AMD A10-5800K preview - iGPU side only

[Eeqmcsq]

AMD has allowed some desktop Trinity benchmarks to be released a week early, but only on the iGPU. For CPU benchmarks, we still have to wait.

http://www.anandtech.com/show/6332/amd-trinity-a10-5800k-a8-5600k-review-part-1
http://www.legitreviews.com/article/2043/1/
http://www.pcper.com/reviews/Processors/AMD-A10-5800K-Performance-Preview-Trinity-Desktop
http://www.tomshardware.com/reviews/trinity-gaming-performance,3304.html

 

[AtenRa]

Haswell has a tough job to do in Gaming and or GPGPU in order to compete not only with Trinity but with 28nm next gen Kaveri GCN graphics.

 

[Centauri]

Considering how well Llano's performance scaled when overclocked w/ fast memory, these initials have me very hopeful for Trinity after some tweaking...

 

[beginner99]

Ok, I'm going to admit if you play minecraft most of the time there is no better bang for the buck than the A10.

For me personally more troubling is that already now you have to run "modern" games (crysis and metro aren't that new..) on low resolution an settings. What about new games in 1 years time?

Assuming A10 will be $135, a pentium with a GT640 like GPU will be less than $50 more. And for $100 more you get something that blows it out of the water. I don't want to brag but lets be honest $100 is very little if you keep that PC for 2 years, that is about $4 per month. Compared to rent, food, health-care, going out it's just pretty negligible amount of money.

Or if you are at the bottom of the performance range it doesn't take much money to get something a lot better.

 

[Yuriman]

At first glace I'm inclined to believe that performance has a lot of room to improve with overclocked memory, but I'm uncertain if I could justify more expensive RAM paired with an APU.

EDIT: Regardless, a nice improvement on the same node. It's a shame it's probably so memory bandwidth limited.

 

[deathBOB]

The Tech Report refuses to take part: http://techreport.com/blog/23638/amd-attempts-to-shape-review-content-with-staged-release-of-info

Must be a stinker.

I don't really understand Trinity's target market. If I'm not gaming, Intel graphics are good enough. If I'm really gaming, Trinity graphics don't cut it. Does a significant middle market actually exist (outside of laptops where this makes more sense)?

 

[NostaSeronx]

At first glace I'm inclined to believe that performance has a lot of room to improve with overclocked memory, but I'm uncertain if I could justify more expensive RAM paired with an APU.

EDIT: Regardless, a nice improvement on the same node. It's a shame it's probably so memory bandwidth limited.

http://www.xbitlabs.com/articles/graphics/display/amd-trinity-graphics_8.html#sect0

1866 to 2133 seems to be the perfect price/performance range. Also, guys did you see that CPU-Z?!

1.472 and 1.44 vees jeez...Undervolters are going to be screaming with glee.

 

[bunnyfubbles]

At first glace I'm inclined to believe that performance has a lot of room to improve with overclocked memory, but I'm uncertain if I could justify more expensive RAM paired with an APU.

EDIT: Regardless, a nice improvement on the same node. It's a shame it's probably so memory bandwidth limited.

the Samsung 30nm "Green" DDR3 can hit 2133 pretty easily and is all but guaranteed to do at least 1866, all for a whopping ~$20-23 per 4GB

 

[Red Hawk]

Good so far. Trinity's GPU isn't really a "leap" from Llano. It's just replacing the VLIW5 GPU with a comparable VLIW4 GPU, and bringing the benefits of VLIW4 with it (and probably other architectural tweaks that were part of the Radeon HD 6000 series, such as improved tessellation). The next true leap will probably be the process shrink to 28 nm (will it be 28 nm? Or smaller?) and switching to GCN architecture. Best case scenario, they manage to fit an entire Cape Verde onto Kaveri.

The Tech Report refuses to take part: http://techreport.com/blog/23638/amd-attempts-to-shape-review-content-with-staged-release-of-info

Must be a stinker.

I don't really understand Trinity's target market. If I'm not gaming, Intel graphics are good enough. If I'm really gaming, Trinity graphics don't cut it. Does a significant middle market actually exist (outside of laptops where this makes more sense)?

It mostly makes sense in laptops (which is also why it makes sense that AMD launched this on laptops first). Desktop, it works for HTPC users and budget gamers really looking to save a buck.

 

[AtenRa]

Hell, even A8-5600K with 256 shaders is 50-60% faster than Core i7 3770K with HD4000 not to mention its very close to Llanos 400 shaders. I really cant wait for the full review in a few days.

 

[Red Hawk]

Yeah, even despite Intel's leaps forward with both HD Graphics 3000 and 4000, they really can't touch AMD. Haswell might surpass Trinity but AMD still has Kaveri waiting in the wings. And lol, Anand chose to use the laughable Geforce GT 640 as the "highest" dedicated graphics card to compare to. While it sometimes surges ahead of Trinity, others it merely ties Trinity or falls behind. Anand could have at least used the Radeon HD 7750, which actually has some use as a gaming card.

 

[SPBHM]

I'm inclined to believe that performance has a lot of room to improve with overclocked memory

and overclocked GPU,
it's seems it can do 20%+ OC with high end cooling and overvolt,

looking at the tests, is clear that is still way ahead of the HD4000 in most cases, but the improvement over llano is not really all that great, even more when you consider that llano runs at 600MHz (and can be OCed to 900+), and trinity at 800, BUT it's the same 32nm process, so I guess it's a good gain, but still, looks slower or close to a card you can buy for less than $50 (6570 DDR3),
and I find interesting that instead of using their own cards, they tried to compare to the renamed GT430/440 DDR3

about the power usage on Anand's review

was the 640 being tested on the A10, or on the Intel platform?

CPU load (prime95) is looking good.

 

[Soulkeeper]

The Tech Report refuses to take part: http://techreport.com/blog/23638/amd-attempts-to-shape-review-content-with-staged-release-of-info

Must be a stinker.

I don't really understand Trinity's target market. If I'm not gaming, Intel graphics are good enough. If I'm really gaming, Trinity graphics don't cut it. Does a significant middle market actually exist (outside of laptops where this makes more sense)?

I agree.
Not to mention tomshardware's preview some 3-4 months ago ... they've reposted it 2 or 3 times since. Apparently AMD got mad at them at first.

I see trinity as a sidegrade: the CPU performance will be uninspiring, gpu performance will be 10-15% higher in some things yet so will load power usage.
I'd go as far as saying long as they use ddr3 they'll be in the same performance class. Untill ddr4 comes along APUs are stuck in the same ballpark performance bracket.

As a llano owner I see absolutly no reason to buy into the FM2 party.

Unless they start making 4 module versions with L3 cache and no gpu.
combing the two strategies to a single socket is what i'd like to see.

 

[NostaSeronx]

Unless they start making 4 module versions with L3 cache and no gpu.
combing the two strategies to a single socket is what i'd like to see.

The L2 in the Bulldozer architecture acts like the L3. The L3 in Zambezi or Vishera acts like a glue for the the L2. This problem is fixed in Steamroller where there is no L3 on any version. Also, you won't see the Fusion platform and Server platform use the same socket till 2015.

 

[hans007]

The Tech Report refuses to take part: http://techreport.com/blog/23638/amd-attempts-to-shape-review-content-with-staged-release-of-info

Must be a stinker.

I don't really understand Trinity's target market. If I'm not gaming, Intel graphics are good enough. If I'm really gaming, Trinity graphics don't cut it. Does a significant middle market actually exist (outside of laptops where this makes more sense)?

Laptops are like 65% of the market now. I figure that alone justifies it

 

[anikhtos]

The apu from amd maybe is a good step for the future but so far it really does not have a market.
Liano apu was not a balanced chip too much gpu power for the cpu part.
And what amd is doing increasing the gpu part again!!!!!!
the apu is held back from the cpu part not the gpu part.

The idea was good. Build a chip system but thats bobcat teritory where amd is better than intel e350-e450 beign a bit better than the atoms cpus and with better igpu.
Amd cancelled the next generation of the bobcat :'( i was really waiting so i could build a cheap system for everyday use and browsing.

But liano-trinitz is a joke.
The memory of the system plays a huge impact on the perfomance.
It would have make sence if trinitz at least suported triple or quadro memory channel
trying to satisfy the gpu part that is hungry for bandwith.
But still limit to double channel. Official support for higher memorz would also helped the platform. But from monez aspect a quadro or triple memorz channel would boost perfomance with less money. Buying an extra module or 2 is cheaper than to buy higher frequensy ram.

But amd really needs to focus on the cpu part of the apus.
Improve that part.

In the future when we will not have graphic cores or cpu cores but plain cores and the software will use them accordinly will be another story.


Also the hyprid crossfire maz sound promising but is so restricted and you can not use a hight card with the igpu so pretty much uselles.
Changing socket because of the diferent interior architecture limits the future proof cost fo the platform. When in a discrete card you simple buy one and replace the old one. With the apu you can not do. New mobo new apu that adds up the cost for a plain gpu perfomance upgrade. Buying a gpu is always an option but then what is the point of the apu!?!?!?!?

And about necesity of socket changing.
How long socket A stayed alive?!?!?
Too long and was a selling point for amd buy amd cause you will upgrade only the cpu. Even very old chipsets supported manz never models of athlons.

Do not praise the milking from intel or amd with changing sockets as a necesity.

Trinity would make sence if it had a small price and could make a compact small form factor pc with less money. But an ivy pentium and a gpu will be at same price range and offer an upgretability for the user.

THERE IS NO MARKET FOR THIS PRODUCT
AMD GIVE US A 4 CORE BOBCAT AT 36 WATT
that would be a killer product.

 

[Hacp]

The apu from amd maybe is a good step for the future but so far it really does not have a market.
Liano apu was not a balanced chip too much gpu power for the cpu part.
And what amd is doing increasing the gpu part again!!!!!!
the apu is held back from the cpu part not the gpu part.

I don't think the graphics were limited by the CPU. The graphics on the APU aren't that blazing that the CPU would be the limiting factor in most benchmarks.

 

[NostaSeronx]

The iGPU is probably ROPs limited rather then CPU limited.

 

[SPBHM]

The iGPU is probably ROPs limited rather then CPU limited.

it would be interesting to see some results of fill rate tests and other theoretical results from 3dm!

but as far as I know this IGP has the equivalent to 8ROPs and 24TMUs (like a 6500/6600 GPU), so while I think the shared memory can be a slight problem, it's probably more limited simply by the amount of power it have (llano had 8/20)...

 

[NTMBK]

it would be interesting to see some results of fill rate tests and other theoretical results from 3dm!

but as far as I know this IGP has the equivalent to 8ROPs and 24TMUs (like a 6500/6600 GPU), so while I think the shared memory can be a slight problem, it's probably more limited simply by the amount of power it have (llano had 8/20)...

Anandtech have some synthetics results:

Does seem to be indicating that fillrate is a bottleneck here.

 

[SPBHM]

how did I miss that? thanks; it looks just like the 5570 (650MHz 8 ROPs, 1600 or 1800?MHz 128bit DDR3)

 

[inf64]

It kinda sucks they opted to limit the early previews to GPU tests only. There is actually no pint in that since we already know the CPU perf. as THG did the complete review some months ago... On the other hand , new iGPU performs very nicely,even the cut down 256SP part is able to outperform top Llano. Also it seems that iGPU scales even better than Llano in terms of both CPU and memory clock so there is huge potential performance in there.

Too bad AT didn't test the dual GPU combo with hybrid CF.

 

[SPBHM]

It kinda sucks they opted to limit the early previews to GPU tests only. There is actually no pint in that since we already know the CPU perf. as THG did the complete review some months ago... On the other hand , new iGPU performs very nicely,even the cut down 256SP part is able to outperform top Llano. Also it seems that iGPU scales even better than Llano in terms of both CPU and memory clock so there is huge potential performance in there.

Too bad AT didn't test the dual GPU combo with hybrid CF.

http://www.techpowerup.com/reviews/AMD/FM2_APU_Preview/5.html

good in 3dm11 (5770 level I think), broken in games it seems...

for mobile the best I can find
http://www.notebookcheck.net/AMD-Radeon-HD-7660G-HD-7670M-Dual-Graphics.81173.0.html

 

[bunnyfubbles]

Yeah, even despite Intel's leaps forward with both HD Graphics 3000 and 4000, they really can't touch AMD. Haswell might surpass Trinity but AMD still has Kaveri waiting in the wings. And lol, Anand chose to use the laughable Geforce GT 640 as the "highest" dedicated graphics card to compare to. While it sometimes surges ahead of Trinity, others it merely ties Trinity or falls behind. Anand could have at least used the Radeon HD 7750, which actually has some use as a gaming card.

on the contrary, I liked seeing the "laughable" GT 640 as there are several benchmarks where it absolutely destroys all the iGPUs, showing that APUs are still light-years behind discrete GPUs; never did I get the idea that Trinity is even remotely close to replacing even the lowest of discrete video cards outside of niche scenarios

and on that note, you might want to look back at those results, it seems like you were seeing things you wanted to see rather than what was actually there; there's only one instance where the 640 loses to an iGPU, and its in OpenGL. In the instances where we could even call it a "tie" the 640 still had the highest results or 'ties' with the 5570. Of which there were 4 such 'tying' results; Shogun 2 and Skyrim (but only at 1366x768 - likely CPU bottleneck), and twice with Civ V (a game where the results are probably CPU limited regardless)

There's basically no need to throw in any more powerful of a GPU when its clear that Trinity is still well behind, as much improved as it may be. There's also a good chance that the CPU is enough of a bottleneck that running a beefier GPU might have had very diminishing returns and unflattering results when the clear goal here is to show just how far ahead AMD is when it comes to iGPU performance; I think its quite clear they're not trying to compete with discrete GPU just yet...

 

[Edgemeal]

I thought these previews were to only show GPU performance?
How Xbit get away with SYSmark2012 bench?

http://www.xbitlabs.com/articles/graphics/display/amd-trinity-graphics_5.html