to pay $1,000 for an AMD ultrabook that's built like the Samsung 9 series or MBA, but it's impossible to find one.

Oh and Bulldozer/PD? Who cares. That is not the market focus IMO. They are fast enough for my uses on the desktop (not that I'd buy one), and I'd rather see innovation in APUs.

With a decent APU from AMD like Llano or better, I'd have everything I need for every build going forward. Mobile, and the best recommendation for my parents or grandparents for desktop builds. For example, just selling a nice all-in-one using an AMD APU rather than a traditional desktop build like we use here at AT appeals to me more at this point.

I have no major qualms using a 8150 FX for my desktop. Faster than what I have now, and faster than what most people use. That said I won't pay for it, unless it was cheaper.
For APU, it's AMD hands down.. and I'd like to see more focus on that, and small improvements on desktop Bulldozer because that market is less relevant.

You can dominate with inferior product, it has nothing to do with the speed of Bulldozer cores.
Intel did it for years with Netbust. It's EVERYTHING ELSE that AMD needs to worry about and focus on, not fixing Bulldozer.

 

[IonusX]

Throwing this out for response.
On desktops the basic gpu need is a moving target, as I understand it. Outside of Crossfire/SLI I do not understand the avaiabilty and pricing of low end cards (8400, 210, etc).
On APUs the problem I have is they do not seem future proof, especially if there are fundamental changes. If GPU memory becomes more important in CPU functions where does that leave APUs dependent on shared memory?
Is this a valid or silly concern?
What effect will increased bandwidth capabilities make on discrete GPU lowend performance?
If I were Intel HD3000 would become the base in chip video and an HD3900(3890?)
would appear much improved.

But for fundamentals, future proof prospects, what shows more promise?

gpu wise apu's do just fine actually trinity's flag will feature something even with a juniper pro (hd 5750) which is NO laughing matter. at the rate amd is improving their apu's graphics their third apu graphics will be in a dead heat with an hd 6870
that means in 2 cpu gens a single amd product will be able to max bfbc2 and saint row 3 @ 1080p /jawdrop
lets see an intel IGP do that mmk? atm they can abrely run sfIV properly (hd 3450 area)
assuming things move at all intel will be replicating a caicos (HD 6450/HD 7400) NEXT YEAR. and then the 6570 the year after.
intel wont be maxing contemporary games anytime soon...
the apu's hybrid cfx is also fantastic an hd 6990D (6670 + a8-3850/3870k) can do some real wonder like play dirt 2 on high and nearly achieve playable fps on crysis 1 on high (varying on res.)
amd's apu's are actually a very smart move and if they would only put more effort into them than it could really move units.

 

[Concillian]

gpu wise apu's do just fine actually trinity's flag will feature something even with a juniper pro (hd 5750) which is NO laughing matter. at the rate amd is improving their apu's graphics their third apu graphics will be in a dead heat with an hd 6870

Well, it'll take 2.3 GHz DDR3 quad channel to pull equal to 1.15 GHz GDDR5 in terms of memory bandwidth... But it will be close. You're certainly right that this is a LOT more powerful than typical discrete GPU options you see on your average low end Dell (usually 5570 or something like that.)

It's great for teenagers who are using their parent's computer and have no option for discrete. It's great for laptops. For the rest of us it's lame.

It'll be a while before we see 6870 level graphics in an IGP though. You need 8 channel DDR3 to even come close to the bandwidth. I don't really see that happening. We need DDR4 or DDR5 on the desktop before we start seeing that level of memory bandwidth on the desktop. In the meantime, it'll get stalled somewhere between a 5770 and 6850.

Really, that's not a bad place for IGP to be stalled though, you can play at 1920x with a GPU that powerful. Not with MSAA and ultra settings, but it's powerful enough to play real games at 1080p with much better visual quality than a console.

amd's apu's are actually a very smart move and if they would only put more effort into them than it could really move units.

You realize they were moving as many as they could make for all of last year, right? They are already "really moving units" and they stopped making 45nm CPUs as a result, so they could increase APU volume. APUs have been a good move for AMD, even though they are not really at all exciting to the kinds of people who read these forums.

 

[MrTeal]

gpu wise apu's do just fine actually trinity's flag will feature something even with a juniper pro (hd 5750) which is NO laughing matter. at the rate amd is improving their apu's graphics their third apu graphics will be in a dead heat with an hd 6870

I think you're being incredibly optimistic.

Finally the desktop Trinity will be 15% faster on the CPU side and 25% faster on the GPU. Although AMD didn't disclose details, it's likely that these numbers are comparing a two-module Piledriver based Trinity to a quad-core Llano.

That's from Anand's CES coverage on the 12th. The A8-3850's GPU is 10-20% slower than a HD5570 on a 3850, matching a 5750 would require a 100%+ increase in GPU power. Getting the performance of a HD6870 into an APU in the next couple years is just crazy. That's a 1.7B transistor die that pulls 150W under load. GCN is an impressive arch, but it's not THAT good.

 

[Ajay]

It isn't L1 cache that's doubling, it's L1 TLB.

Piledriver is similar to Ivy Bridge in a sense that per clock the improvements will be marginal. The big gains they claim over previous chips is when its compared to Llano, since Trinity will have a significant clock speed advantage.

Exactly! I think the most likely reason that Trinity is performing above expectations is that GF is getting their act together and hitting higher clocks. Though, the PD arch may be better suited to higher clocks as well.

I'm not surprised that Vishera will consist of 5 modules since that is AMD's main advantage over Intel at the moment in certain heavily threaded server workloads.

Hopefully, BDII will have a higher sustained decode rate to support those 5 modules. Cache performance remains a significant problem, but I doubt that AMD will be able to fix all three levels until BDIII, *if* they execute very well.

 

[Nemesis 1]

from the way things are looking trinity beats llano by 25% and it has none of the software fun stuff vishera gets it also doesnt get as much l1 or l3 cache and it gets min. 2/5ths the core modules of a vishera
as long as they dont goof on transistor count and use their chance at coding instructions properly then vishera will be a huge improvement over bulldozer across the board. expect its 8core piledriver to outpace the 8150 by easily 30-40% which puts the vishera 8 core well into i7SB/SB-E turf. and thats before you even discuss the 5 module vishera. so piledriver will be ebtter but will intel waste it with IB.. no doubt
i expect a similar environment in the piledriver/IB era that we saw in the am3 vs 1156/1136 era.

what happens after that well assuming amd doesnt botch it again that trend will continue indefinitely

Please Link to trinity beating llamp by 25% . If its AMD giving out these numbers 25% would be the highest gain . NOt average . Intel shows IVB being UPto 35% faster in cpu performance . But Anand says only 5%. I have seen with these metrics AMD is always hyped higher and intel always lower . To this day many still say SB is only 10% faster than th generation befor . When most know its more like 25% faster

 

[Nemesis 1]

Exactly! I think the most likely reason that Trinity is performing above expectations is that GF is getting their act together and hitting higher clocks. Though, the PD arch may be better suited to higher clocks as well.

I'm not surprised that Vishera will consist of 5 modules since that is AMD's main advantage over Intel at the moment in certain heavily threaded server workloads.

Hopefully, BDII will have a higher sustained decode rate to support those 5 modules. Cache performance remains a significant problem, but I doubt that AMD will be able to fix all three levels until BDIII, *if* they execute very well.

BS . If AMD had an advantage here . They would be gaining server market share . But their not . AMD is still losing server market share period.