Tuesday, February 24th 2015

AMD's Excavator Core is Leaner, Faster, Greener

AMD gave us a technical preview of its next-generation "Carrizo" APU, which is perhaps the company's biggest design leap since "Trinity." Built on the 28 nm silicon fab process, this chip offers big energy-efficiency gains over the current-generation "Kaveri" silicon, thanks to some major under-the-hood changes.

The biggest of these is the "Excavator" CPU module. 23 percent smaller in area than "Steamroller," (same 28 nm process), Excavator features a new high-density library design, which reduces die-area of the module. Most components are compacted. The floating-point scheduler is 38% smaller, fused multiply-accumulate (FMAC) units compacted by 35%, and instruction-cache controller compacted by another 35%. The "Carrizo" silicon itself uses GPU-optimized high-density metal stack, which helps with the compaction. Each "Excavator" module features two x86-64 CPU cores, which are structured much in the same way as AMD's previous three CPU core generations.
The compaction in components doesn't necessarily translate into lower transistor-counts over "Kaveri." In fact, Carrizo features 3.1 billion transistors (Haswell-D has 1.4 billion). Other bare-metal energy optimizations include an 18% leakage reduction over previous generation, using faster RVT components, which enables 10% higher clock speeds at the same power-draw (as "Kaveri"). A new adaptive-voltage algorithm reduces CPU power draw by 19%, and iGPU power draw by 10%. AMD introduced a few new low-power states optimized for mobile devices such as >9-inch tablets and ultra-compact notebooks, which reduces overall package power draw to less than 1.5W when idling, and a little over 10W when active. In all, AMD is promising a conservative 5% IPC uplift for Excavator over Steamroller, but at a staggering 40% less power, and 23% less die-area.
The integrated-GPU is newer than the one featured on "Kaveri." It features 8 compute units (512 stream processors) based on Graphics CoreNext 1.3 architecture, with Mantle and DirectX 12 API support, and H.265 hardware-acceleration, with more than 3.5 times the video transcoding performance increase over "Kaveri." For notebook and tablet users, AMD is promising "double-digit percentage" improvements in battery life.

Now, if only AMD can put six of these leaner, faster, and greener "Excavator" modules onto an AM3+ chip.
Add your own comment

85 Comments on AMD's Excavator Core is Leaner, Faster, Greener

#26
TRWOV
I can't believe I'm actually agreeing with Jorge but yes, last two nodes haven't got Intel much performance. Even if IPC has increased maximum attainable clock speeds have decreased too so it evens out although that might have more to do with the change from solder to TIM on Intel's part rather than the node itself.

Anyone having a Core 3xxx doesn't have much reason to upgrade to Core 4xxx or 5xxx.
Posted on Reply
#27
Jorge
AquinusIntel and AMD measure node sizes differently. They're actually not as dissimilar as you think they are, but that isn't to say that Intel doesn't have a smaller process than AMD, it's just over exaggerated because of how it's measured. I wish I could find the article, but IIRC it describes how Intel measures more of the width of any given circuit where AMD excludes the very outside edge where Intel includes it, or something along those lines. So in reality, something like 16nm for Intel would be something like 20nm by how AMD measures it. It's important to keep that in mind and if I end up finding the article, I'll add it.
...and as documented by the past three Intel node reductions, no tangible performance is gained with die shrinks below ~32Nm. The die shrink is primarily for lower power consumption, increased transistor density and lower unit costs. That's why AMD is in no hurry to rush to smaller node sizes.

AMD is doing a very good job on their APU designs. Carrizo should make a lot of people very happy. AMD is scheduled to officially release these chippies in Q2 of '15. They are actually in production now.

For those not up top speed on die shrink sizes, there is almost zero performance gain now that we are below 32Nm. The biggest gains in die shrink are lower power and higher transistor density, which reduce CPU/APU costs. With AMD's advanced power management on Carrizo, power consumption is reduced even further than in prior low power APU models.

Carrizo APUs will be available in both mobile and desktop versions with mobile models being released first.
Posted on Reply
#28
Exceededgoku
JorgeAMD is doing a very good job on their APU designs. Carrizo should make a lot of people very happy. AMD is scheduled to officially release these chippies in Q2 of '15. They are actually in production now.

For those not up top speed on die shrink sizes, there is almost zero performance gain now that we are below 32Nm. The biggest gains in die shrink are lower power and higher transistor density, which reduce CPU/APU costs. With AMD's advanced power management on Carrizo, power consumption is reduced even further than in prior low power APU models.

Carrizo APUs will be available in both mobile and desktop versions with mobile models being released first.
Yes of course power requirements are lower, which can (not necessarily does) allow for greater performance.

The fact that Intel hasn't improved much is down to market complacency and general optimisations being prioritised.

I'd love to see AMD release something even half as good as some of the latest Intel CPU's. Until these are released and in consumer hands I won't be making any judgements, as the last time they did this they released the original FX that ended up being a turd (although over it's lifetime it's improved somewhat...).

This is of course ignoring the price differences (which AMD has aced)!
Posted on Reply
#29
Ralfies
I would switch to AMD if they offered even a 4 module excavator CPU on AM3+.
Posted on Reply
#30
Sony Xperia S
ExceededgokuYes of course power requirements are lower, which can (not necessarily does) allow for greater performance.

The fact that Intel hasn't improved much is down to market complacency and general optimisations being prioritised.
If Intel sit back, relax and wait for AMD, that might turn to a very severe mistake which could cost them a lot.

Those recent launches from them for some reason didn't show anything than negligible improvements both in power consumption and performance, so I guess they don't even use the process nodes' full potential which is a shame.
Posted on Reply
#31
newtekie1
Semi-Retired Folder
JorgeUnfortunately AMD dropped the ball and should have released an Excavator based AM3+, but they are NOT going to do this. Zen will be the upgrade for AM3+ users but it will require a new socket/mobo. While Zen will be a big performance jump from the current FX Vishera series, it's taken YEARS too long for AMD to deliver this performance upgrade. AMD should have definitely offered an Excavator cored AM3+ late in 2014 so AM3+ users had a viable upgrade path.
The northbridge on the motherboard has become the huge issue with AM3+. These chips are designed to have the nothbridge directly attached to them. Trying to modify them to work with an already existing outdated northbridge would be a lot of work. Plus, AM3+ users would still be stuck with PCI-E 2.0. It really is time to let AM3+ go. What we really need is a new high end socket and some high end versions of these APUs. I know they said they didn't want to release AM4 until DDR4 was mainstream, but I'd really like to see them do it sooner rather than later. Put Excavator on AM4, give it 8-cores and a weaker GPU(because most people will use a dedicated GPU anyway so power efficiency is more important) and let it use DDR3. When DDR4 is mainstream move to AM4+.
Posted on Reply
#32
Dave65
TRWOVAMD themselves stated that Piledriver was the end of the line for AM3+

I think the next big CPU from AMD will feature a new socket and new architecture. Maybe Zen?
As it should be, AM3+ is ancient!
Posted on Reply
#33
Dustproblem
A really nice gaming-capable mobile/ultrabook chip.
I really need a new notebook badly; Web browsing, twitch streaming, Youtube and gaming are top priorities for me.
My current laptop gets so sluggish running flash packed websites and source quality videos on twitch.
It can't even run 1080p 60fps youtube videos, without crashing the browser and the extensions running on Chrome.
Hope these ones come in equipped with SSDs, unlike their predecessors Kaveri laptops.
Posted on Reply
#34
Sony Xperia S
newtekie1The northbridge on the motherboard has become the huge issue with AM3+. These chips are designed to have the nothbridge directly attached to them. Trying to modify them to work with an already existing outdated northbridge would be a lot of work. Plus, AM3+ users would still be stuck with PCI-E 2.0. It really is time to let AM3+ go. What we really need is a new high end socket and some high end versions of these APUs. I know they said they didn't want to release AM4 until DDR4 was mainstream, but I'd really like to see them do it sooner rather than later. Put Excavator on AM4, give it 8-cores and a weaker GPU(because most people will use a dedicated GPU anyway so power efficiency is more important) and let it use DDR3. When DDR4 is mainstream move to AM4+.
In 2016-2017 DDR4 should be the way-to-go for a system designed to last at least couple of years.
With Zen AMD needs to go directly to DDR4 and forget about DDR3.
Posted on Reply
#35
BiggieShady
Sony Xperia SThis might be true only in a limited area of the frequency / power curve....
Power consumption, yes, I also suspect that decrease is their best case scenario. With IPC, if they are telling the truth, 5% is across all frequencies.
Posted on Reply
#36
theGryphon
AMD is following the right strategy by going for the mobile market. That's where the volume (of sales) is, and that's where the money can come from. The money they desperately need to fund further R&D. Desktop market (especially high-end) is way smaller than the excitement of these tech forums, filled with enthusiast system builders (including myself), indicate.

AM3+ is now dead. AMD should come up with a new socket when the new node is ready for mass high performance production, and they can deliver desktop/server CPUs with many cores with at least somewhat competitive performance. That may not be possible without the new node, so releasing a DOA desktop platform too soon would be a terrible mistake, putting the nail in the coffin.
Posted on Reply
#37
Assimilator
comperiusNo, I read Q3 2015 is out of the picture and the CPU's wil arrive in late Q4 2015 or Q1 2016. That's why I wrote this info. You can check this info online via google search.
www.guru3d.com/news-story/intel-14nm-skylake-desktop-cpus-delayed-to-august.html
Jorge...and as documented by the past three Intel node reductions, no tangible performance is gained with die shrinks below ~32Nm. The die shrink is primarily for lower power consumption, increased transistor density and lower unit costs. That's why AMD is in no hurry to rush to smaller node sizes.
Considering the appalling power consumption of AMD's CPUs, perhaps they should be in a bit more of a hurry.
Posted on Reply
#38
$ReaPeR$
good news :) its time for some competition on the CPU market. it has been stagnant for too long..
Posted on Reply
#39
ZoneDymo
DustproblemA really nice gaming-capable mobile/ultrabook chip.
I really need a new notebook badly; Web browsing, twitch streaming, Youtube and gaming are top priorities for me.
My current laptop gets so sluggish running flash packed websites and source quality videos on twitch.
It can't even run 1080p 60fps youtube videos, without crashing the browser and the extensions running on Chrome.
Hope these ones come in equipped with SSDs, unlike their predecessors Kaveri laptops.
Well its not like 1080p 60fps is easy peasy
Posted on Reply
#40
arbiter
TheinsanegamerNYeah, you have to replace the socket (and chipset) eventually, but AMD didn't do that. They killed AM3+ without offering a newer socket, or offering 6-8 core chips for FM2+. This is after they tried to pump up everyone on the MOAR CORES system. If I have to buy a quad core, the i5 smears the a10 across the walls. I'm still waiting for an upgrade from the fx6300.....
Problem that is with doing that on FM2+, is it has GPU on it as well which takes up a ton of space. Probably could make one without it and have it work but probably be confusing for some people. Intel on their high end 6 and 8 core parts don't have gpu cause would use to much space.
Posted on Reply
#42
ThE_MaD_ShOt
It would be great if they did what they did with Am2+/Am3 cpu's and have a memory controller on the cpu that could use ddr3 or ddr4. That way to can release something now and it use ddr3 and when ddr4 becomes mainstream release an updated mobo that is ddr4.
Posted on Reply
#43
Aquinus
Resident Wat-man
I personally would like AMD to release a new socket that's LGA with the PCI-E root complex moved to the CPU and with more PCI-E lanes like 32 or 36. Enough to drive 4x 8-lane ports.
newtekie1The northbridge on the motherboard has become the huge issue with AM3+. These chips are designed to have the nothbridge directly attached to them. Trying to modify them to work with an already existing outdated northbridge would be a lot of work. Plus, AM3+ users would still be stuck with PCI-E 2.0. It really is time to let AM3+ go. What we really need is a new high end socket and some high end versions of these APUs. I know they said they didn't want to release AM4 until DDR4 was mainstream, but I'd really like to see them do it sooner rather than later. Put Excavator on AM4, give it 8-cores and a weaker GPU(because most people will use a dedicated GPU anyway so power efficiency is more important) and let it use DDR3. When DDR4 is mainstream move to AM4+.
So what you're really saying is that the PCI-E root complex should be moved to the CPU and the interconnect between the south bridge would use what, A-Link/PCI-E like it has in the past?

The funny thing about all of this is that it's exactly how APUs work and how Intel CPUs work. AM3+ is just a dying platform because of the archaic design of having the PCI-E root complex on the motherboard (which is awesome for cheap servers I might add. It's half of the reason my gateway still has the Phenom II and a 790FX in it, PCI-E galore! I could add 8 ethernet ports and still have PCI-E to spare.)
Posted on Reply
#44
librin.so.1
I would have gotten an AMD laptop with the older chips. And I would get a laptop with these new chips.
Now, if only I could BUY a laptop with an AMD APU in these parts, that'd be great.
Me and quite a few people from around here I know are looking for [decent] laptops with AMD APUs. But it's Intels as far as the eye can see.
"There's a demand. But zero supply for the last few years."
ARG.
Posted on Reply
#45
arbiter
ThE_MaD_ShOtIt would be great if they did what they did with Am2+/Am3 cpu's and have a memory controller on the cpu that could use ddr3 or ddr4. That way to can release something now and it use ddr3 and when ddr4 becomes mainstream release an updated mobo that is ddr4.
Sounds good in theory but problem is its 240pin vs 288pin ram slots, so need 2 separate ram slots which limits the ram can install. On top of that space on the cpu for the 2 diff memory controllers that would each need their own pins and traces in the board. It makes things more complex on cpu pin-out and board traces. Not really worth it to make one. The memory controlers are on cpu for both intel and amd.
Posted on Reply
#46
newtekie1
Semi-Retired Folder
AquinusSo what you're really saying is that the PCI-E root complex should be moved to the CPU and the interconnect between the south bridge would use what, A-Link/PCI-E like it has in the past?

The funny thing about all of this is that it's exactly how APUs work and how Intel CPUs work. AM3+ is just a dying platform because of the archaic design of having the PCI-E root complex on the motherboard (which is awesome for cheap servers I might add. It's half of the reason my gateway still has the Phenom II and a 790FX in it, PCI-E galore! I could add 8 ethernet ports and still have PCI-E to spare.)
Yes, that is what I'm saying. And yes, I would like to see 32 PCI-E lanes from the CPU and 16 from the PCH(southbridge).
arbiterSounds good in theory but problem is its 240pin vs 288pin ram slots, so need 2 separate ram slots which limits the ram can install. On top of that space on the cpu for the 2 diff memory controllers that would each need their own pins and traces in the board. It makes things more complex on cpu pin-out and board traces. Not really worth it to make one. The memory controlers are on cpu for both intel and amd.
That isn't what he means. He doesn't want a single motherboard with both types of RAM. He just wants the CPU to support both types. So you can use a motherboard with DDR3 at first, and then get a new motherboard with DDR4 in the future when DDR4 becomes mainstream.
Posted on Reply
#47
Schmuckley
Diagrams don't mean a thing until it's in my grubby little hand :P
Posted on Reply
#48
ThE_MaD_ShOt
arbiterSounds good in theory but problem is its 240pin vs 288pin ram slots, so need 2 separate ram slots which limits the ram can install. On top of that space on the cpu for the 2 diff memory controllers that would each need their own pins and traces in the board. It makes things more complex on cpu pin-out and board traces. Not really worth it to make one. The memory controlers are on cpu for both intel and amd.
Yes it would be two different mobos a ddr3 board for now then a ddr4 board when ddr4 becomes mainstream. The boards don't need both kinds of ram slots. Just the cpu be able to run on a board with either. Also Amd has already done this with cpus using 2 different memory specs, the Phenom II cpus could be used in boards that either where ddr2 or ddr3.
Posted on Reply
#49
BiggieShady
AquinusIntel and AMD measure node sizes differently. They're actually not as dissimilar as you think they are, but that isn't to say that Intel doesn't have a smaller process than AMD, it's just over exaggerated because of how it's measured. I wish I could find the article, but IIRC it describes how Intel measures more of the width of any given circuit where AMD excludes the very outside edge where Intel includes it, or something along those lines. So in reality, something like 16nm for Intel would be something like 20nm by how AMD measures it. It's important to keep that in mind and if I end up finding the article, I'll add it.
Quoted for truth.
16 nm node doesn't have transistors that are 2 times smaller than at 32 nm node. FinFET Transistors at 16 nm node just have thinner fins so marketing comes to play and starts measuring the fin width and BAM new silicon node is here ... where in fact there's nothing new except tapered fins.
Main optimizations these days come from figuring out how to arrange the transistors in order for them to be able to share a fin or two.
Posted on Reply
#50
ThE_MaD_ShOt
cheesy999That's because the ddr2 controller was on the motherboard and the ddr3 one was on the cpu, they could presumably do something similar again though...
Nah the controllers where on the cpu this is from Amd's web site.

Posted on Reply
Add your own comment
Mar 15th, 2025 19:52 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts