IP Performance has certainly increased but the FP is still down a bit ... the memory performance is still slower (if i read that correctly) too.
On the basis of past performance these benchies seem realistic?
Discuss (and thanks Juan) guys?
The L1 Icache looks off, 96KB is a very odd size to have as it tends to be evenly divisible by powers of 2. After the fiasco that was BD I'm wary of anything posted as "engineering sample".
I've written a Java app to compare CPUs and GPUs, and I have some results for a stock i7-2600K and an 8350 overclocked to 4.4GHz. When using Java, it is a pretty convincing win for the 8350. I conclude that the Java just-in-time compiler was not written by Intel!
My results would be improved by more data points, I do admit. You can always go to the download page on the site and contribute some more data if you like Be aware that it is a work-in-progress and has bugs (e.g. crashes on GTX 770). Also, you have to have Java installed.
Well this 2600k seam to be holding it own against that same 8350!
Thanks for submitting the result! Since 8350rocks submitted the result, I've tweaked the app to squeeze a bit more performance from the CPUs. A fairer comparison would be Palladin's 8350 which is overclocked higher and tested with a more recent version of the software:
As you might expect, the 8350 and 2600k are neck and neck on integer performance but the 2600k pulls ahead when we look at floating point and really dominates on memory (the memory test is cache intensive).
Ags1 :
I'm more interested in what the CPUs do at stock, but so far all the 8350 results have been OC'd. I was simply assuming your 2600k was overclcked too - it isn't? I had no intention of being unfair...
Ags1 :
I do have a comparison of i5-3570 (at stock reportedly) to 8350 (OC'd):
I've written a Java app to compare CPUs and GPUs, and I have some results for a stock i7-2600K and an 8350 overclocked to 4.4GHz. When using Java, it is a pretty convincing win for the 8350. I conclude that the Java just-in-time compiler was not written by Intel!
My results would be improved by more data points, I do admit. You can always go to the download page on the site and contribute some more data if you like Be aware that it is a work-in-progress and has bugs (e.g. crashes on GTX 770). Also, you have to have Java installed.
Well this 2600k seam to be holding it own against that same 8350!
Thanks for submitting the result! Since 8350rocks submitted the result, I've tweaked the app to squeeze a bit more performance from the CPUs. A fairer comparison would be Palladin's 8350 which is overclocked higher and tested with a more recent version of the software:
As you might expect, the 8350 and 2600k are neck and neck on integer performance but the 2600k pulls ahead when we look at floating point and really dominates on memory (the memory test is cache intensive).
Ags1 :
I'm more interested in what the CPUs do at stock, but so far all the 8350 results have been OC'd. I was simply assuming your 2600k was overclcked too - it isn't? I had no intention of being unfair...
Ags1 :
I do have a comparison of i5-3570 (at stock reportedly) to 8350 (OC'd):
I'm not arguing with you, I think everyone agrees the i7's will beat an 8350. But n some metrics at least they are pretty close.
It's not an argument I'm after , you used the words convincing and fair while being totally unfair,
if both CPU were running at the same clock speeds and the 8350 won that would be fair, and
I'm still wondering what was convincing to you, that if you leave the 2600k at stock clock an 8350
overclocked can beat it?
Well we would expect the $330 USD CPU to be better then the $199 USD CPU. Though honestly the real value is the fx6350 at $139 USD. AMD is the value brand, I never expect them to the absolute best at anything, I only expect them to be very cheap for what they do provide. This is why I compare the fx8x series with the i5 as they are priced to compete with them at that level and the fx6x with the i3. The fx4300 is kinda misplaced as it doesn't really compete well with the lower pentiums and you'd most likely be better off going with APU.
Well to be fair when the 8350 was released it sold for 269.00, and the 2600k could be
purchased for 309.00 that's about a 40.00 dollar difference between the two CPU's.
We also expect the 8 core 5GHz overclocker to be able to handle almost 2 year old 4 core with
HT 2600k with no problem.
Now let's look at the 9590 for 899.00 when released should we expect more performance from this CPU?,
or the same cpu at 399.99 at it's current selling price?
I've written a Java app to compare CPUs and GPUs, and I have some results for a stock i7-2600K and an 8350 overclocked to 4.4GHz. When using Java, it is a pretty convincing win for the 8350. I conclude that the Java just-in-time compiler was not written by Intel!
My results would be improved by more data points, I do admit. You can always go to the download page on the site and contribute some more data if you like Be aware that it is a work-in-progress and has bugs (e.g. crashes on GTX 770). Also, you have to have Java installed.
Well this 2600k seam to be holding it own against that same 8350!
Thanks for submitting the result! Since 8350rocks submitted the result, I've tweaked the app to squeeze a bit more performance from the CPUs. A fairer comparison would be Palladin's 8350 which is overclocked higher and tested with a more recent version of the software:
As you might expect, the 8350 and 2600k are neck and neck on integer performance but the 2600k pulls ahead when we look at floating point and really dominates on memory (the memory test is cache intensive).
Ags1 :
I'm more interested in what the CPUs do at stock, but so far all the 8350 results have been OC'd. I was simply assuming your 2600k was overclcked too - it isn't? I had no intention of being unfair...
Ags1 :
I do have a comparison of i5-3570 (at stock reportedly) to 8350 (OC'd):
I'm not arguing with you, I think everyone agrees the i7's will beat an 8350. But n some metrics at least they are pretty close.
It's not an argument I'm after , you used the words convincing and fair while being totally unfair,
if both CPU were running at the same clock speeds and the 8350 won that would be fair, and
I'm still wondering what was convincing to you, that if you leave the 2600k at stock clock an 8350
overclocked can beat it?
Well we would expect the $330 USD CPU to be better then the $199 USD CPU. Though honestly the real value is the fx6350 at $139 USD. AMD is the value brand, I never expect them to the absolute best at anything, I only expect them to be very cheap for what they do provide. This is why I compare the fx8x series with the i5 as they are priced to compete with them at that level and the fx6x with the i3. The fx4300 is kinda misplaced as it doesn't really compete well with the lower pentiums and you'd most likely be better off going with APU.
Well to be fair when the 8350 was released it sold for 269.00, and the 2600k could be
purchased for 309.00 that's about a 40.00 dollar difference between the two CPU's.
We also expect the 8 core 5GHz overclocker to be able to handle almost 2 year old 4 core with
HT 2600k with no problem.
Now let's look at the 9590 for 899.00 when released should we expect more performance from this CPU?,
or the same cpu at 399.99 at it's current selling price?
jed :
Ags1 :
jed :
Ags1 :
jed :
Ags1 :
It's hard to trust benchmarks, so I wrote my own
I've written a Java app to compare CPUs and GPUs, and I have some results for a stock i7-2600K and an 8350 overclocked to 4.4GHz. When using Java, it is a pretty convincing win for the 8350. I conclude that the Java just-in-time compiler was not written by Intel!
My results would be improved by more data points, I do admit. You can always go to the download page on the site and contribute some more data if you like Be aware that it is a work-in-progress and has bugs (e.g. crashes on GTX 770). Also, you have to have Java installed.
Well this 2600k seam to be holding it own against that same 8350!
Thanks for submitting the result! Since 8350rocks submitted the result, I've tweaked the app to squeeze a bit more performance from the CPUs. A fairer comparison would be Palladin's 8350 which is overclocked higher and tested with a more recent version of the software:
As you might expect, the 8350 and 2600k are neck and neck on integer performance but the 2600k pulls ahead when we look at floating point and really dominates on memory (the memory test is cache intensive).
Ags1 :
I'm more interested in what the CPUs do at stock, but so far all the 8350 results have been OC'd. I was simply assuming your 2600k was overclcked too - it isn't? I had no intention of being unfair...
Ags1 :
I do have a comparison of i5-3570 (at stock reportedly) to 8350 (OC'd):
You also have my 3770k running @ 4.7GHz daily against the 8350 @ 4.7 you think.
http://www.headline-benchmark.com/results/45e8da85-12a3-44d3-b6f0-11b9eccb4525/9a411141-526d-4452-8e78-1c9891c2efa0
You could get the 8350 for slightly over $200 USD when it was released, it was quickly set to $200 USD. Your thinking the original FX8150 that was horrible overpriced for it's performance. The 3770K is priced at $339 USD right now which is about what the 2600K was going for during that time span. There was a big debate on whether purchasing the i7-2600K would of given a meaningful performance difference vs the much cheaper i5-2500K which is what the FX8 was slotted to compete with.
You don't compare clock vs clock you compare what you get for your money. That's why I tell people to stay away from the fx9 as it's just a golden binned fx8350 factory OC'd and a blatant cash grab.
Also the BD/PD uArch will never be able to perform the same as the SB/IB/HW uArch at the same clock. BD/PD has 2 ALU's per core for 16 ALU's per chip with four FPU's capable of 2x instructions each. The SB/IB/HW is 3 ALU's per core for 12 per chip with four FPU's capable of doing 3x instructions each. The BD/PD design is largely held back by it's poor L2/L3 cache performance with a hit being taken from the front end decoder. That's the price they had to pay for shoving 16 ALU's on a chip. Being at 32nm doesn't help either, that's a big reason why 4.8 is a reasonable clock ceiling.
BTW I've had mine to 5.0ghz but I needed to run the cooling system too high and I hate fan noise. My entire rig is designed to be quiet.
I'm not arguing with you, I think everyone agrees the i7's will beat an 8350. But n some metrics at least they are pretty close.
It's not an argument I'm after , you used the words convincing and fair while being totally unfair,
if both CPU were running at the same clock speeds and the 8350 won that would be fair, and
I'm still wondering what was convincing to you, that if you leave the 2600k at stock clock an 8350
overclocked can beat it?
Thanks for the benchmarks but those are not real world applications we need actual games to be tested and actual programs such as Photoshop, handbrake, and many more. As always i take synthetic benchmarks with a grain of salt rarely do they represent real world tests
Around 15% performance per clock increase this is actually less then i thought i guess we should still wait a bit more, keep in mind the A10 6800K is already clocked at 4.1Ghz
As it was explained before, that benchmark is very sensible to memory. The one that I selected uses 8GB for the i5-2500k. Yours uses 16GB and therefore the i5-2500k get bigger scores. In both cases Kaveri only uses 4GB of slow memory. Therefore my link gives a better comparison of Kaveri to an i5 than yours.
Unlike you want make the silly claim that an i5-2500k is fastest than an i5-2500k you are comparing apples to oranges.
I'm not arguing with you, I think everyone agrees the i7's will beat an 8350. But n some metrics at least they are pretty close.
It's not an argument I'm after , you used the words convincing and fair while being totally unfair,
if both CPU were running at the same clock speeds and the 8350 won that would be fair, and
I'm still wondering what was convincing to you, that if you leave the 2600k at stock clock an 8350
overclocked can beat it?
Well we would expect the $330 USD CPU to be better then the $199 USD CPU. Though honestly the real value is the fx6350 at $139 USD. AMD is the value brand, I never expect them to the absolute best at anything, I only expect them to be very cheap for what they do provide. This is why I compare the fx8x series with the i5 as they are priced to compete with them at that level and the fx6x with the i3. The fx4300 is kinda misplaced as it doesn't really compete well with the lower pentiums and you'd most likely be better off going with APU.
Well to be fair when the 8350 was released it sold for 269.00, and the 2600k could be
purchased for 309.00 that's about a 40.00 dollar difference between the two CPU's.
We also expect the 8 core 5GHz overclocker to be able to handle almost 2 year old 4 core with
HT 2600k with no problem.
Now let's look at the 9590 for 899.00 when released should we expect more performance from this CPU?,
or the same cpu at 399.99 at it's current selling price?
jed :
Ags1 :
jed :
Ags1 :
jed :
Ags1 :
It's hard to trust benchmarks, so I wrote my own
I've written a Java app to compare CPUs and GPUs, and I have some results for a stock i7-2600K and an 8350 overclocked to 4.4GHz. When using Java, it is a pretty convincing win for the 8350. I conclude that the Java just-in-time compiler was not written by Intel!
My results would be improved by more data points, I do admit. You can always go to the download page on the site and contribute some more data if you like Be aware that it is a work-in-progress and has bugs (e.g. crashes on GTX 770). Also, you have to have Java installed.
Well this 2600k seam to be holding it own against that same 8350!
Thanks for submitting the result! Since 8350rocks submitted the result, I've tweaked the app to squeeze a bit more performance from the CPUs. A fairer comparison would be Palladin's 8350 which is overclocked higher and tested with a more recent version of the software:
As you might expect, the 8350 and 2600k are neck and neck on integer performance but the 2600k pulls ahead when we look at floating point and really dominates on memory (the memory test is cache intensive).
Ags1 :
I'm more interested in what the CPUs do at stock, but so far all the 8350 results have been OC'd. I was simply assuming your 2600k was overclcked too - it isn't? I had no intention of being unfair...
Ags1 :
I do have a comparison of i5-3570 (at stock reportedly) to 8350 (OC'd):
You also have my 3770k running @ 4.7GHz daily against the 8350 @ 4.7 you think.
http://www.headline-benchmark.com/results/45e8da85-12a3-44d3-b6f0-11b9eccb4525/9a411141-526d-4452-8e78-1c9891c2efa0
You could get the 8350 for slightly over $200 USD when it was released, it was quickly set to $200 USD. Your thinking the original FX8150 that was horrible overpriced for it's performance. The 3770K is priced at $339 USD right now which is about what the 2600K was going for during that time span. There was a big debate on whether purchasing the i7-2600K would of given a meaningful performance difference vs the much cheaper i5-2500K which is what the FX8 was slotted to compete with.
You don't compare clock vs clock you compare what you get for your money. That's why I tell people to stay away from the fx9 as it's just a golden binned fx8350 factory OC'd and a blatant cash grab.
Also the BD/PD uArch will never be able to perform the same as the SB/IB/HW uArch at the same clock. BD/PD has 2 ALU's per core for 16 ALU's per chip with four FPU's capable of 2x instructions each. The SB/IB/HW is 3 ALU's per core for 12 per chip with four FPU's capable of doing 3x instructions each. The BD/PD design is largely held back by it's poor L2/L3 cache performance with a hit being taken from the front end decoder. That's the price they had to pay for shoving 16 ALU's on a chip. Being at 32nm doesn't help either, that's a big reason why 4.8 is a reasonable clock ceiling.
BTW I've had mine to 5.0ghz but I needed to run the cooling system too high and I hate fan noise. My entire rig is designed to be quiet.
Here's the flipside to your argument, anyone who purchased the 2600k when first released for business like
myself had 1 year and 10 month's of productivity and profit before the 8350 was released and to be truthful,
anyone running a 2600k today is still above anything AMD has to offer right now including 9590.
When time is money one can't wait almost 2 years for performance close to the 2600k,
in less then 1 month I recovered all monies spent on the entire system not just the cpu.
BTW my 2600k highest overclock was 5.2 and the 3770k was 4.9
IP Performance has certainly increased but the FP is still down a bit ... the memory performance is still slower (if i read that correctly) too.
On the basis of past performance these benchies seem realistic?
Discuss (and thanks Juan) guys?
That bench where Kaveri is at i5-2500k level agrees with anything that we know and has been leaked.
The same bench gives ~30% IPC gain over Piledriver when compared to Trinity APU as discussed before. This 30% IPC coincides with leaked benchmarks from the cosmology site (also discussed before).
In my BSN* article I got i5-2500k level by assuming 20% IPC and 4GHz base clock. The same score is obtained by 30% IPC and 3.7GHz base clock. Therefore leaked benchmarks coincide with my predictions.
Finally 30% IPC means that Kaveri will be ~20% faster than Richland (because has 10% more frequency) and this 20% coincides with the 20% claimed by AMD in the leaked slide of the Russian talk. Again discussed here.
Therefore all the info available coincides in the same point.
The expected 30% IPC splits into ~20% due to double decoder in the module plus ~10% due to improvements in the execution units, caches, and rest of architecture.
SR Module ~ 1.3x PD module
SR Core ~ 1.1x PD core
In a single core test as that of above one must expect about 10% gain over Piledriver and 4.2/3.7 is about 13%.
IP Performance has certainly increased but the FP is still down a bit ... the memory performance is still slower (if i read that correctly) too.
On the basis of past performance these benchies seem realistic?
Discuss (and thanks Juan) guys?
The L1 Icache looks off, 96KB is a very odd size to have as it tends to be evenly divisible by powers of 2. After the fiasco that was BD I'm wary of anything posted as "engineering sample".
I am starting to think the memory problem is a structural one (design) that they have settled with, and decided to work on other areas instead. Intel's big advantage with the cache memory speed continues, and AMD won't catch up until they can address it. I imagine they can get more speed out of their design for the cache, but that the heat generated / power is something they have not sorted out ... thats my 2 cents worth anyway.
Looking all the way back to Barcelona they still haven't caught up with their previous gen - read the Scotty Wasson article.
As it was explained before, that benchmark is very sensible to memory. The one that I selected uses 8GB for the i5-2500k. Yours uses 16GB and therefore the i5-2500k get bigger scores. In both cases Kaveri only uses 4GB of slow memory. Therefore my link gives a better comparison of Kaveri to an i5 than yours.
Unlike you want make the silly claim that an i5-2500k is fastest than an i5-2500k you are comparing apples to oranges.
Here's the flipside to your argument, anyone who purchased the 2600k when first released for business like
myself had 1 year and 10 month's of productivity and profit before the 8350 was released and to be truthful,
anyone running a 2600k today is still above anything AMD has to offer right now including 9590.
When time is money one can't wait almost 2 years for performance close to the 2600k,
in less then 1 month I recovered all monies spent on the entire system not just the cpu.
BTW my 2600k highest overclock was 5.2 and the 3770k was 4.9
Uhm... I think those who bought an i7 2600k or 2700k at original prices paid for having the best mainstream CPU of its time, which is very rightful to Intel no matter how unfair it might seem software wise to AMD's 8150 and all. And you have a valid point, but as valid as it might be, it is moot: when the i7 3770K was introduced, the 8350 was also introduced a couple of months later IIRC. Then, the value proposition changed a lot. I won't repeat what we all know, but Sandy was uncontested favoring Intel at its time (no sane person seeking a certain performance level would have waited 1 year), but Ivy has PD to stand its ground in Perf/$. Haswell could as well be a match to it taking price into the equation.
Now, in your "performance is king" kind of argument for business. Then why not an SB-E or IB-E right away for any performance build? They do offer better raw performance from the get go. Even the good old 980X is still prevalent in the """""aging""""" X58 chipset (and just to remark, notice the quotes around "aging"). We all know why; targeted performance of applications. The i7s are a (tad) better than the 8350 (and 6300/6350) when OC'ed; that's a fact, but they do cost more as well. I don't really know how you value 30s (maybe less? don't remember) of rendering time, for instance, in favor of any i7 over an 8350 (OC'ed for both camps), but I would argue that the difference might be meaningless in the long run, contrary to popular belief. You will spend more time IDLEing (watching YouTube, talking with co-workers, posting in tech forums , etc) than the CPU making intensive work worthy of the original price difference.
As it was explained before, that benchmark is very sensible to memory. The one that I selected uses 8GB for the i5-2500k. Yours uses 16GB and therefore the i5-2500k get bigger scores. In both cases Kaveri only uses 4GB of slow memory. Therefore my link gives a better comparison of Kaveri to an i5 than yours.
Unlike you want make the silly claim that an i5-2500k is fastest than an i5-2500k you are comparing apples to oranges.
Probably it's because of turbo and different thermal room for different rigs.
But still I don't know how to explain things like this: http://browser.primatelabs.com/geekbench3/compare/120097?baseline=75030
Explained before and again the message that you quote. 8GB vs 16GB. The i3 with double ram get higher scores. The same that the i5-2500k with double RAM got higher scores.
Anyone has something to comment on nVidia GameWorks? it sounds like something really really Evil... can anyone share any visions on the matter?
It seems that Nvidia is taking a page out of Intels evil playbook with this tactic... If you can't beat them then cheat! I still can't understand why anyone would ever support these 2 companies (intel & nvidia) and their sleazy underhanded practices!
I fail to see the issue; nothing more then a set of libraries that are optimized for Geforce cards. If NVIDIA want's to create a library that biases their own product, good for them. Now convince everyone to use it.
I mean, sheesh, are we REALLY going to start an open source flame war here?
No flame wars here I am just stating the fact that Nvidia is using an intel shady tactic. I have no problem with AMD or Nvidia optimizing a game towards their GPUs, where I run into an issue is when a company is intentionally tries to cripple the competition like what Nvidia is doing here and what intel did with their compilers.