Rams, Fsb, Utt Figņas.

[DoubleT]

Tātad izlasīju bodes sadaļā noreka postu un gribētu uzzināt par RAMU un FSB ( Front Side Bus cik sapratu ) ...

Kur varu zināt ar cik FSB rukā lai zinu cik RAMu likt vēl iekšā.?

tātad nav tolka kompī ar 400 FSB ātrumu likt iekšā 2GB RAMu domājot ka kompis būs ūber kruts ja.?

vēl viens jaut. cik mHz ir droši over-clokot CPU.? pagaidām man ir 2.3 ghz celerons ;]]

vēl... vai varētu būt ka kompis rukā mazliet lēnāk, ja defaulti tajā dinozaurā bija kkāda 32MB videne un 256MB rams, bet ieliku citu videni ( no vecā PC ) 256MB radeonu 9600...

[Norek]

Lai gari nerakstītu, te ir perfekts raksturojums ar piemēriem. Ja Tu šo izlasīsi 3 reizes, tad viss taps skaidrs. Pats uzzināju pāris lietas, kuras likās miglā tītas.

The FSB speed is the speed of the link between the processor and chipset.

The memory speed is the speed of, well, the memory and its bus to the chipset.

Modern chipsets can run them independently without a significant performance hit. Chipsets support specific memory speeds. For instance, the i945 series chipsets support DDR2-400 through DDR2-667. The i975X chipset drops support for DDR2-400, and supports DDR2-533 and DDR2-667, as do the 965 series chipsets. The "Bearlake" P35/G3x/Q3x/X38 chipsets support DDR2-533 through DDR2-800. The main concern is that the memory speed should not bottleneck the performance of the system.

With Intel-based processors, the processor has to talk to the memory through the chipset, where the memory controller resides. So there are two pipes, one between the processor and the memory controller hub (chipset) and the other between the memory controller hub and the RAM. These are the FSB and memory buses respectively. As you might imagine, you'd want their bandwidth (data rate) to match.

Now when someone talks about a "1066 MHz FSB", what they're really talking about is a FSB clocked at 266 MHz, with four transfers a cycle (hence 266 x 4 = 1066 MT/s, million transfers a second). Each transfer is 64 bits. This comes to 8,533 MB/s. So you'd want a memory solution that can match or exceed this bandwidth, otherwise you are potentially creating a bottleneck with slow RAM.

As it happens, modern memory controllers can talk to two channels of memory, which means that if you pair up your memory modules, one in each of the pair of channels, your memory controller will be able to talk to two memory modules per transfer.

As an example, let me talk about PC2-4200, aka DDR2-533. This RAM has a clock rate of 266 MHz and transfers twice a clock cycle. Each transfer is 64 bits wide. So a single DDR2-533 channel can provide 4,266 MB/s. (This is also why it's called PC2-4200. Similarly, DDR2-667 can provide 5,333 MB/s and is often sold as PC2-5300, DDR2-800 can provide 6,400 MB/s and is sold as PC2-6400, and DDR2-1066 can provide 8,533 MB/s and is sold as PC2-8500.)

If you have paired RAM, your memory controller will be able to talk to the RAM transferring up to 8,533 MB/s. Note that this matches a 1066 MT/s FSB, so in this case the system is balanced -- neither the FSB nor the RAM speed is a bottleneck in the data flow between the processor and the RAM. If you put in faster RAM, say DDR2-667 or DDR2-800, you won't see a benefit in the memory bandwidth you can utilize since the FSB becomes a bottleneck.

However, this doesn't mean you won't see a benefit in performance if you go for faster memory. When a memory controller makes a request to memory, there are several cycles it has to wait ("latency") before it can get data. Different grades of RAM at a particular speed (say, DDR2-667) are sold with different latency characteristics. For instance, one such latency parameter is the column access strobe (CAS) latency, which is typically 4 or 5 cycles for DDR2 SDRAM. Now, this delay is expressed in terms of memory clock cycles, so with DDR2-533 CL4 memory, we're talking about 4 cycles / 266 million cycles per second = about 15 nanoseconds. With DDR2-667 CL5 memory, which is about the same price these days, you get about the same latency (5 cycles / 333 million cycles per second = 15 ns). With DDR2-667 CL4 memory, it's 12 ns. With DDR2-800 CL4 memory it's 10 ns. With DDR2-1066 CL5 memory it's 9.375 ns. Note that these numbers don't depend on how much data you can shove through the pipes (bandwidth), but rather the clock rate.

The latency is a significant factor since the processor is doing nothing between the first request and the first response, and since memory access tends to be bursty, with lots of stops and starts, this adds up rather quickly.

If you're overclocking an Intel system, you are raising the FSB clock rate. This also increases the bandwidth the processor can use, so it helps to have faster RAM in this situation since the processor continues to be matched by the RAM in bandwidth.

The Athlon 64 and Athlon 64 X2 processors are somewhat different in architecture. In their case, the memory controller is integrated on to the processor itself, so there is no FSB that limits the bandwidth used -- the limit is the speed of memory that the processor itself supports. Currently this is DDR2-800 for current "K8" generation Athlons. For the upcoming "K10" generation Athlon and Phenom processors, it is DDR2-1066.

The AMD processors are a bit tricky when it comes to memory clock rates, however. These processors derive their memory controller's clock rate from the processor's clock speed. (This processor clock speed is in turn determined by the HyperTransport bus that feeds off the system clock generator, with a multiplier on the processor. This factors in when overclocking these processors, which I will talk about later.)

As it happens, AMD's processors are clocked at 100 MHz intervals between speed grades, so you have 1.6 GHz, 1.8 GHz, 1.9 GHz, 2.0 GHz, 2.1 GHz, 2.2 Ghz, 2.3 GHz, 2.4 GHz, 2.5 GHz, 2.6 GHz, 2.7 GHz, 2.8 GHz, 3.0 GHz and 3.2 GHz processors. There isn't a clean way to use an integer divisor on these clock rates (the cleanest way to derive a slower clock rate is to simply "divide" the clock by an integer) and achieve clean speeds of 266 MHz (DDR2-533), 333 MHz (DDR2-667) or 400 MHz (DDR2-800), so the processor tries to come as close to the rated clock rate but no higher. As an example, take the 2.6 GHz Athlon 64 X2 5200+ used with DDR2-667 RAM. If the clock is divided by 7, you get 371 MHz, which is too high for the RAM's rating (333 MHz). So the processor clock is divided by 8 to get 325 MHz, so the DDR2-667 is running at DDR2-650 speed.

I have a chart at http://ethos.battleaxe.net/~prototyped/ddr2-amd.html .

Now when you overclock such a processor, it doesn't take into account the fact that it's overclocked, so for instance, if you were to push that Athlon 64 X2 5200+ to 3.0 GHz by raising the HyperTransport bus clock rate from 200 MHz to 230 MHz, the processor will try to clock the memory assuming its stock speed of 2.6 GHz.

Fortunately, the speed of the RAM itself is something that the processor can be configured to assume. So, at a BIOS "DDR2-533" setting, at 2.6 GHz, it'd try to use a 10x divider, to run the RAM at 260 MHz (DDR2-520 speed). Overclocked to 3.0 GHz, it'd try to use the same 10x divider, so you'd end up with a memory clock rate of 300 MHz. Similarly at a "DDR2-667" setting it'd try to use an 8x divider, so at 3.0 GHz it'd end up trying to run the memory at 375 MHz. Finally at a "DDR2-800" setting it'd try to use a 7x divider, so at 3.0 GHz, the memory would run at 428 MHz.

So if you have DDR2-667 rated memory with an Athlon 64 X2 5200+ and want to overclock the processor without going over the memory's speed rating, you can set the motherboard to get the memory controller to use a "DDR2-533 ratio", and at 3.0 GHz, the RAM will run at 300 MHz (DDR2-600 speed), within its rating.

The discrepancy gets larger with higher overclocks. For instance, if you took an Athlon 64 X2 4000+ (2.1 GHz) processor up to 2.7 GHz, a 28% overclock, you'd need to run DDR2-533 at "DDR2-400" ratio (245 MHz, or DDR2-490 speed), DDR2-667 at "DDR2-533" ratio (337.5 MHz, at which the DDR2-667 is already beyond rated spec, but will probably still run reliably), DDR2-800 at "DDR2-667" ratio (386 MHz, or DDR2-771 speed) or DDR2-1066 at "DDR2-800" ratio (450 MHz, or DDR2-900 speed).

[rusty]

īsais variants:

ram un fsb frekvencēm nav jāsakrīt (ja arī sakrīt, tad bišk labāk preformē).

tev jāskatās ko mātesplate atbalsta - cik mhz un kādu tipu.

overclocing - aizmirst.

ja nespēj aizmirst, tad overclockers.ru vai iemet aci boota.

literatūra ar ko sāk overclockingam

Edited January 29, 2009 by rusty

[DoubleT]

man karoč iekšā sēž DDR-1 ramucis ar 200mhz ;]] bet tikai 256MB gribēju ko mazliet "niknāku" iemest bet tagad gribēju kļūt skaidrs kas ar to FSB un kā tur ;]]

+ to jaut. par videni.? es gļukoju vai tā varētu būt ( i promise mom, i will not eat the shrooms anymore )

[rusty]

videokartes guid book'a rakstīts cik nepieciešams ram viņas uzturēšanai.

un jā. tā varētu būt. nu salīdzināt tavu ar manu nav pareizi, bet manai (geforce 9600 gt) min req ir 128mb, bet priekš best preformance vajag 2gb.

[DoubleT]

nu ok par videni iebraucu aplūrēšu ATI lapā cik vajag kas un kā ...

tātad caur Everest var uzzināt cik FSB ātrums ir.? gribētos zināt tad varētu jums pateikt un tad jūs gudrie pateiktu cik reāli man vēl likt RAMu iekšā lai ir uz tip-top a to man CS uz 40fps iet ( kad šauj 15 ) a procis 2.3 pohožē normāli vainu meklēju RAMā a maybe ne tur meklēju , bet nu pats PC ar tāds mazliet lēns .... tagad vēl doma likt 1TB HDD iekšā ;]] šim m79.lv tagad skidons normāls

EDĪTE: jup manai videnei homepage raksztīts laikam ka 128mb vajag ;]]

http://ati.amd.com/products/radeon9600/rad...ro/compare.html

Edited January 30, 2009 by DoubleT

[rusty]

saproti - tavā situācijā ir pilnīgi pofig par ram un fsb frekvencēm! tev vnk pietrūks fiziskās atmiņas, lai normāli ko darbinātu!

noskaidro ko tev mātesplate atbalsta (kādu ram) un iegādājies! 1 gb tev būs gana.

[DoubleT]

noskaidro ko tev mātesplate atbalsta (kādu ram) un iegādājies! 1 gb tev būs gana.

labi skaidrs. pačolēšu kad ieripos naudiņa kontā. ;]]