FX-8370 - non E version few tests

[Abwx]

Not sure what case you are making here, but it doesn't seem to be one for the 8370E, from your observation we should be recommending the i3 or FX-6300?

Agree on this point, let say a FX6 or a i5 at about i3 prices are more than enough, here one could get eventualy a FX8320 wich is 20% more expensive than a FX6, with 33% more cores it has intrinsicaly a better price/perf ratio.

Those who value absolute perf/prices would be better served with Pentium, Celerons or AM1 APUs, the A8 7600 is really AMD s most attractive APU offering but clearly not in that segment price wise.

That is for the average users of course, hardcore gamers or people that need huge CPU throughputs are a different matter, on another note it s just amazing how a hobby like gaming could be more compute hungry than professional applications...

 

[Abwx]

o

I know you weren't directly addressing me, but I want to respond. Surely you must know that the web is replete with numbers that indicate AMD's inferior position in general. Even the articles you cite, when read in their entirely, support this. If only the 8370's competition was the 2500K, then the decision would be a lot harder.



http://www.anandtech.com/bench/CPU/1052

I ought to put at least one single-thread bench in here to make my point:

http://www.anandtech.com/bench/CPU/1028

Here's an interesting read that helps illustrate the difference between approaches, and a chart that maps out ranges of loading and approximate throughput for dozens of different CPUs. Should be taken with a grain of salt because the methodology is not revealed:

http://www.overclock.net/t/1493307/relative-access-to-execution-throughput-comparison-chart

I ll start with Handbrake with the 2500K doing a strong showing since it s only 12.44% slower than the FX8350 but from memory i didnt remind the FX being so mediocre in HB compared to the 2500K, because when i check Handbrake at Techreport i get this, that is the 2500K is 35% slower, so really, what happened at AT ??? :

Second is your single thread exemple wich use CB, as already said, and also aknowledged by Maxon, it is optimised for Intel but nevermind even if there s not a single person in the universe that will use this soft in ST, let s admit that there are people that will do it, but then as said i proposed to average the FP MT scores of CB and Povray such that one uarch is not favoured by the soft, so let s look at another ST FP bench, that is povray, as said i think that doing so is quite fair :

2500K is better but nothing like CB, isnt it..??.

You think that we should compile the whole thing using ICC eventualy..?.
We could get about CB scores hierarchy this way, it would be fun to also use an FP test that is AMD friendly, why always the same that get the sweet and easy to digest code..?

Last, but not least, is this huge ock.net data outputs, since both the 2700K and FXs are 8 threads CPUs i ll use the 8 thread throughput, i must admit that it would take me more time to do a synthesis of the whole thing but since we are not at work we have the right to some lazyness.

Let s take both the 2700K and the FX at 4.8Ghz and 8 threads, apparently the 2700K is...4.5% better at the same frequency.

The 2500K at 4.8 has 20% lower throughput than the 8350/4.8 and about 25% less than the 2700K/4.8, am i right.??..

I didnt comment on povray MT but i ll post AT test in Vishera review as a hint for the FX8320 and 6300, you can downgrade all scores accordingly to match the 8370E 3.3 base frequency in MT but also its 4.3 for ST tasks.

 

[Enigmoid]

Why is the pentium performing better than the 4360?

Can we try and stay on topic.

Anyway it looks like a good chip addressing a lot of the problems with the 8350. Overclocks well too.

 

[Abwx]

Why is the pentium performing better than the 4360?

.

Because they made a typo....

Can we try and stay on topic.

Anyway it looks like a good chip addressing a lot of the problems with the 8350. Overclocks well too.

Only the chipset is somewhat lacking actualy, fortunately for thoses who will use it in 95W boards they have generaly less power hungry chipsets than the 990 series.

 

[crashtech]

o

I ll start with Handbrake with the 2500K doing a strong showing since it s only 12.44% slower than the FX8350 but from memory i didnt remind the FX being so mediocre in HB compared to the 2500K, because when i check Handbrake at Techreport i get this, that is the 2500K is 35% slower, so really, what happened at AT ??? :

Second is your single thread exemple wich use CB, as already said, and also aknowledged by Maxon, it is optimised for Intel but nevermind even if there s not a single person in the universe that will use this soft in ST, let s admit that there are people that will do it, but then as said i proposed to average the FP MT scores of CB and Povray such that one uarch is not favoured by the soft, so let s look at another ST FP bench, that is povray, as said i think that doing so is quite fair :

2500K is better but nothing like CB, isnt it..??.

You think that we should compile the whole thing using ICC eventualy..?.
We could get about CB scores hierarchy this way, it would be fun to also use an FP test that is AMD friendly, why always the same that get the sweet and easy to digest code..?

Last, but not least, is this huge ock.net data outputs, since both the 2700K and FXs are 8 threads CPUs i ll use the 8 thread throughput, i must admit that it would take me more time to do a synthesis of the whole thing but since we are not at work we have the right to some lazyness.

Let s take both the 2700K and the FX at 4.8Ghz and 8 threads, apparently the 2700K is...4.5% better at the same frequency.

The 2500K at 4.8 has 20% lower throughput than the 8350/4.8 and about 25% less than the 2700K/4.8, am i right.??..

I didnt comment on povray MT but i ll post AT test in Vishera review as a hint for the FX8320 and 6300, you can downgrade all scores accordingly to match the 8370E 3.3 base frequency in MT but also its 4.3 for ST tasks.

So if your point is to prove that 4M/8T Visheras are better than the 2500K clock for clock when fully loaded, what is there to argue about? It's true. But my point is that there are many scenarios where the CPU is heavily loaded but not fully saturated. It's hard to test these scenarios unless you want to take an ST bench and run n+1 instances of it and record the results. Even that is not representative of a real use scenario, but it would come closer and be an interesting metric. My earlier point still stands, I think, though it is tough to back up empirically: Clock for clock, SB and up Intel quads, even w/o HT, will outperform Vishera up to around 6 fully loaded threads.

 

[SlowSpyder]

So this is quickly becoming another AMD vs. Intel thread. Cool, we needed another, I don't think we've had one in something like three days.

 

[Abwx]

But my point is that there are many scenarios where the CPU is heavily loaded but not fully saturated. It's hard to test these scenarios unless you want to take an ST bench and run n+1 instances of it and record the results. Even that is not representative of a real use scenario, but it would come closer and be an interesting metric.

Why a non realistic scenario would be an interesting metrics..?

My earlier point still stands, I think, though it is tough to back up empirically: Clock for clock, SB and up Intel quads, even w/o HT, will outperform Vishera up to around 6 fully loaded threads.

So now it s clock for clock, it s no more stock frequencies, i guess that this is a way to cut the scores, but if AMD designed a high frequency CPU why should it be used at Intel stock frequencies.?

But apparently it s not enough, we must account only for 6 cores, heck, they must abide by Intel s core counts for the bench to be of any relevance..??.

Because it s the only mean to make a point..?

And you are using FP results as a general rule, i told you that Vishera is much better at integer tasks being a server CPU by design, Vishera has been massacred by badly threaded softs particularly in integer tasks and you are stuck with thoses old images, like this one, look at the 2500K score :

Let s see what happen with a correctly threaded integer task that is about the same :

What happened to the 2500k.???.

But for the fun let s reduce the 8350 score by 61% to get to 6 cores at 3.3 and the result is....14519, i thought that it should had been outmatched by the 2500k.??.

Your earlier point doesnt stand, sorry.

Now please can we get specificaly on topic since we are again at 3.3Ghz, that is, the base frequency of the 8370E whose score should be about 19300 in this bench.

So this is quickly becoming another AMD vs. Intel thread. Cool, we needed another, I don't think we've had one in something like three days.

Actualy i m discussing the 8370E perfs using all Vishera s (s)cores, anybody can deduct the scores at 3.3...

 

[crashtech]

OK, AMD FTW, and apologies to all for the OT material.

 

[Abwx]

OK, AMD FTW, and apologies to all for the OT material.

That was a nice demonstration of yours, seriously, i must admit that i did ignore some numbers before this exchange, of course all is not that satisfying in AMD s FXs but one should render to Caesar what is Caesar s, the chip is more potent currently that it was two years ago, in this respect the 8370E landed quite at a favourable time, too bad its plateform is not up to its qualities, sure that it will please the overclocking crowd but it s a waste that they didnt update their plateform for regular people like me who do not overclock at all.

 

[Gikaseixas]

So this is quickly becoming another AMD vs. Intel thread. Cool, we needed another, I don't think we've had one in something like three days.

Intel fanboys can't resist seeing a AMD thread. They're consistent and always manage to wreck what could be a nice thread.

Back on topic... I didn't know that even the 8370E could be pushed that high (5ghz plus). It would have been nice if AMD updated motherboard chipsets at least. These things deserve a better / newer platform.

 

[DrMrLordX]

AM3+ either needs a new chipset or just needs lower prices. Lower prices seem perhaps more realistic.

But the cheapest 8+2 990FX board you can get is around $110-$120 (you can get a much cheaper 970 8+2 board, but that's an even older chipset).

 

[AtenRa]

AM3+ either needs a new chipset or just needs lower prices. Lower prices seem perhaps more realistic.

But the cheapest 8+2 990FX board you can get is around $110-$120 (you can get a much cheaper 970 8+2 board, but that's an even older chipset).

9xx series Chipsets were all released together at the same time.

 

[escrow4]

Cannonlake will be compatible with 1150 MBs.?

Who cares? You are stuck with AM3+ which is ancient. At least Intel doesn't let its chipsets stagnate.

 

[Abwx]

Who cares? You are stuck with AM3+ which is ancient. At least Intel doesn't let its chipsets stagnate.

So there will be new chipsets for LGA1150..?

 

[AtenRa]

Who cares? You are stuck with AM3+ which is ancient. At least Intel doesn't let its chipsets stagnate.

AMD 990FX motherboards (no PLX) (released 2011)

PCIe Gen 2.0 x32 lanes (2x 16 or 4x 8)

Quad CF support (up to 4x GPUs)
Quad SLI support (up to 4x GPUs)

USB3 x4 ports (through extra controller)
6x SATA 6 ports

Memory : 4x slots up to 32GB ECC/non ECC

---------------------------------------------------

Intel Z68 Motherboards (no PLX) (Released 2011)

PCIe Gen 2.0 x16 lanes (1x 16 or 2x 8)

Quad CF support (Only dual GPU)
Quad SLI support (Only Dual GPU)

USB3 x4 ports (through extra controller)
2x SATA 6 ports

Memory : 4x slots up to 32GB non ECC

----------------------------------------------------------

Intel z97 Motherboards (no PLX) (Released 2014)

PCIe Gen 3.0 x16 lanes (1x 16 or 2x 8)

Quad CF support (Only dual GPUs)
Quad SLI support (Only dual GPUs)

USB3 x6 ports (natively)
6x SATA 6 ports

Memory : 4x slots up to 32GB non ECC

------------------------------------------------

So Intel needed three years to finally reach parity with 990FX features (except natively USB3 support). They still dont support 4x GPUs or ECC on those Chipsets.

Also, there are new AM3+ motherboards with even M2 SSD Gen 2.0 (10Mbps) slot like ASROCK Fatal1ty 990FX Killer.

 

[Gikaseixas]

So Intel needed three years to finally reach parity with 990FX features (except natively USB3 support). They still dont support 4x GPUs or ECC on those Chipsets.

This is true but Intel's SATA performance makes a difference and the chipset sips power thus allowing for better thermal and consumption characteristics. Point is, AMD needs an upgrade there, regardless of how many bullet points are there...

 

[AtenRa]

This is true but Intel's SATA performance makes a difference and the chipset sips power thus allowing for better thermal and consumption characteristics. Point is, AMD needs an upgrade there, regardless of how many bullet points are there...

Well the difference in SATA performance is only academic between AMD 950 Chipset and latest Intel.

http://www.tomshardware.com/reviews/storage-controller-performance-ssd,3540.html

I do agree they had to bring a new chipset for AM3+ since they were planing to keep that Socket for so long especially with USB-3 natively support and PCIe Gen 3.0.
But since AM3+ is a dual Chipset configuration, power consumption would not get substantially lower. Perhaps 10-15W tops if they would use 32nm SOI process on both North and South bridge Chipsets.

 

[ShintaiDK]

Well the difference in SATA performance is only academic between AMD 950 Chipset and latest Intel.

http://www.tomshardware.com/reviews/storage-controller-performance-ssd,3540.html

Arh, its quite abit more than that arcording to your link.

Even the article says you are going to have a better experience with Z77/Z87 than SB950.

 

[AtenRa]

Arh, its quite abit more than that arcording to your link.

Even the article says you are going to have a better experience with Z77/Z87 than SB950.

I can also quote the following from the same review since you had to only cherry pick the worst you could find.

All those benchmarks really show that in real life the performance difference is only academic. You will not even feel the difference using the same SSD with AMD or Intel SATA-6 controllers in real life workloads.
Next time try and be more objective.

 

[Gikaseixas]

AtenRa there's a fluid snappiness on Intel platforms that you won't see on a 990FX board. This is purely subjective but again, just like I always say that you would never be able to distinguish between Intel i5/i7/ AMD FX 8xxx / 9xxx in gaming experiences, I cannot say the same for desktop usage.
I game on my FX machine and enjoy as much as I do on my i7 but I prefer the Intel rig for my other tasks. I could be wrong about the entire thing but I doubt

 

[Gikaseixas]

Perhaps 10-15W tops if they would use 32nm SOI process on both North and South bridge Chipsets.

Are they on 45nm still? What is Intel using for Z97 / Z87?

 

[AtenRa]

Are they on 45nm still? What is Intel using for Z97 / Z87?

They are on 65nm with 9xx series chipsets. Z77 was also made at 65nm but it doesnt have PCIe Gen 3.0 lanes, those are in the CPU die.
The problem is not the chipset alone, the problem is they need to upgrade to FM2+ stile of socket with PCIe integrated in to CPU die in order to have less latency and lower power consumption due to smaller lithographic process.
But they only keep AM3+ in order to still keep 32nm process volume as high as they can.
They could easily produce a Quad Module 8 Thread 28nm FM2+ Die close to 240mm2(no L3) or less. But then AM3+ and 32nm would stop to exist with whatever consequence that would have to both AMD and GloFo. Well more to GloFo since 32nm SOI was strictly used by AMD alone.

 

[ShintaiDK]

Are they on 45nm still? What is Intel using for Z97 / Z87?

Intel is using 32nm for the PCH today. AMD is using 65nm for both chips. unless it got IGP, then the north is 55nm. The last update besides the rebrand of northbridge controllers was in Q2 2011 for the southbridge.

The HT link is limited to 10.4GB/sec and the A-link Express is 2GB/sec. The extra jump for anythign southbridge related isnt helping either.

 

[ShintaiDK]

All those benchmarks really show that in real life the performance difference is only academic. You will not even feel the difference using the same SSD with AMD or Intel SATA-6 controllers in real life workloads.
Next time try and be more objective.

Performance never matters does it?

 

[Gikaseixas]

thanks for the info guys, I didn't know much about this, always thought it was 45nm for AMD, 32nm for Intel