• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Bulldozer Finally!

acme15 said:
8 cores at 4.2ghz with 8mb of L2 and 8mb of L3 cache, and yet it still scores significantly lower that the 3.7ghz 2500k in every benchmark, and yet the bulldozer costs more?!

Are AMD still under living under a rock? why release a cpu which is supposed to rival the 2600k when it is falling short of the 2500k by miles?? :confused:
Click to expand...

it was never going to compete with Sandy in single threaded applications, and the multi-threaded applications are sketchy at best, it does really well in applications that make good use of its architecture and does badly in situations where that same architecture hamper performance. the biggest problems are sure each module is pretty effective use of space, two 'cores' and shared floating-point unit in similar sort of space to previous cores, but then they go an attach an acre of cache to the die and make it truly monstrously enormous!

the second and biggest problem is the insane (obvious) amount of leakage occurring in the design, probably caused by the problems the guys over at Global Foundries are having that are hampering the power efficiency and intended frequencies, if Global sort their process out and Bulldozer gets to the frequencies it was always intended to run at (at stock) then it should be much much more competitive against the older K10.5 architecture, and let these stock speeds run at lower voltages with less heat output so its not all doom and gloom. also newer applications that take advantage of the design you see Bulldozer getting some pretty good results, some occasions beating everything except the ultra elite Extreme Editions and such, and gaming benchmarks are pretty mixed at times, again in the exact way you'd expect for the design, would love to see an FX-8150 against a 2600K in Supreme Commander as an example, cause that game loves cores (got quite superior performance switching to 1055T from Q6600 in that game in particular!).

the thing is, if they had access to more fabrication lines and such they would keep K10.5+ alive because a lot of the time it performs better than Bulldozer in gaming, and having the Bulldozer processors around for media based, heavily multi-threaded environments, let it mature and as applications get more threaded performance should come toward the architecture, like Battlefield 3 for an example should do well on the FX series of processors.

think they need to look at why they gave it so much cache though, it has a massive amount of cache which is taking up far far too much die space, does each modules need 2MB each? does the whole thing need 8MB L3? :confused:
 
Gashman said:
think they need to look at why they gave it so much cache though, it has a massive amount of cache which is taking up far far too much die space, does each modules need 2MB each? does the whole thing need 8MB L3? :confused:
Click to expand...

The large amount of cache is for the server market- as is the whole design tbh. There are large chucks of the uncore that are totally unused in desktop chips that will be used in servers.
Also I would assume that cache is very easy to make. Years before a process is ready for CPU production wafers of SRAM are made on it to learn about the process. It's basically the first building block.
 
Owenb said:
The large amount of cache is for the server market- as is the whole design tbh. There are large chucks of the uncore that are totally unused in desktop chips that will be used in servers.
Also I would assume that cache is very easy to make. Years before a process is ready for CPU production wafers of SRAM are made on it to learn about the process. It's basically the first building block.
Click to expand...

yeah that is a great point there about the server market, but still cache = transistors and Bulldozer was a transistor heavy design from the off, just seems like a weird thing to do for a desktop processor. suppose it stops them having to have multiple lines for the same sort of processor, but decreasing the cache (assuming it had no negative performance impact) would give them better yields surely, also reduce the overall power consumption of the chip? :confused:
 
acme15 said:
8 cores at 4.2ghz with 8mb of L2 and 8mb of L3 cache, and yet it still scores significantly lower that the 3.7ghz 2500k in every benchmark, and yet the bulldozer costs more?!

Are AMD still under living under a rock? why release a cpu which is supposed to rival the 2600k when it is falling short of the 2500k by miles?? :confused:
Click to expand...

It doesn't lose to 2500K in every benchmark.
 
kitch9 said:
It doesn't lose to 2500K in every benchmark.
Click to expand...

it does in gaming I think, though we all know why that is! the thing is in other benchmarks it does pretty well, it does very well in the ones it should be doing well in and quite poorly in the single-threaded ones, not sure why everyone is so surprised though, considering each integer core only has about 66.6% the resources of a single K10.5 core, the method of keeping the IPC decrease from being obvious was to make pre-fetching and branch prediction more efficient so it doesn't waste time and significantly ramp up the clock speeds, the latter is being hindered by Global's dodgy 32NM process at the moment.

would expect Bulldozer to monster games like Supreme Commander, Battlefield 3 and so on, anything that is multi-threaded, but it will do badly in things like Super-PI and games that don't make use of its architecture. the annoying thing is everyone seems so surprised by the results, knowing full well what Bulldozer is designed to do and what it sacrifices in order to be very strong in that respect, the only way it'll compete in single core performance is if the clock speeds go through the roof, with the architecture it is supposed to be possible mind.

but then lets go back to Phenom II against the earlier i5 and i7 processors, nobody gives it a second thought but its pretty competitive, in benchmarks the 920 and the 1100T trade blows quite nicely, the 920 wins some and the 1100T wins others, the world is just becoming far too pro-Intel at the moment! :p
 
Gashman said:
would expect Bulldozer to monster games like Supreme Commander, Battlefield 3 and so on, anything that is multi-threaded
Click to expand...

The trouble is best case scenario it will match or slightly beat an equivalently priced 2600K which kicks it silly everywhere else.

If BD was 25-50% faster in heavily multithreaded apps I could understand the architecture... but as it stands it's like Intel taking a 2600K, halving the resources inside each core and then adding double the cores just so that they can scream about having 8 cores - single threaded performance plummets without benefitting multithreaded performance at all.
 
the whole reason they removed a lot of the hardware from each integer core was because the conclusion was came to that a lot of the time it is sat waiting for other parts of the processor and so on, so the whole thing was streamlined to increase the overall efficiency of the design, K10.5 for example rivals Sandy Bridge in terms of overall hardware, just Intel have much better scheduling and branch prediction, which makes all the difference, hell on paper K10.5 from a hardware point of view could be very competitive if AMD sat down and sorted out its few shortcomings.

look at it from another point of view, reviews so far, we have a wide variety of numbers, hell a lot of the time they differ from site to site, we have reports of better performance on different motherboards, most of the reviews I have read have the CHV board as their testing board, which is the apparently slower or bugged one, so I am of two minds at the moment.
 
Gandeh said:
Not me, I chickened out at the last minute, went 1100T
Click to expand...

Did the same, 1100T will pretty much hold its value, if you pick one up at 140 get 15 for the Deus Ex code you get a great chip for £125.

Then just wait for the next revision, which hopefully they will have started to work out some of the shortcomings!
 
OCUK has FX6 stock, no FX8 to be seen yet.
I'm surprised at how early they've managed to get FX6 stock :o

They're also not the only E-tailor to be receiving stock.

But no FX8 anywhere yet.
 
Last edited:
Martini1991 said:
OCUK has FX6 stock, no FX8 to be seen yet.
I'm surprised at how early they've managed to get FX6 stock :o

They're also not the only E-tailor to be receiving stock.

But no FX8 anywhere yet.
Click to expand...

I bet that tells the story about yields - they have a lot of dies with one module defective.

Not surprised they're struggling with a 2 billion transistor CPU at 32nm.
 
Martini1991 said:
OCUK has FX6 stock, no FX8 to be seen yet.
I'm surprised at how early they've managed to get FX6 stock :o

They're also not the only E-tailor to be receiving stock.

But no FX8 anywhere yet.
Click to expand...

TBH,it could be down to the fact that the FX8100 series is the most popular model followed by the FX4100 series. The FX6100 series seems to be no better than a Phenom II X6 ATM.
 
CAT-THE-FIFTH said:
TBH,it could be down to the fact that the FX8100 series is the most popular model followed by the FX4100 series. The FX6100 series seems to be no better than a Phenom II X6 ATM.
Click to expand...

Why would the FX4 be popular? It's still worse than the Phenom II X4. I bet a Phenom II X3/X2 could give it a good chase in gaming.
The FX6 is crap in comparison to Intel and current AMD CPU's out at that price point too.

They simply don't have the numbers to ship.

What's worse about BD CPU's, is they seem to be limited in what the CPU NB can achieve, you can take Phenom II's to 3GHZ on the CPU-NB fairly easy, my old 1055T did 3.2GHZ for example, that'd spank the FX6 all over.
 
You must log in or register to reply here.
Back
Top Bottom