@number6

One point to note is that write speed is much more important for VRAM since most of the CPU Video traffic is in that direction.

_________________
Doing stupid things for fun...

@Karlos

Quote:

One point to note is that write speed is much more important for VRAM since most of the CPU Video traffic is in that direction.

Sure, but a 4:1 difference with a fellow Micro owner?

And 30%-50% increase on some other stats.

#6

Last edited by number6 on 06-Nov-2010 at 11:12 PM.

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

For fun, with OS4Emu:

Pegasos II 1GHz

RAGEMEM v0.37 - compiled 11/06/2010

CPU: PowerPC 7447 0.0 @ 1000 Mhz
Caches Sizes: L1: 32 KB - L2: 256 KB - L3: 1 MB
Cache Line: 0

--- CPU ---
MAX MIPS: 2989

--- L1 ---
READ32: 3794 MB/Sec
READ64: 7543 MB/Sec
WRITE32: 3255 MB/Sec
WRITE64: 2535 MB/Sec

--- L2 ---
READ32: 1783 MB/Sec
READ64: 2333 MB/Sec
WRITE32: 1598 MB/Sec
WRITE64: 2025 MB/Sec

--- L3 ---
READ32: 250 MB/Sec
READ64: 253 MB/Sec
WRITE32: 467 MB/Sec
WRITE64: 519 MB/Sec

--- RAM ---
READ32: 210 MB/Sec
READ64: 210 MB/Sec
WRITE32: 418 MB/Sec
WRITE64: 418 MB/Sec
WRITE: 429 MB/Sec (Tricky)

--- VIDEO BUS ---
READ: 52 MB/Sec
WRITE: 220 MB/Sec

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

PowerBook 1.67GHz

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Unknown 0.0 @ 1666 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 0

--- CPU ---
MAX MIPS: 4992

--- L1 ---
READ32: 6347 MB/Sec
READ64: 12690 MB/Sec
WRITE32: 5440 MB/Sec
WRITE64: 4233 MB/Sec

--- L2 ---
READ32: 2884 MB/Sec
READ64: 3738 MB/Sec
WRITE32: 2601 MB/Sec
WRITE64: 3296 MB/Sec

--- RAM ---
READ32: 400 MB/Sec
READ64: 400 MB/Sec
WRITE32: 770 MB/Sec
WRITE64: 770 MB/Sec
WRITE: 803 MB/Sec (Tricky)

--- VIDEO BUS ---
READ: 32 MB/Sec
WRITE: 159 MB/Sec

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

Mac mini 1.42GHz

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Unknown 0.0 @ 1416 Mhz
Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 0

--- CPU ---
MAX MIPS: 4237

--- L1 ---
READ32: 5386 MB/Sec
READ64: 10768 MB/Sec
WRITE32: 4616 MB/Sec
WRITE64: 3593 MB/Sec

--- L2 ---
READ32: 2439 MB/Sec
READ64: 3183 MB/Sec
WRITE32: 2204 MB/Sec
WRITE64: 2791 MB/Sec

--- RAM ---
READ32: 388 MB/Sec
READ64: 404 MB/Sec
WRITE32: 775 MB/Sec
WRITE64: 775 MB/Sec
WRITE: 815 MB/Sec (Tricky)

--- VIDEO BUS ---
READ: 32 MB/Sec
WRITE: 180 MB/Sec

Last edited by Fab on 06-Nov-2010 at 11:30 PM.

@Fab

Suddenly my main memory scores don't look so absurd. Well, that's not true, but at least they are not alone in their absurdity.

The Peg2 uses a different Northbridge, right (Marvel?). Yet you also seem to get much slower read than write. Stranger still, your write performance is less than my Articia based system (though your read speeds are better), despite having a faster CPU. What's the FSB speed on that machine?

_________________
Doing stupid things for fun...

@Karlos

Having slower read than write is perfectly normal.

And regarding write speed, let's say that there are different kinds of memory setups, some optimized for useless benchmarks and some optimized for real world usage.

Typically, this 6xx MB/s write speed on amigaones is quite meaningless.

@Fab

Having *far* slower reads than writes is normal for VRAM but I'm not used to seeing it on regular SDRAM / DRAM setups where most systems I've tested in the past tend to show either even performance or biased towards reading. Contrast the G4/SDRAM in my A1 to the 603/FPM memory in my A1200. The latter exhibits much more the behaviour I'd expect to see. The differential in read v write on the former is far outside what I'd expect and is much more akin to a VRAM setup.

Of course, without seeing the benchmark code, it's not really possible to comment the impact of data cache etc.

_________________
Doing stupid things for fun...

@Number6
My test (setA1 actived)

Micro A1 @ 800Mhz with 512mb ram

RAGEMEM v0.37 - compiled 11/06/2010

CPU: IBM PowerPC 750 FX 2.2 @ 800Mhz

Caches Sizes: L1: 32 KB - L2: 512 KB - L3: none
Cache Line: 32

---CPU---
MAX MIPS: 1595


---L1---

READ32: 3046 MB/Sec
READ64: 6090 MB/Sec
WRITE32: 2708 MB/Sec
WRITE64: 4062 MB/Sec

---L2---

READ32: 1587 MB/Sec
READ64: 2154 MB/Sec
WRITE32: 1423 MB/Sec
WRITE64: 2012 MB/Sec


---RAM---

READ32: 246 MB/Sec
READ64: 246 MB/Sec
WRITE32: 138 MB/Sec
WRITE64: 138 MB/Sec
WRITE: 669 MB/Sec (Tricky)

---VIDEO BUS---

READ: 13 MB/Sec
WRITE 134 MB/Sec

Your video bus is the highest that we have seen from micro A1

@bhop
And another Micro

Micro A1 running at 800Mhz with 256mb ram

RAGEMEM v0.37 - compiled 11/06/2010

CPU: IBM PowerPC 750GX 1.1 @ 800Mhz

Caches Sizes: L1: 32 KB - L2: 1024KB - L3: none
Cache Line: 32

---CPU---
MAX MIPS: 1593


---L1---

READ32: 3031 MB/Sec
READ64: 6078 MB/Sec
WRITE32: 2697 MB/Sec
WRITE64: 4059 MB/Sec

---L2---

READ32: 1050 MB/Sec
READ64: 1255 MB/Sec
WRITE32: 874 MB/Sec
WRITE64: 1037 MB/Sec


---RAM---

READ32: 199 MB/Sec
READ64: 199MB/Sec
WRITE32: 125MB/Sec
WRITE64: 125MB/Sec
WRITE: 671MB/Sec (Tricky)

---VIDEO BUS---

READ: 13 MB/Sec
WRITE 135 MB/Sec

_________________
Everything is proceeding as I have foreseen

Some people are like slinkies. Not really good for anything but they bring a smile to your face when pushed down the stairs.

@bhop

Quote:

Your video bus is the highest that we have seen from micro A1

And higher than almost every posted result regardless of card/platform

And -you- are showing a huge difference in L2 cache with Micro FX vs the GX.

Quote:

My test (setA1 actived)

That's unusual bump in the #s from seta1, imo. Others (including the author of setA1 himself), seemed to indicate 10% would be a reasonable improvement.
Sure, he wrote that 5 years ago, but that 10% seems in line with what sundown is showing me on a chart comparing some entries in this thread. Perhaps your entries in setA1 are different than sundowns?

Oh, welcome to Amigworld posting!

Sundown's GX results are still puzzling, although his vcore is way low for sake of stabilization. And his memory is surely different than mine. (there's a joke there somewhere)

#6

Last edited by number6 on 07-Nov-2010 at 02:00 PM.

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@ajs

Looks quite similar to my GX, except for that one Video Bus entry.
Thanks for posting. It's hard to even get a sense of things with so few posts.

@redfox/@oldfart/@Chris_Y

I guess you realize I would enjoy seeing your entries here. (nudge)

@crisot

I would not mind being informed of what we should be expecting from these numbers.

@Karlos

Quote:

The Peg2 uses a different Northbridge, right (Marvel?).

Marvell Discovery II northbridge and Via VT8231 southbridge

@thread

Repeating link to the test:
http://crisot.free.fr/universe/files/RageMem.lha

#6

Last edited by number6 on 07-Nov-2010 at 02:36 PM.
Last edited by number6 on 07-Nov-2010 at 01:54 PM.
Last edited by number6 on 07-Nov-2010 at 01:13 PM.
Last edited by number6 on 07-Nov-2010 at 01:04 PM.

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@number6

Quote:

@crisot I would not mind being informed of what we should be expecting from these numbers.

Hummm... I did this bench firstly to see the AmigaOne memory bandwidth compared to more recent platforms like the Pegasos 2 or the Sam440/460, just to see if the SDR->DDR->DDR2 really make the difference.

When AOne was available, many people said it was crap, because its SDR performances where very low compared to the same ram on X86. And that's true. But actually, even if newer Os4 hardware with DDR/DDR2 are a bit faster than AOne, they are also really far from real DDR/DDR2 performances in X86 world. PowerMAC benchmarks posted by Fab show quite the same "problem" on Mac. Maybe this comes from the PowerPC architecture, I don't know...

Also, I've seen a Sam440 running for the first time a few weeks ago in a french party, and found, with all my respect for the Sam440, that it delivers horribly slow performances, much slower than what I excepted. This benchmark was a way to see ram speed, video bus speed, cache speed, on this hardware.

So you see my first goal was not very noble. Just curiosity, "who got the biggest".

Thoses benchmarks are reliable for big ram buffer comparison: read, write, copy of a big buffer to another. There is no "benchmark optimisation". But performances may be different with a "random acces" memory test, where latency is more important than bandwidth. Hierarchy may not be the same.

_________________
AmigaOne XE --- 7455A@1400 --- Radeon 9000 Pro --- RAM 2GB --- SSD 40GB

@Crisot

Quote:

So you see my first goal was not very noble. Just curiosity, "who got the biggest".

ok. Then we'll let others comment, if they so desire. I'm still seeing variances amongst near identical h/w that seem beyond reasonable.
Correct me if I'm wrong about this.

btw-seta1 is showing approximately 15% increase, if I average the few who have tested it recently or used it in the past. But as I stated in irc, soak testing for 6 years without using it does not = running one test -with- it. Only a handful of people I see posting have used it from day1, and they can better speak about their own experiences.

#6

Last edited by number6 on 07-Nov-2010 at 04:43 PM.

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@Crisot

Quote:

Also, I've seen a Sam440 running for the first time a few weeks ago in a french party, and found, with all my respect for the Sam440, that it delivers horribly slow performances, much slower than what I excepted. This benchmark was a way to see ram speed, video bus speed, cache speed, on this hardware.

Would be interesting to know the specs of that Sam 440.
My 800 Mhz Sam 440ep-flex model is quite snappy.
It seems to be one were the L1 write speeds results are funny.

Another funny one I noticed posted by Slayer on Amigans.net with a 800Mhz Sam 44ep-flex
where his Ram write32 and write64 results are the same . His L1 write
speeds seem ok though.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

@Spectre660

Those 2 really stood out to me as well.

#6

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@Crisot

Interesting results. what are the theoretical max speed for DDR/DDR2 and how much are the OS4 systems close to the max.

Is it possible to compare similar systems to the wintel world and do we come close to those results.

how much each model deviates from each other, I hope we dont get into a situation of a 'benchmark' before we buy attitude.

_________________
Download 499.26 Mbps, 659.94 Mbps Upload :)

@Crisot

Well, I think the poor memory performance of the Peg2 comes from DDR RAM being only used in Single Data Mode. A Powermac G4 with SDRAM has nearly the same memory performance.

/me have to find time to run this test, because it seems no-one has run this on a 666Mhz SAM & onboard GFX RAM...

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@Geennaam

Note my results from the same model and speed Machine


---> RAM ---
READ32: 299 MB/Sec
READ64: 299 MB/Sec
WRITE32: 166 MB/Sec
WRITE64: 329 MB/Sec
WRITE: 902 MB/Sec (Tricky)

Quote:

Geennaam wrote: @Spectre660 Quote: Another funny one I noticed posted by Slayer on Amigans.net with a 800Mhz Sam 44ep-flex where his Ram write32 and write64 results are the same Keep in mind that the 440ep has an 32bit memory interface as opposed to the 64bit memory bus of an AONE and PegII. So there's nothing to gain from 64bit write since they're translated to 32bit anyway. The one thing that I think is funny is that the video write performance on an SAM is only 50 MB/s. The graphics PCI bus is supposed to run at 66MHz, so a theoretical performance of 266MB/s.

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

Tested before/after running 'SetA1 fastmem fastcas'
Btw, my gfx card is a Radeon 7500 AGP w/128MB

RAGEMEM v0.37 - compiled 11/06/2010

CPU: Motorola MPC 7445/7455 Apollo 2.1 @ 800 Mhz
Caches Sizes: L1: 32 KB - L2: 256 KB - L3: none
Cache Line: 32

--- CPU ---     
MAX MIPS:  2396 

--- L1 ---
READ32:  3031/3045 MB/Sec
READ64:  6082/6087 MB/Sec
WRITE32: 2606/2609 MB/Sec
WRITE64: 2030/2031 MB/Sec

--- L2 ---
READ32:  1402/1407 MB/Sec
READ64:  1828/1835 MB/Sec
WRITE32: 1259/1260 MB/Sec
WRITE64: 1591/1590 MB/Sec

--- RAM ---
READ32:  181/220 MB/Sec
READ64:  181/220 MB/Sec
WRITE32: 255/255 MB/Sec
WRITE64: 644/644 MB/Sec
WRITE:   664/664 MB/Sec (Tricky)

--- VIDEO BUS ---
READ:     16/16  MB/Sec
WRITE:   169/169 MB/Sec

_________________
AlexC's free OS4 software collection

AmigaOne XE/X1000/X5000/UAE-PPC OS4 laptop/X-10 Home Automation

this is the part im interested in Blizzard PPC video bus vs Cyberstorm video bus results from another website and is not my card.

depending on how the card is config the PCI bus speed is done by the 68k oscilllator .

what speed is the 68040/68060?

http://www.amigans.net/modules/newbb/viewtopic.php?topic_id=4433&viewmode=flat&order=ASC&type=&mode=0&start=40

Last edited by delshay on 10-Nov-2010 at 02:31 AM.
Last edited by delshay on 10-Nov-2010 at 02:29 AM.
Last edited by delshay on 10-Nov-2010 at 02:24 AM.
Last edited by delshay on 10-Nov-2010 at 02:22 AM.
Last edited by delshay on 10-Nov-2010 at 02:20 AM.

_________________
The Machine: Bride Of The Pin•Bot by Williams Electronics